diff --git a/README.md b/README.md
index ddc3f47..5a3d391 100644
--- a/README.md
+++ b/README.md
@@ -49,7 +49,7 @@ in the terminal (usually **http://127.0.0.1:1313/**); **Ctrl+C** stops both.
 | Choose | What you can do |
 | --- | --- |
 | Member profiles | Update your details, research areas, links, awards, and portrait. |
-| Publications | Add papers, abstracts, authors, tags, and related datasets. |
+| Publications | Add papers, abstracts, authors, research outcomes, tags, and related datasets. |
 | Personal Pages | Build a page from a template or preview your custom website. |
 | Datasets and downloads | Add descriptions, related papers, download links, and files. |
 
@@ -85,8 +85,17 @@ to share that.
 ### Add or edit publications
 
 Choose **Publications**, select a paper or **New entry**, and fill in
-the title, year, authors, venue, paper URL, and publication type.
+the title, year, authors, venue, paper links, and publication type.
 Use **Year** for the paper's publication year.
+Use **Published / official URL** for the published version's DOI, publisher page,
+or official publication record. Use **Preprint URL** for the corresponding arXiv
+or other preprint repository page. Provide at least one; both are welcome.
+The published version takes priority, with a less prominent preprint link beside
+it. If only a preprint is available, leave the published URL blank. 
+Explain significant title or version differences in **Additional
+notes**. When a paper is published, add its
+published URL and update the venue, year, and publication type; keep its preprint
+URL as a secondary link.
 Choose **Publication type** from the dropdown. If the correct type is missing,
 contact the site owner to have it added.
 For authors, search website members and their published name variants, or type
@@ -95,6 +104,13 @@ order shown on the paper. Tags and other item lists work the same way; you can
 also paste a list with line breaks.
 Add research areas and related datasets where relevant.
 
+If the paper has public code, a benchmark, data, or a project website, choose
+**Add research outcome**. Fill in its name, HTTPS link, and a short plain-text
+description of what is available. You can add more than one outcome or leave
+the section empty. Save and preview the publication to check the links and text.
+Outcomes appear on the publication page and in its selected research areas;
+changing those selections also updates where its outcomes appear.
+
 The page identifier is assigned automatically from the original title and cannot
 be edited. If it is already in use, a number is added (for example, `paper-2`).
 Changing the title later keeps the original URL and adds a redirect from a URL
@@ -168,6 +184,7 @@ Link to an external repository for datasets larger than **20 MB**.
 | --- | --- |
 | Homepage text | `content/_index.md` |
 | Research themes | `content/research/` |
+| Research software and benchmark outcomes | Publication editor, saved in `content/publications/` |
 | Navigation and site settings | `hugo.yaml` |
 
 Use GitHub's file editor on a branch or fork, then submit a PR.
diff --git a/assets/css/main.css b/assets/css/main.css
index e79efbb..ec9aede 100644
--- a/assets/css/main.css
+++ b/assets/css/main.css
@@ -12,6 +12,7 @@
   --accent: #86a7d8;
   --copper: #9e6339;
   --sage: #54724b;
+  --teal: #23746e;
   --shadow: 0 18px 42px rgba(13, 31, 60, 0.08);
   --shadow-soft: 0 10px 24px rgba(13, 31, 60, 0.06);
   --radius: 1.35rem;
@@ -730,6 +731,20 @@ ul {
   color: var(--primary);
 }
 
+.publication-links {
+  align-items: center;
+}
+
+.publication-links .button-secondary:hover {
+  background: var(--bg-alt);
+  color: var(--primary-deep);
+}
+
+.publication-links-compact .publication-preprint {
+  color: var(--muted);
+  font-size: 0.95em;
+}
+
 .hero-card .button {
   background: var(--success);
 }
@@ -1046,6 +1061,10 @@ a.chip {
   border-top: 4px solid var(--primary);
 }
 
+.accent-teal {
+  border-top: 4px solid var(--teal);
+}
+
 .person-card {
   display: flex;
   flex-direction: column;
@@ -1217,6 +1236,13 @@ button:disabled { cursor: default; opacity: 0.6; }
   font-size: 1.9rem;
 }
 
+.research-support-logos { display: flex; flex-wrap: wrap; align-items: center; gap: 2rem; margin-block: 1rem; }
+.research-support-logos img { display: block; object-fit: contain; }
+.research-diagrams { display: grid; grid-template-columns: repeat(auto-fit, minmax(min(100%, 20rem), 1fr)); gap: 2rem; margin-block: 2rem; }
+.research-diagrams figure { margin: 0 auto; max-width: 25rem; }
+.research-diagrams img { display: block; width: 100%; height: auto; }
+.research-diagrams figcaption { color: var(--muted); font-size: 0.9rem; margin-top: 0.75rem; }
+
 /* Long display equations scroll within their panel, including on phones. */
 .math-block {
   max-width: 100%;
diff --git a/assets/js/site.js b/assets/js/site.js
index e547dbd..d02a81e 100644
--- a/assets/js/site.js
+++ b/assets/js/site.js
@@ -73,6 +73,9 @@ document.addEventListener("DOMContentLoaded", () => {
     const emptyState = document.querySelector("[data-publication-empty]");
     const controls = document.querySelector("[data-publication-controls]");
     const researchButtons = [...document.querySelectorAll("[data-publication-research]")];
+    const researchAliases = new Map(researchButtons.flatMap((button) =>
+      (button.dataset.publicationResearchAliases || "").split(",").filter(Boolean)
+        .map((alias) => [alias, button.dataset.publicationResearch])));
     const resultCount = document.querySelector("[data-publication-count]");
     const clearButton = document.querySelector("[data-publication-clear]");
     const normalize = (text) => text.normalize("NFD").replace(/\p{M}/gu, "").replace(/[‘’]/g, "'").toLowerCase();
@@ -197,8 +200,13 @@ document.addEventListener("DOMContentLoaded", () => {
       const url = new URL(location.href);
       const params = url.searchParams;
       searchInput.value = params.get("q") || params.get("author") || "";
-      const area = params.get("research");
+      const requestedArea = params.get("research");
+      const area = researchAliases.get(requestedArea) || requestedArea;
       activeResearch = researchButtons.some((button) => button.dataset.publicationResearch === area) ? area : "all";
+      if (requestedArea !== area) {
+        url.searchParams.set("research", area);
+        history.replaceState(null, "", url);
+      }
       const page = params.get("page") || url.pathname.match(/\/page\/(\d+)\//)?.[1] || "1";
       currentPage = /^[1-9]\d*$/.test(page) && Number.isSafeInteger(Number(page)) ? Number(page) : 1;
       const filtered = searchInput.value || activeResearch !== "all";
diff --git a/content/_index.md b/content/_index.md
index 2e80603..11c4639 100644
--- a/content/_index.md
+++ b/content/_index.md
@@ -1,7 +1,7 @@
 +++
 title = "Home"
 hero_subtitle = "The Middleware Systems Research Group at the University of Toronto"
-hero_tagline = "Systems research that spans data, learning, and quantum computing"
+hero_tagline = "Systems research that spans data, learning, quantum computing, and reliability"
 +++
 
 Traditionally, the “M” in MSRG stood for “middleware”—the Middleware Systems Research Group at the University of Toronto. But don’t worry, “middleware is literally everywhere.” Over the years, that “M” has evolved to mean “mobile,” “microservices,” “machine learning,” and, last but not least, “matter,” as in entangled particles in space-time. What *more* could you ask for?
diff --git a/content/people/alexander-erben.md b/content/people/alexander-erben.md
index a31cd86..7e7b7fe 100644
--- a/content/people/alexander-erben.md
+++ b/content/people/alexander-erben.md
@@ -6,7 +6,7 @@ slug = "alexander-erben"
 role = "PhD Student"
 status = "alumni"
 location = "MSRG alumnus"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 orcid = "https://orcid.org/0000-0002-0153-7251"
 linkedin = "https://www.linkedin.com/in/alexandererben/"
 end_date = "2024"
diff --git a/content/people/ben-cheng.md b/content/people/ben-cheng.md
index 2fbfeff..413b0f6 100644
--- a/content/people/ben-cheng.md
+++ b/content/people/ben-cheng.md
@@ -5,7 +5,7 @@ role = "Undergraduate Student"
 status = "current"
 start_date = "2023-09"
 location = "Toronto, Canada"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 email = ""
 interests = ["Machine Learning Systems", "Data Management"]
 portrait = "/images/people/ben-cheng.jpg"
diff --git a/content/people/chenkai-zhang.md b/content/people/chenkai-zhang.md
index 17a2e48..d53849d 100644
--- a/content/people/chenkai-zhang.md
+++ b/content/people/chenkai-zhang.md
@@ -9,6 +9,6 @@ email = "zhangchenkai.zhang@mail.utoronto.ca"
 interests = ["AI Agent Reliability", "Fault Diagnosis"]
 author_names = ["Chenkai Zhang"]
 portrait = ""
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 linkedin = "https://www.linkedin.com/in/chenkaizhang1/"
 +++
diff --git a/content/people/fei-pan.md b/content/people/fei-pan.md
index d9ef379..18e7c54 100644
--- a/content/people/fei-pan.md
+++ b/content/people/fei-pan.md
@@ -7,6 +7,6 @@ status = "alumni"
 start_date = "2022-01"
 location = "Toronto, Canada"
 researchgate = "https://www.researchgate.net/profile/Fei-Pan-2"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 interests = ["Distributed Systems", "Byzantine Fault Tolerance", "Consensus Protocols"]
 +++
diff --git a/content/people/hans-arno-jacobsen.md b/content/people/hans-arno-jacobsen.md
index e7e6a45..bdceeaa 100644
--- a/content/people/hans-arno-jacobsen.md
+++ b/content/people/hans-arno-jacobsen.md
@@ -5,7 +5,7 @@ role = "Professor"
 status = "current"
 start_date = "2000-01"
 location = "Toronto, Canada"
-research = ["quantum-computing-systems", "distributed-machine-learning", "data-management"]
+research = ["quantum-computing-systems", "rca-aiops", "data-management"]
 email = "jacobsen@eecg.utoronto.ca"
 homepage = "https://www.eecg.toronto.edu/~jacobsen"
 orcid = "https://orcid.org/0000-0003-0813-0101"
diff --git a/content/people/herbert-woisetschlaeger.md b/content/people/herbert-woisetschlaeger.md
index 8b19dc0..ab945f8 100644
--- a/content/people/herbert-woisetschlaeger.md
+++ b/content/people/herbert-woisetschlaeger.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "alumni"
 start_date = "2022-09"
 location = "Munich, Germany"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 interests = ["Federated Learning", "Efficient Machine Learning"]
 portrait = "/images/people/herbert-woisetschlaeger.jpg"
 homepage = "https://research.ibm.com/people/herbert-woisetschlaeger"
diff --git a/content/people/jana-vatter.md b/content/people/jana-vatter.md
index 3d26ffe..406d266 100644
--- a/content/people/jana-vatter.md
+++ b/content/people/jana-vatter.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "alumni"
 start_date = "2022"
 location = "Munich, Germany"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 interests = ["Graph Neural Networks", "Distributed Deep Learning", "Graph Processing"]
 portrait = "/images/people/jana-vatter.jpg"
 linkedin = "https://www.linkedin.com/in/jana-vatter/"
diff --git a/content/people/mengli-duan.md b/content/people/mengli-duan.md
index d6757d0..196ebcd 100644
--- a/content/people/mengli-duan.md
+++ b/content/people/mengli-duan.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "current"
 start_date = "2026-09"
 location = "Toronto, Canada"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 email = "mengli.duan@mail.utoronto.ca"
 interests = ["Machine Learning Systems"]
 author_names = ["Mengli Duan", "Dawn (Mengli) Duan", "Dawn Duan"]
diff --git a/content/people/michalis-bachras.md b/content/people/michalis-bachras.md
index 689255b..25bac30 100644
--- a/content/people/michalis-bachras.md
+++ b/content/people/michalis-bachras.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "current"
 start_date = "2023-09"
 location = "Toronto, Canada"
-research = ["data-management"]
+research = ["data-management", "rca-aiops"]
 interests = ["Data Management", "Distributed Systems"]
 portrait = "/images/people/michalis-bachras.jpg"
 linkedin = "https://www.linkedin.com/in/michalis-bachras/"
diff --git a/content/people/nikolai-merkel.md b/content/people/nikolai-merkel.md
index dc8a341..1bafd48 100644
--- a/content/people/nikolai-merkel.md
+++ b/content/people/nikolai-merkel.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "alumni"
 start_date = "2020-08"
 location = "Munich, Germany"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 interests = ["Graph Partitioning", "Graph Neural Networks", "Distributed Systems"]
 portrait = "/images/people/nikolai-merkel.jpg"
 homepage = "https://www.cs.cit.tum.de/dis/alumni/nikolai-merkel/"
diff --git a/content/people/rev-liu.md b/content/people/rev-liu.md
index 57dda5d..92f9958 100644
--- a/content/people/rev-liu.md
+++ b/content/people/rev-liu.md
@@ -1,12 +1,14 @@
 +++
-name = "Rev Liu"
+name = "Ruiwu Liu"
 slug = "rev-liu"
-role = "MSc Student"
+role = "PhD Student"
 status = "current"
-start_date = "2025-09"
+start_date = "2026-09"
 location = "Toronto, Canada"
+research = ["rca-aiops"]
 email = "rev.liu@mail.utoronto.ca"
-interests = []
+interests = ["VLM", "RL", "AIOps"]
 author_names = []
-portrait = ""
+portrait = "/images/people/ruiwu-liu.jpg"
+linkedin = "https://www.linkedin.com/in/ruiwu-liu-b1534827b/"
 +++
diff --git a/content/people/ruben-mayer.md b/content/people/ruben-mayer.md
index d099e90..3915eaa 100644
--- a/content/people/ruben-mayer.md
+++ b/content/people/ruben-mayer.md
@@ -5,7 +5,7 @@ slug = "ruben-mayer"
 role = "Post-Doc"
 status = "alumni"
 location = "MSRG alumnus"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 orcid = "https://orcid.org/0000-0001-9870-7466"
 dblp = "https://dblp.org/pid/132/0330"
 homepage = "https://ruben-mayer.com/"
diff --git a/content/people/yifang-tian.md b/content/people/yifang-tian.md
index ecbcab4..db49b85 100644
--- a/content/people/yifang-tian.md
+++ b/content/people/yifang-tian.md
@@ -5,7 +5,7 @@ role = "PhD Student"
 status = "current"
 start_date = "2022-01"
 location = "Toronto, Canada"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 interests = ["Machine Learning Systems", "Distributed Systems"]
 portrait = "/images/people/yifang-tian.jpg"
 email = "yifang.tian@mail.utoronto.ca"
diff --git a/content/people/yifeng-he.md b/content/people/yifeng-he.md
index 57fa204..3d6a69e 100644
--- a/content/people/yifeng-he.md
+++ b/content/people/yifeng-he.md
@@ -5,6 +5,7 @@ role = "MSc Student"
 status = "current"
 start_date = "2025-09"
 location = "Toronto, Canada"
+research = ["rca-aiops"]
 email = "yifeng.he@mail.utoronto.ca"
 interests = []
 author_names = ["Yifeng He", "Yifeng (Jason) He"]
diff --git a/content/people/yiran-li.md b/content/people/yiran-li.md
index 50cb7af..ccf948e 100644
--- a/content/people/yiran-li.md
+++ b/content/people/yiran-li.md
@@ -5,7 +5,7 @@ role = "Post-Doc"
 status = "current"
 start_date = "2025-11"
 location = "Toronto, Canada"
-research = ["quantum-computing-systems", "distributed-machine-learning", "data-management"]
+research = ["quantum-computing-systems", "rca-aiops", "data-management"]
 interests = ["Graph Analytics", "Graph Learning", "Quantum Data Management", "AI Agent Reliability"]
 email = ""
 homepage = "https://cyanidecentral.github.io/"
diff --git a/content/people/yuanchen-gao.md b/content/people/yuanchen-gao.md
new file mode 100644
index 0000000..eb2efca
--- /dev/null
+++ b/content/people/yuanchen-gao.md
@@ -0,0 +1,15 @@
++++
+name = "Yuanchen Gao"
+slug = "yuanchen-gao"
+role = "PhD Student"
+status = "current"
+location = "HKUST, Hong Kong"
+research = ["rca-aiops"]
+interests = ["Root Cause Analysis", "Automated Program Repair", "AIOps"]
+author_names = ["Yuanchen Gao"]
+email = "ygaocv@connect.ust.hk"
+# PhD role and current RCA participation confirmed by the maintainer, 2026-11-01.
+# HKUST group: https://cse.hkust.edu.hk/~charlesz/
+# Affiliation and public email: https://arxiv.org/html/2608.17018v1
+# MSRG visit dates, portrait, and personal academic accounts are unverified.
++++
diff --git a/content/people/zongxin-liu.md b/content/people/zongxin-liu.md
index a06582b..4b35763 100644
--- a/content/people/zongxin-liu.md
+++ b/content/people/zongxin-liu.md
@@ -6,7 +6,7 @@ status = "alumni"
 start_date = "2021-09"
 end_date = "2023-06"
 location = "Toronto, Canada"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 email = ""
 linkedin = "https://www.linkedin.com/in/zongxin-liu-155615b6/"
 interests = ["Distributed Systems", "CRDTs", "Data Consistency"]
diff --git a/content/publications/2ps-high-quality-edge-partitioning-with-two-phase-streaming.md b/content/publications/2ps-high-quality-edge-partitioning-with-two-phase-streaming.md
index cfce69e..6e3624e 100644
--- a/content/publications/2ps-high-quality-edge-partitioning-with-two-phase-streaming.md
+++ b/content/publications/2ps-high-quality-edge-partitioning-with-two-phase-streaming.md
@@ -5,6 +5,6 @@ authors = ["Ruben Mayer", "Kamil Orujzade", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/2001.07086"
+preprint_url = "https://arxiv.org/abs/2001.07086"
 abstract = "Graph partitioning is an important preprocessing step to distributed graph processing. In edge partitioning, the edge set of a given graph is split into $k$ equally-sized partitions, such that the replication of vertices across partitions is minimized. Streaming is a viable approach to partition graphs that exceed the memory capacities of a single server. The graph is ingested as a stream of edges, and one edge at a time is immediately and irrevocably assigned to a partition based on a scoring function. However, streaming partitioning suffers from the uninformed assignment problem: At the time of partitioning early edges in the stream, there is no information available about the rest of the edges. As a consequence, edge assignments are often driven by balancing considerations, and the achieved replication factor is comparably high. In this paper, we propose 2PS, a novel two-phase streaming algorithm for high-quality edge partitioning. In the first phase, vertices are separated into clusters by a lightweight streaming clustering algorithm. In the second phase, the graph is re-streamed and edge partitioning is performed while taking into account the clustering of the vertices from the first phase. Our evaluations show that 2PS can achieve a replication factor that is comparable to heavy-weight random access partitioners while inducing orders of magnitude lower memory overhead."
 +++
diff --git a/content/publications/a-comprehensive-survey-of-machine-unlearning-techniques-for-large-language-models.md b/content/publications/a-comprehensive-survey-of-machine-unlearning-techniques-for-large-language-models.md
index fadb062..3242554 100644
--- a/content/publications/a-comprehensive-survey-of-machine-unlearning-techniques-for-large-language-models.md
+++ b/content/publications/a-comprehensive-survey-of-machine-unlearning-techniques-for-large-language-models.md
@@ -4,7 +4,7 @@ year = 2025
 authors = ["Jiahui Geng", "Qing Li", "Herbert Woisetschlaeger", "Zongxiong Chen", "Fengyu Cai", "Yuxia Wang", "Preslav Nakov", "Hans-Arno Jacobsen", "Fakhri Karray"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2503.01854"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2503.01854"
 abstract = "This study investigates the machine unlearning techniques within the context of large language models (LLMs), referred to as *LLM unlearning*. LLM unlearning offers a principled approach to removing the influence of undesirable data (e.g., sensitive or illegal information) from LLMs, while preserving their overall utility without requiring full retraining. Despite growing research interest, there is no comprehensive survey that systematically organizes existing work and distills key insights; here, we aim to bridge this gap. We begin by introducing the definition and the paradigms of LLM unlearning, followed by a comprehensive taxonomy of existing unlearning studies. Next, we categorize current unlearning approaches, summarizing their strengths and limitations. Additionally, we review evaluation metrics and benchmarks, providing a structured overview of current assessment methodologies. Finally, we outline promising directions for future research, highlighting key challenges and opportunities in the field."
 +++
diff --git a/content/publications/a-crowdsourcing-approach-for-the-inference-of-distribution-grids.md b/content/publications/a-crowdsourcing-approach-for-the-inference-of-distribution-grids.md
index ac12e8e..fde318a 100644
--- a/content/publications/a-crowdsourcing-approach-for-the-inference-of-distribution-grids.md
+++ b/content/publications/a-crowdsourcing-approach-for-the-inference-of-distribution-grids.md
@@ -4,7 +4,7 @@ year = 2018
 authors = ["Pezhman Nasirifard", "José Rivera", "Qunjie Zhou", "Klaus Bernd Schreiber", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the Ninth International Conference on Future Energy Systems"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3208903.3208927"
 abstract = "Maintaining a complete and up-to-date model of the distribution grid is a challenging task, and the scarcity of open models represents a significant bottleneck for researchers in this area. In this work, we address these challenges by introducing a crowdsourcing framework for the collection of open data on distribution grid devices and an algorithm to infer the topological model of the distribution grids. We use the crowd and smartphones to collect an image and the geographical position of power distribution grid devices. Since power distribution lines are usually underground and cannot be mapped, we use spatial data analytics on the collected data in combination with other open data sources to infer the topology of the distribution grid. This paper describes and evaluates our crowdsourcing and inference approach. To evaluate our approach, we organized and conducted a crowdsourcing campaign to map and infer a sizeable district in Munich, Germany. The results are compared with the ground truth of the distribution system operator. Our field experiments show that using the crowd to recognize power distribution elements, a precision of up to 82% and a recall of up to 65% can be obtained. The numerical evaluation of our inference algorithm demonstrates that the model we inferred based on the acquired official DSO grid dataset achieves a power length accuracy of 88% compared to the ground truth. These results confirm our approach as a practical method to infer real power distribution grid models."
 +++
diff --git a/content/publications/a-memory-bandwidth-efficient-hybrid-radix-sort-on-gpus.md b/content/publications/a-memory-bandwidth-efficient-hybrid-radix-sort-on-gpus.md
index 9a8f3a7..aa46d1d 100644
--- a/content/publications/a-memory-bandwidth-efficient-hybrid-radix-sort-on-gpus.md
+++ b/content/publications/a-memory-bandwidth-efficient-hybrid-radix-sort-on-gpus.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3035918.3064043"
 abstract = "Sorting is at the core of many database operations, such as index creation, sort-merge joins, and user-requested output sorting. As GPUs are emerging as a promising platform to accelerate various operations, sorting on GPUs becomes a viable endeavour. Over the past few years, several improvements have been proposed for sorting on GPUs, leading to the first radix sort implementations that achieve a sorting rate of over one billion 32-bit keys per second. Yet, state-of-the-art approaches are heavily memory bandwidth-bound, as they require substantially more memory transfers than their CPU-based counterparts. Our work proposes a novel approach that almost halves the amount of memory transfers and, therefore, considerably lifts the memory bandwidth limitation. Being able to sort two gigabytes of eight-byte records in as little as 50 milliseconds, our approach achieves a 2.32-fold improvement over the state-of-the-art GPU-based radix sort for uniform distributions, sustaining a minimum speed-up of no less than a factor of 1.66 for skewed distributions. To address inputs that either do not reside on the GPU or exceed the available device memory, we build on our efficient GPU sorting approach with a pipelined heterogeneous sorting algorithm that mitigates the overhead associated with PCIe data transfers. Comparing the end-to-end sorting performance to the state-of-the-art CPU-based radix sort running 16 threads, our heterogeneous approach achieves a 2.06-fold and a 1.53-fold improvement for sorting 64 GB key-value pairs with a skewed and a uniform distribution, respectively."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1611.01137"
 +++
diff --git a/content/publications/a-serverless-publish-subscribe-system.md b/content/publications/a-serverless-publish-subscribe-system.md
index 5c2e47e..fbd9216 100644
--- a/content/publications/a-serverless-publish-subscribe-system.md
+++ b/content/publications/a-serverless-publish-subscribe-system.md
@@ -5,6 +5,6 @@ authors = ["Pezhman Nasirifard", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/2210.07897"
+preprint_url = "https://arxiv.org/abs/2210.07897"
 abstract = "Operating a scalable and reliable server application, such as publish/subscribe (pub/sub) systems, requires tremendous development efforts and resources. The emerging serverless paradigm simplifies the development and deployment of highly available applications by delegating most operational concerns to the cloud providers. The serverless paradigm describes a programming model where the developers break the application downs into smaller microservices that run on the cloud in response to events. This paper proposes designing a serverless pub/sub system based on the IBM Bluemix cloud platform. Our pub/sub system performs topic-based, content-based, and function-based matchings. The function-based matching is a novel matching approach where the subscribers can define a highly customizable subscription function that the broker applies to the publications in the cloud. The evaluations of the designed system verify the practicality of the designed system. However, the vendor-specific constraints of the IBM Bluemix resources are a bottleneck to the scalability of the broker."
 +++
diff --git a/content/publications/a-user-tunable-machine-learning-framework-for-step-wise-synthesis-planning.md b/content/publications/a-user-tunable-machine-learning-framework-for-step-wise-synthesis-planning.md
index 463bcfa..b81f245 100644
--- a/content/publications/a-user-tunable-machine-learning-framework-for-step-wise-synthesis-planning.md
+++ b/content/publications/a-user-tunable-machine-learning-framework-for-step-wise-synthesis-planning.md
@@ -4,8 +4,10 @@ year = 2026
 authors = ["Shivesh Prakash", "Nandan Patel", "Hans-Arno Jacobsen", "Viki Kumar Prasad"]
 venue = "Digital Discovery"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1039/d5dd00562k"
 abstract = "We introduce MHNpath, a machine learning-driven retrosynthetic tool designed for computer-aided synthesis planning. Leveraging modern Hopfield networks and novel comparative metrics, MHNpath efficiently prioritizes reaction templates, improving the scalability and accuracy of retrosynthetic predictions. The tool incorporates a tunable scoring system that allows users to prioritize pathways based on cost, reaction temperature, and toxicity, thereby facilitating the design of greener and cost-effective reaction routes. We demonstrate its effectiveness through case studies involving complex molecules from ChemByDesign, showcasing its ability to predict novel synthetic and enzymatic pathways. Furthermore, we benchmark MHNpath against existing frameworks using the PaRoutes dataset, achieving a solution rate of 85.4% and replicating 69.2% of experimentally validated “gold-standard” pathways. Our case studies show that the tool can generate shorter, cheaper, moderate-temperature routes employing green solvents, as exemplified by molecules such as dronabinol, arformoterol, and lupinine."
 abstract_license_url = "https://creativecommons.org/licenses/by/3.0/"
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2504.02191"
 +++
diff --git a/content/publications/adversarial-robustness-in-distributed-quantum-machine-learning.md b/content/publications/adversarial-robustness-in-distributed-quantum-machine-learning.md
index e54a268..c3b8166 100644
--- a/content/publications/adversarial-robustness-in-distributed-quantum-machine-learning.md
+++ b/content/publications/adversarial-robustness-in-distributed-quantum-machine-learning.md
@@ -8,4 +8,6 @@ research = ["quantum-computing-systems"]
 tags = ["quantum-systems"]
 external_url = "https://doi.org/10.1007/978-3-032-11153-1_11"
 url = "/publications/adversarial-robustness-in-distributed-quantum-machine-learning/"
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2508.11848"
 +++
diff --git a/content/publications/adversarial-robustness-of-partitioned-quantum-classifiers.md b/content/publications/adversarial-robustness-of-partitioned-quantum-classifiers.md
index 04d66ae..9b14920 100644
--- a/content/publications/adversarial-robustness-of-partitioned-quantum-classifiers.md
+++ b/content/publications/adversarial-robustness-of-partitioned-quantum-classifiers.md
@@ -6,5 +6,5 @@ venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["quantum-computing-systems"]
 tags = ["quantum-systems"]
-external_url = "https://arxiv.org/abs/2502.20403"
+preprint_url = "https://arxiv.org/abs/2502.20403"
 +++
diff --git a/content/publications/an-end-to-end-performance-comparison-of-seven-permissioned-blockchain-systems.md b/content/publications/an-end-to-end-performance-comparison-of-seven-permissioned-blockchain-systems.md
index 6f23392..9ecd5e0 100644
--- a/content/publications/an-end-to-end-performance-comparison-of-seven-permissioned-blockchain-systems.md
+++ b/content/publications/an-end-to-end-performance-comparison-of-seven-permissioned-blockchain-systems.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3590140.3629106"
 abstract = "The emergence of numerous blockchain solutions, offering innovative approaches to optimise performance, scalability, privacy, and governance, complicates performance analysis. Reasons for the difficulty of benchmarking blockchains include, for example, the high number of system parameters to configure and the effort to deploy a blockchain network. In addition, performance data, which mostly comes from system vendors, is often opaque. We provide a reproducible evaluation of the performance of seven permissioned blockchain systems across different parameter settings. We employ an end-to-end approach, where the clients sending the transactions are fully involved in the data collection approach. Our results underscore the unique characteristics and limitations of the systems we examined. Due to the insights given, our work forms the basis for continued research to optimise the performance of blockchain systems."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2311.15433"
 +++
diff --git a/content/publications/an-experimental-comparison-of-partitioning-strategies-for-distributed-graph-neural-network-training.md b/content/publications/an-experimental-comparison-of-partitioning-strategies-for-distributed-graph-neural-network-training.md
index 66da8d8..d767169 100644
--- a/content/publications/an-experimental-comparison-of-partitioning-strategies-for-distributed-graph-neural-network-training.md
+++ b/content/publications/an-experimental-comparison-of-partitioning-strategies-for-distributed-graph-neural-network-training.md
@@ -4,8 +4,10 @@ year = 2025
 authors = ["Nikolai Merkel", "Daniel Stoll", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "EDBT"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["graph-systems"]
 external_url = "https://doi.org/10.48786/edbt.2025.14"
 abstract = "Recently, graph neural networks (GNNs) have gained much attention as a growing area of deep learning capable of learning on graph-structured data. However, the computational and memory requirements for training GNNs on large-scale graphs make it necessary to distribute the training. A prerequisite for distributed GNN training is to partition the input graph into smaller parts that are distributed among multiple machines of a compute cluster. Although graph partitioning has been studied with regard to graph analytics and graph databases, its effect on GNN training performance is largely unexplored. As a consequence, it is unclear whether investing computational efforts into high-quality graph partitioning would pay off in GNN training scenarios. In this paper, we study the effectiveness of graph partitioning for distributed GNN training. Our study aims to understand how different factors such as GNN parameters, mini-batch size, graph type, features size, and scale-out factor influence the effectiveness of graph partitioning. We conduct experiments with two different GNN systems using vertex and edge partitioning. We found that high-quality graph partitioning is a very effective optimization to speed up GNN training and to reduce memory consumption. Furthermore, our results show that invested partitioning time can quickly be amortized by reduced GNN training time, making it a relevant optimization for most GNN scenarios. Compared to research on distributed graph processing, our study reveals that graph partitioning plays an even more significant role in distributed GNN training, which motivates further research on the graph partitioning problem."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2308.15602"
 +++
diff --git a/content/publications/analyzing-common-electronic-structure-theory-algorithms-for-distributed-quantum-computing.md b/content/publications/analyzing-common-electronic-structure-theory-algorithms-for-distributed-quantum-computing.md
index 8fd2144..afbf6ca 100644
--- a/content/publications/analyzing-common-electronic-structure-theory-algorithms-for-distributed-quantum-computing.md
+++ b/content/publications/analyzing-common-electronic-structure-theory-algorithms-for-distributed-quantum-computing.md
@@ -8,4 +8,6 @@ research = ["quantum-computing-systems"]
 tags = ["quantum-systems"]
 external_url = "https://doi.org/10.1109/qce65121.2025.10351"
 abstract = "To move towards the utility era of quantum computing, many corporations have posed distributed quantum computing (DQC) as a framework for scaling the current generation of devices for practical applications. One of these applications is quantum chemistry, also known as electronic structure theory, which has been poised as a “killer application” of quantum computing. To this end, we analyze five electronic structure methods, including the unitary coupled-cluster singles and doubles (UCCSD), unitary pair coupled-cluster doubles (UpCCD), unitary pair coupled-cluster with generalized singles and doubles (UpCCGSD), separable pair approximation plus generalized singles (SPA+GS), and local unitary cluster Jastrow (LUCJ) ansätze. The key benefit of these methods is that they can be found in common packages, such as Tequila and ffsim, that interface easily with the Qiskit Circuit Cutting addon. Herein, we highlight the challenges and aptitude of the aforementioned electronic structure methods for distribution using quasiprobability simulation with local operations (LO). The key findings of our work are that many of these algorithms cannot be efficiently parallelized using LO, and new methods must be developed to apply electronic structure theory within a DQC framework."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2507.01902"
 +++
diff --git a/content/publications/analyzing-geospatial-distribution-in-blockchains.md b/content/publications/analyzing-geospatial-distribution-in-blockchains.md
index 32c09f5..cfd7a79 100644
--- a/content/publications/analyzing-geospatial-distribution-in-blockchains.md
+++ b/content/publications/analyzing-geospatial-distribution-in-blockchains.md
@@ -8,4 +8,6 @@ research = ["data-management"]
 tags = ["blockchain"]
 external_url = "https://doi.org/10.1109/dapps57946.2023.00022"
 abstract = "Blockchains are decentralized; are they genuinely? We analyze blockchain decentralization's often-overlooked but quantifiable dimension: geospatial distribution of transaction processing. Blockchains bring with them the potential for geospatially distributed transaction processing. They enable validators from geospatially distant locations to partake in consensus protocols; we refer to them as minority validators. Based on our observations, in practice, most validators are often geographically concentrated in close proximity. Furthermore, we observed that minority validators tend not to meet the performance requirements, often misidentified as crash failures. Consequently, they are subject to punishment by jailing (removal from the validator set) and/or slashing (penalty in native tokens). Our emulations, under controlled conditions, demonstrate the same results, raising serious concerns about the potential for the geospatial centralization of validators. To address this, we developed a solution that easily integrates with consensus protocols, and we demonstrated its effectiveness."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2305.17771"
 +++
diff --git a/content/publications/analyzing-initialization-strategies-for-the-local-unitary-cluster-jastrow-ansatz-within-the-quantum-centric-supercomputing-framework.md b/content/publications/analyzing-initialization-strategies-for-the-local-unitary-cluster-jastrow-ansatz-within-the-quantum-centric-supercomputing-framework.md
index 59ebc02..685ab58 100644
--- a/content/publications/analyzing-initialization-strategies-for-the-local-unitary-cluster-jastrow-ansatz-within-the-quantum-centric-supercomputing-framework.md
+++ b/content/publications/analyzing-initialization-strategies-for-the-local-unitary-cluster-jastrow-ansatz-within-the-quantum-centric-supercomputing-framework.md
@@ -2,9 +2,11 @@
 title = "Analyzing Initialization Strategies for the Local Unitary Cluster Jastrow Ansatz within the Quantum-Centric Supercomputing Framework"
 year = 2026
 authors = ["Grier M. Jones", "Maforikan J. Amoussou", "Maximilian O. Leach", "Hans-Arno Jacobsen"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
+venue = "Journal of Chemical Theory and Computation 22(19), pp. 10078–10090"
+publication_type = "Journal Article"
 research = ["quantum-computing-systems"]
-external_url = "https://arxiv.org/abs/2606.14933"
+external_url = "https://doi.org/10.1021/acs.jctc.6c01195"
+# ACS publisher metadata: https://api.crossref.org/works/10.1021/acs.jctc.6c01195
 abstract = "In this study, we analyze the choice of local unitary cluster Jastrow (LUCJ) ansatz initialization and sensitivity of the sample-based quantum diagonalization (SQD) algorithm within the quantum-centric supercomputing (QCSC) framework. We examine six initialization strategies, including those based on coupled-cluster singles and doubles (CCSD), Møller-Plesset second-order perturbation theory (MP2), data-driven coupled-cluster (DDCC), and trivial (zeroes and random) initializations, across twelve molecular systems and three basis sets (STO-3G, cc-pVDZ, and aug-cc-pVDZ). We find that while the mean absolute percentage errors (MAPEs) between the alternative and CCSD-initialized t2-amplitudes span many orders of magnitude, the resulting SQD energies are largely insensitive to this variation. In particular, most initializations recover energies within chemical accuracy (+/-1.6 mEh) of the CCSD reference, with convergence improving as the basis set size increases. Notably, random initialization achieves performance competitive with CCSD across all basis sets, while zeroes initialization, despite having smaller deviations from CCSD, yields the worst energy agreement. Our results highlight that the proximity to the CCSD initialization is not a reliable predictor of the quality of electronic energies. These findings establish that configuration recovery within SQD, rather than circuit initialization, is the dominant factor governing energy accuracy, and suggest that computationally cheaper initialization strategies are viable alternatives to CCSD for QCSC workflows"
+preprint_url = "https://arxiv.org/abs/2606.14933"
 +++
diff --git a/content/publications/bert4nilm-a-bidirectional-transformer-model-for-non-intrusive-load-monitoring.md b/content/publications/bert4nilm-a-bidirectional-transformer-model-for-non-intrusive-load-monitoring.md
index 00862d4..9888936 100644
--- a/content/publications/bert4nilm-a-bidirectional-transformer-model-for-non-intrusive-load-monitoring.md
+++ b/content/publications/bert4nilm-a-bidirectional-transformer-model-for-non-intrusive-load-monitoring.md
@@ -4,7 +4,7 @@ year = 2020
 authors = ["Zhenrui Yue", "Camilo Requena Witzig", "Daniel Jorde", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 5th International Workshop on Non-Intrusive Load Monitoring"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3427771.3429390"
 abstract = "Non-intrusive load monitoring (NILM) based energy disaggregation is the decomposition of a system's energy into the consumption of its individual appliances. Previous work on deep learning NILM algorithms has shown great potential in the field of energy management and smart grids. In this paper, we propose BERT4NILM, an architecture based on bidirectional encoder representations from transformers (BERT) and an improved objective function designed specifically for NILM learning. We adapt the bidirectional transformer architecture to the field of energy disaggregation and follow the pattern of sequence-to-sequence learning. With the improved loss function and masked training, BERT4NILM outperforms state-of-the-art models across various metrics on the two publicly available datasets UK-DALE and REDD."
 +++
diff --git a/content/publications/beyond-message-passing-modern-gnn-architectures-for-online-planner-selection.md b/content/publications/beyond-message-passing-modern-gnn-architectures-for-online-planner-selection.md
index 591a9f3..f19e2fe 100644
--- a/content/publications/beyond-message-passing-modern-gnn-architectures-for-online-planner-selection.md
+++ b/content/publications/beyond-message-passing-modern-gnn-architectures-for-online-planner-selection.md
@@ -4,7 +4,7 @@ year = 2026
 authors = ["Jana Vatter", "Ruben Mayer", "Hans-Arno Jacobsen", "Horst Samulowitz", "Michael Katz"]
 venue = "Proceedings of the International Conference on Automated Planning and Scheduling"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1609/icaps.v36i1.42845"
 abstract = "As planning is computationally hard, the performance of automated planners varies greatly across planning tasks. Thus, the ability to predict planner performance on a given task is of great importance. While various learning methods have been applied in cost-optimal planning, Graph Neural Networks (GNNs) were found to perform well. However, existing work only explores a limited range of homogeneous GNN architectures and focuses primarily on the model perspective. We address these limitations by approaching the problem from both data and model perspectives. From the data perspective, we analyze the planners' performance data in terms of Shapley values to assess the potential contribution of individual planners to a portfolio. Our insights enable us to effectively reduce the portfolio from 17 to 6 planners, improving practicality and performance. From the model perspective, our work extends previous investigations of homogeneous graphs by modeling planning tasks as heterogeneous graphs and applying the heterogeneous Relational Graph Convolutional Network (RGCN) and Relational Graph Attention Network (RGAT) models. To analyze the problem in more depth, we thoroughly investigate the impact of GNN model, graph representation, node features, and prediction task. Going further, we propose a hybrid approach in which graph representations obtained by GNNs are used as input to a classical machine learning model (XGBoost), resulting in both a more resource-efficient and accurate approach. Our best model (RGCN+XGBoost) achieves 91.7% accuracy, a substantial improvement over previous methods with 87%, while requiring fewer computational resources. Overall, we demonstrate the effectiveness of heterogeneous GNN-based online planner selection methods, opening up new exciting avenues for future research."
 +++
diff --git a/content/publications/beyond-performance-measuring-the-environmental-impact-of-analytical-databases.md b/content/publications/beyond-performance-measuring-the-environmental-impact-of-analytical-databases.md
index b0f3a62..332a93e 100644
--- a/content/publications/beyond-performance-measuring-the-environmental-impact-of-analytical-databases.md
+++ b/content/publications/beyond-performance-measuring-the-environmental-impact-of-analytical-databases.md
@@ -5,6 +5,6 @@ authors = ["Michail Bachras", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/2504.18980"
+preprint_url = "https://arxiv.org/abs/2504.18980"
 abstract = "The exponential growth of data is making query processing increasingly critical for modern computing infrastructure, yet the environmental impact of database operations remains poorly understood and largely overlooked. This paper presents ATLAS, a comprehensive methodology for measuring and quantifying the environmental footprint of analytical database systems, considering both operational impacts and manufacturing costs of hardware components. Through extensive empirical evaluation of four distinct database architectures (DuckDB, MonetDB, Hyper, and StarRocks), we uncover how fundamental architectural decisions affect environmental efficiency. Our findings reveal that environmental considerations in database operations are multifaceted, encompassing both immediate operational impacts and long-term sustainability implications. We demonstrate that architectural choices can significantly influence both power consumption and environmental sustainability, while deployment location emerges as a critical factor that can amplify or diminish these architectural advantages."
 +++
diff --git a/content/publications/blockaim-a-neural-network-based-intelligent-middleware-for-large-scale-iot-data-placement-decisions.md b/content/publications/blockaim-a-neural-network-based-intelligent-middleware-for-large-scale-iot-data-placement-decisions.md
index f53efc3..cab9cc0 100644
--- a/content/publications/blockaim-a-neural-network-based-intelligent-middleware-for-large-scale-iot-data-placement-decisions.md
+++ b/content/publications/blockaim-a-neural-network-based-intelligent-middleware-for-large-scale-iot-data-placement-decisions.md
@@ -4,7 +4,7 @@ year = 2023
 authors = ["Syed Muhammad Danish", "Kaiwen Zhang", "Hans-Arno Jacobsen"]
 venue = "IEEE Transactions on Mobile Computing"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 external_url = "https://doi.org/10.1109/tmc.2021.3071576"
 abstract = "Current Internet of Things (IoT) infrastructures rely on cloud storage however, relying on a single cloud provider puts limitations on the IoT applications and Service Level Agreement (SLA) requirements. Recently, multiple decentralized storage solutions (e.g., based on blockchains) have entered the market with distinct architecture, Quality of Service (QoS) parameters and at lower price compared to the cloud storage. In this work, we introduce BAM: a neural network-based middleware designed for intelligent selection of storage technology for IoT applications. We first propose a blockchain-based data placement protocol and theoretically model a decision optimization problem, which jointly considers cloud, multi-cloud and decentralized storage technologies to select the appropriate medium to store large-scale IoT data, while ensuring data integrity, traceability, auditability and decision verifiability. We then propose a neural network-based maintenance reconfiguration, which aims to optimize the computational complexity of the middleware design along with the blockchain transaction and storage overhead by learning and predicting the applications parameters. We also propose the aggregation rate feedback functionality in our design and model it as a linear optimization problem to improve data quality and precision. Finally, we provide a reference implementation and perform extensive experiments, which demonstrate the effectiveness of the proposed design."
 +++
diff --git a/content/publications/cabinet-dynamically-weighted-consensus-made-fast.md b/content/publications/cabinet-dynamically-weighted-consensus-made-fast.md
index 4d44a23..16ba2e9 100644
--- a/content/publications/cabinet-dynamically-weighted-consensus-made-fast.md
+++ b/content/publications/cabinet-dynamically-weighted-consensus-made-fast.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.14778/3718057.3718071"
 abstract = "Conventional consensus algorithms, such as Paxos and Raft, encounter inefficiencies when applied to large-scale distributed systems due to the requirement of waiting for replies from a majority of nodes. To address these challenges, we propose Cabinet, a novel consensus algorithm that introduces dynamically weighted consensus, allocating distinct weights to nodes based on any given failure thresholds. Cabinet dynamically adjusts nodes' weights according to their responsiveness, assigning higher weights to faster nodes. The dynamic weight assignment maintains an optimal system performance, especially in large-scale and heterogeneous systems where node responsiveness varies. We evaluate Cabinet against Raft with distributed MongoDB and PostgreSQL databases using YCSB and TPC-C workloads. The evaluation results show that Cabinet outperforms Raft in throughput and latency under increasing system scales, complex networks, and failures in both homogeneous and heterogeneous clusters, offering a promising high-performance consensus solution."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2503.08914"
 +++
diff --git a/content/publications/can-graph-reordering-speed-up-graph-neural-network-training.md b/content/publications/can-graph-reordering-speed-up-graph-neural-network-training.md
index d52cc25..977f5bb 100644
--- a/content/publications/can-graph-reordering-speed-up-graph-neural-network-training.md
+++ b/content/publications/can-graph-reordering-speed-up-graph-neural-network-training.md
@@ -4,8 +4,10 @@ year = 2024
 authors = ["Nikolai Merkel", "Pierre Toussing", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the VLDB Endowment"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["graph-systems", "performance", "distributed-training"]
 external_url = "https://doi.org/10.14778/3705829.3705846"
 abstract = "Graph neural networks (GNNs) are a type of neural network capable of learning on graph-structured data. However, training GNNs on large-scale graphs is challenging due to iterative aggregations of high-dimensional features from neighboring vertices within sparse graph structures combined with neural network operations. The sparsity of graphs frequently results in suboptimal memory access patterns and longer training time. Graph reordering is an optimization strategy aiming to improve the graph data layout. It has shown to be effective to speed up graph analytics workloads, but its effect on the performance of GNN training has not been investigated yet. The generalization of reordering to GNN performance is nontrivial, as multiple aspects must be considered: GNN hyper-parameters such as the number of layers, the number of hidden dimensions, and the feature size used in the GNN model, neural network operations, large intermediate vertex states, and GPU acceleration. In our work, we close this gap by performing an empirical evaluation of 12 reordering strategies in two state-of-the-art GNN systems, PyTorch Geometric and Deep Graph Library. Our results show that graph reordering is effective in reducing training time for CPU- and GPU-based training, respectively. Further, we find that GNN hyper-parameters influence the effectiveness of reordering, that reordering metrics play an important role in selecting a reordering strategy, that lightweight reordering performs better for GPU-based than for CPU-based training, and that invested reordering time can in many cases be amortized."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2409.11129"
 +++
diff --git a/content/publications/challenger-2-0.md b/content/publications/challenger-2-0.md
index d9bd3cc..03b0e99 100644
--- a/content/publications/challenger-2-0.md
+++ b/content/publications/challenger-2-0.md
@@ -4,7 +4,7 @@ year = 2024
 authors = ["Jawad Tahir", "Chiheb Baili", "Matej Svaral", "Johannes Friedlein", "Christoph Doblander", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 18th ACM International Conference on Distributed and Event-based Systems"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 tags = ["benchmarking", "resilience", "event-processing"]
 external_url = "https://doi.org/10.1145/3629104.3666027"
 abstract = "The DEBS Grand Challenge (GC) is a yearly programming competition organized by the DEBS community. The participants of the GC are provided with a dataset and are required to build a solution generating insights from the data. Participants deploy their solutions on the provided virtual machines (VMs). The dataset is disseminated and the solutions' performance is measured using Challenger, an RPC-based service. Developer surveys show a lower adaption of RPC, which may limit the audience of the GC. Furthermore, provisioning of VMs blocks the compute resources, setting a limit on the number of participants. Lastly, Challenger lacks the functionality to test the fault-tolerance capabilities of the solutions, which is a strict non-functional requirement for the solutions."
diff --git a/content/publications/choosing-a-classical-planner-with-graph-neural-networks.md b/content/publications/choosing-a-classical-planner-with-graph-neural-networks.md
index 984c0c3..8876592 100644
--- a/content/publications/choosing-a-classical-planner-with-graph-neural-networks.md
+++ b/content/publications/choosing-a-classical-planner-with-graph-neural-networks.md
@@ -4,7 +4,7 @@ year = 2024
 authors = ["Jana Vatter", "Ruben Mayer", "Hans-Arno Jacobsen", "Horst Samulowitz", "Michael Katz"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2402.04874"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2402.04874"
 abstract = "Online planner selection is the task of choosing a solver out of a predefined set for a given planning problem. As planning is computationally hard, the performance of solvers varies greatly on planning problems. Thus, the ability to predict their performance on a given problem is of great importance. While a variety of learning methods have been employed, for classical cost-optimal planning the prevailing approach uses Graph Neural Networks (GNNs). In this work, we continue the line of work on using GNNs for online planner selection. We perform a thorough investigation of the impact of the chosen GNN model, graph representation and node features, as well as prediction task. Going further, we propose using the graph representation obtained by a GNN as an input to the Extreme Gradient Boosting (XGBoost) model, resulting in a more resource-efficient yet accurate approach. We show the effectiveness of a variety of GNN-based online planner selection methods, opening up new exciting avenues for research on online planner selection."
 +++
diff --git a/content/publications/coddora-co2-based-occupancy-detection-model-trained-via-domain-randomization.md b/content/publications/coddora-co2-based-occupancy-detection-model-trained-via-domain-randomization.md
index cc1af17..073eeff 100644
--- a/content/publications/coddora-co2-based-occupancy-detection-model-trained-via-domain-randomization.md
+++ b/content/publications/coddora-co2-based-occupancy-detection-model-trained-via-domain-randomization.md
@@ -4,7 +4,7 @@ year = 2024
 authors = ["Manuel Weber", "Farzan Banihashemi", "Davor Stjelja", "Peter Mandl", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "2024 International Joint Conference on Neural Networks (IJCNN)"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1109/ijcnn60899.2024.10650820"
 abstract = "Information about human presence in indoor spaces is crucial for building energy optimization. While there has been a considerable amount of research on using neural networks to automatically detect occupancy from CO2 sensors, their application in practice is limited due to the scarcity of labeled training data. In this paper, we propose Coddora, an off-the-shelf deep learning model pretrained on data from randomized room simulations. Coddora enables quick adaptation to real-world rooms, requiring only minimal data collection. Our contribution includes two model variants for application via fine-tuning or zero-shot classifying, as well as the synthetic dataset providing data from simulations with 100,000 room models."
 +++
diff --git a/content/publications/decentralization-in-pos-blockchain-consensus-quantification-and-advancement.md b/content/publications/decentralization-in-pos-blockchain-consensus-quantification-and-advancement.md
index c288d14..c19f8d1 100644
--- a/content/publications/decentralization-in-pos-blockchain-consensus-quantification-and-advancement.md
+++ b/content/publications/decentralization-in-pos-blockchain-consensus-quantification-and-advancement.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.1109/tnsm.2025.3561098"
 abstract = "Decentralization is a foundational principle of permissionless blockchains, with consensus mechanisms serving a critical role in its realization. This study quantifies the decentralization of consensus mechanisms in proof-of-stake (PoS) blockchains using a comprehensive set of metrics, including Nakamoto coefficients, Gini, Herfindahl-Hirschman Index (HHI), Shapley values, and Zipf’s coefficient. Our empirical analysis across ten prominent blockchains reveals significant concentration of stake among a few validators, posing challenges to fair consensus. To address this, we introduce two alternative weighting models for PoS consensus: Square Root Stake Weight (SRSW) and Logarithmic Stake Weight (LSW), which adjust validator influence through non-linear transformations. Results demonstrate that SRSW and LSW models improve decentralization metrics by an average of 51% and 132%, respectively, supporting more equitable and resilient blockchain systems."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2504.14351"
 +++
diff --git a/content/publications/decentralizing-permissioned-blockchain-with-delay-towers.md b/content/publications/decentralizing-permissioned-blockchain-with-delay-towers.md
index 8f4ac90..7f47aac 100644
--- a/content/publications/decentralizing-permissioned-blockchain-with-delay-towers.md
+++ b/content/publications/decentralizing-permissioned-blockchain-with-delay-towers.md
@@ -6,6 +6,6 @@ venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
 tags = ["blockchain"]
-external_url = "https://arxiv.org/abs/2203.09714"
+preprint_url = "https://arxiv.org/abs/2203.09714"
 abstract = "Growing excitement around permissionless blockchains is uncovering its latent scalability concerns. Permissioned blockchains offer high transactional throughput and low latencies while compromising decentralization. In the quest for a decentralized, scalable blockchain fabric, i.e., to offer the scalability of permissioned blockchain in a permissionless setting, we present L4L to encourage decentralization over the permissioned Libra network without compromising its sustainability. L4L employs delay towers, -- puzzle towers that leverage verifiable delay functions -- for establishing identity in a permissionless setting. Delay towers cannot be parallelized due to their sequential execution, making them an eco-friendly alternative. We also discuss methodologies to replace validators participating in consensus to promote compliant behavior. Our evaluations found that the cost of enabling decentralization over permissioned networks is almost negligible. Furthermore, delay towers offer an alternative to existing permissionless consensus mechanisms without requiring airdrops or pre-sale of tokens."
 +++
diff --git a/content/publications/dgfindex-for-smart-grid-enhancing-hive-with-a-cost-effective-multidimensional-range-index.md b/content/publications/dgfindex-for-smart-grid-enhancing-hive-with-a-cost-effective-multidimensional-range-index.md
index 9fb818c..f0602ea 100644
--- a/content/publications/dgfindex-for-smart-grid-enhancing-hive-with-a-cost-effective-multidimensional-range-index.md
+++ b/content/publications/dgfindex-for-smart-grid-enhancing-hive-with-a-cost-effective-multidimensional-range-index.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.14778/2733004.2733021"
 abstract = "In Smart Grid applications, as the number of deployed electric smart meters increases, massive amounts of valuable meter data is generated and collected every day. To enable reliable data collection and make business decisions fast, high throughput storage and high-performance analysis of massive meter data become crucial for grid companies. Considering the advantage of high efficiency, fault tolerance, and price-performance of Hadoop and Hive systems, they are frequently deployed as underlying platform for big data processing. However, in real business use cases, these data analysis applications typically involve multidimensional range queries (MDRQ) as well as batch reading and statistics on the meter data. While Hive is high-performance at complex data batch reading and analysis, it lacks efficient indexing techniques for MDRQ. In this paper, we propose DGFIndex, an index structure for Hive that efficiently supports MDRQ for massive meter data. DGFIndex divides the data space into cubes using the grid file technique. Unlike the existing indexes in Hive, which stores all combinations of multiple dimensions, DGFIndex only stores the information of cubes. This leads to smaller index size and faster query processing. Furthermore, with pre-computing user-defined aggregations of each cube, DGFIndex only needs to access the boundary region for aggregation query. Our comprehensive experiments show that DGFIndex can save significant disk space in comparison with the existing indexes in Hive and the query performance with DGFIndex is 2-50 times faster than existing indexes in Hive and HadoopDB for aggregation query, 2-5 times faster than both for non-aggregation query, 2-75 times faster than scanning the whole table in different query selectivity."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1404.5686"
 +++
diff --git a/content/publications/diba-a-re-configurable-stream-processor.md b/content/publications/diba-a-re-configurable-stream-processor.md
index 15c5777..55137a4 100644
--- a/content/publications/diba-a-re-configurable-stream-processor.md
+++ b/content/publications/diba-a-re-configurable-stream-processor.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.1109/tkde.2024.3381192"
 abstract = "Stream processing acceleration is driven by the continuously increasing volume and velocity of data generated on the Web and the limitations of storage, computation, and power consumption. Hardware solutions provide better performance and power consumption, but they are hindered by the high research and development costs and the long time to market. In this work, we propose our re-configurable stream processor (Diba), a complete rethinking of a previously proposed customized and flexible query processor that targets real-time stream processing. Diba uses a unidirectional dataflow not dedicated to any specific type of query (operator) on streams, allowing a straightforward placement of processing components on a general data path that facilitates query mapping. In Diba, the concepts of the distribution network and processing components are implemented as two separate entities connected using generic interfaces. This approach allows the adoption of a versatile architecture for a family of queries rather than forcing a rigid chain of processing components to implement such queries. Our experimental evaluations of representative queries from TPC-H yielded processing times of 300, 1220, and 3520 milliseconds for data streams with scale factor sizes of one, four, and ten gigabytes, respectively."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2304.01659"
 +++
diff --git a/content/publications/distributed-quantum-computing-for-chemical-applications.md b/content/publications/distributed-quantum-computing-for-chemical-applications.md
index 648c632..0cc39c9 100644
--- a/content/publications/distributed-quantum-computing-for-chemical-applications.md
+++ b/content/publications/distributed-quantum-computing-for-chemical-applications.md
@@ -8,4 +8,6 @@ research = ["quantum-computing-systems"]
 tags = ["quantum-systems", "quantum-chemistry", "distributed-computing"]
 external_url = "https://doi.org/10.1109/qce60285.2024.10270"
 abstract = "In recent years, interest in quantum computing has increased due to technological advances in quantum hardware and algorithms. Despite the promises of quantum advantage, the applicability of quantum devices has been limited to few qubits on hardware that experiences decoherence due to noise. One proposed method to get around this challenge is distributed quantum computing (DQC). Like classical distributed computing, DQC aims at increasing compute power by spreading the compute processes across many devices, with the goal to minimize the noise and circuit depth required by quantum devices. In this paper, we cover the fundamental concepts of DQC and provide insight into where the field of DQC stands with respect to the field of chemistry -- a field which can potentially be used to demonstrate quantum advantage on noisy-intermediate scale quantum devices."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2408.05351"
 +++
diff --git a/content/publications/dualtable-a-hybrid-storage-model-for-update-optimization-in-hive.md b/content/publications/dualtable-a-hybrid-storage-model-for-update-optimization-in-hive.md
index 53a08aa..2a7dcdb 100644
--- a/content/publications/dualtable-a-hybrid-storage-model-for-update-optimization-in-hive.md
+++ b/content/publications/dualtable-a-hybrid-storage-model-for-update-optimization-in-hive.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1109/icde.2015.7113381"
 abstract = "Hive is the most mature and prevalent data warehouse tool providing SQL-like interface in the Hadoop ecosystem. It is successfully used in many Internet companies and shows its value for big data processing in traditional industries. However, enterprise big data processing systems as in Smart Grid applications usually require complicated business logics and involve many data manipulation operations like updates and deletes. Hive cannot offer sufficient support for these while preserving high query performance. Hive using the Hadoop Distributed File System (HDFS) for storage cannot implement data manipulation efficiently and Hive on HBase suffers from poor query performance even though it can support faster data manipulation. There is a project based on Hive issue Hive-5317 to support update operations, but it has not been finished in Hive's latest version. Since this ACID compliant extension adopts same data storage format on HDFS, the update performance problem is not solved. In this paper, we propose a hybrid storage model called DualTable, which combines the efficient streaming reads of HDFS and the random write capability of HBase. Hive on DualTable provides better data manipulation support and preserves query performance at the same time. Experiments on a TPC-H data set and on a real smart grid data set show that Hive on DualTable is up to 10 times faster than Hive when executing update and delete operations."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1404.6878"
 +++
diff --git a/content/publications/dynamic-loss-based-sample-reweighting-for-improved-large-language-model-pretraining.md b/content/publications/dynamic-loss-based-sample-reweighting-for-improved-large-language-model-pretraining.md
index e3a072d..07b4bb1 100644
--- a/content/publications/dynamic-loss-based-sample-reweighting-for-improved-large-language-model-pretraining.md
+++ b/content/publications/dynamic-loss-based-sample-reweighting-for-improved-large-language-model-pretraining.md
@@ -4,7 +4,9 @@ year = 2025
 authors = ["Daouda Sow", "Herbert Woisetschläger", "Saikiran Bulusu", "Shiqiang Wang", "Hans-Arno Jacobsen", "Yingbin Liang"]
 venue = "ICLR"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://openreview.net/forum?id=gU4ZgQNsOC"
 abstract = "Pretraining large language models (LLMs) on vast and heterogeneous datasets is crucial for achieving state-of-the-art performance across diverse downstream tasks. However, current training paradigms treat all samples equally, overlooking the importance or relevance of individual samples throughout the training process. Existing reweighting strategies, which primarily focus on group-level data importance, fail to leverage fine-grained instance-level information and do not adapt dynamically to individual sample importance as training progresses. In this paper, we introduce novel algorithms for dynamic, instance-level data reweighting aimed at improving both the efficiency and effectiveness of LLM pretraining. Our methods adjust the weight of each training sample based on its loss value in an online fashion, allowing the model to dynamically focus on more informative or important samples at the current training stage. In particular, our framework allows us to systematically devise reweighting strategies deprioritizing redundant or uninformative data, which we find tend to work best. Furthermore, we develop a new theoretical framework for analyzing the impact of loss-based reweighting on the convergence of gradient-based optimization, providing the first formal characterization of how these strategies affect convergence bounds. We empirically validate our approach across a spectrum of tasks, from pretraining 7B and 1.4B parameter LLMs to smaller-scale language models and linear regression problems, demonstrating that our loss-based reweighting approach can lead to faster convergence and significantly improved performance."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2502.06733"
 +++
diff --git a/content/publications/efficient-federated-learning-methods-for-foundation-model-training.md b/content/publications/efficient-federated-learning-methods-for-foundation-model-training.md
index 7859fea..50e2d71 100644
--- a/content/publications/efficient-federated-learning-methods-for-foundation-model-training.md
+++ b/content/publications/efficient-federated-learning-methods-for-foundation-model-training.md
@@ -4,8 +4,10 @@ year = 2024
 authors = ["Herbert Woisetschläger", "Alexander Erben", "Shiqiang Wang", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI)"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["federated-learning", "foundation-models", "systems-survey"]
 external_url = "https://www.ijcai.org/proceedings/2024/919"
 abstract = "Federated Learning (FL) has become an established technique to facilitate privacy-preserving collaborative training across a multitude of clients. However, new approaches to FL often discuss their contributions involving small deep-learning models only and focus on training full models on clients. In the wake of Foundation Models (FM), the reality is different for many deep learning applications. Typically, FMs have already been pre-trained across a wide variety of tasks and can be fine-tuned to specific downstream tasks over significantly smaller datasets than required for full model training. However, access to such datasets is often challenging. By its design, FL can help to open data silos. With this survey, we introduce a novel taxonomy focused on computational and communication efficiency, the vital elements to make use of FMs in FL systems. We discuss the benefits and drawbacks of parameter-efficient fine-tuning (PEFT) for FL applications, elaborate on the readiness of FL frameworks to work with FMs, and provide future research opportunities on how to evaluate generative models in FL as well as the interplay of privacy and PEFT."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2401.04472"
 +++
diff --git a/content/publications/electrical-appliance-classification-using-deep-convolutional-neural-networks-on-high-frequency-current-measurements.md b/content/publications/electrical-appliance-classification-using-deep-convolutional-neural-networks-on-high-frequency-current-measurements.md
index 32b1ef7..b772007 100644
--- a/content/publications/electrical-appliance-classification-using-deep-convolutional-neural-networks-on-high-frequency-current-measurements.md
+++ b/content/publications/electrical-appliance-classification-using-deep-convolutional-neural-networks-on-high-frequency-current-measurements.md
@@ -4,7 +4,7 @@ year = 2018
 authors = ["Daniel Jorde", "Thomas Kriechbaumer", "Hans-Arno Jacobsen"]
 venue = "2018 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1109/smartgridcomm.2018.8587452"
 abstract = "Monitoring the energy demand of appliances can raise consumer awareness and therefore reduce energy consumption. Using a single-point measurement of mains energy consumption can keep costs and hardware complexity to a minimum. This data stream of raw voltage and current measurements can be used in machine learning tasks to extract information. We apply Deep Convolutional Neural Networks on an electrical appliance classification task, using raw high frequency start up events from two datasets. We further present Data Augmentation techniques to improve the model performance and evaluate different data normalization techniques. We achieve a perfect classification on WHITED and a Fl-Score of 0.69 on PLAID."
 +++
diff --git a/content/publications/embedpart-embedding-driven-graph-partitioning-for-scalable-graph-neural-network-training.md b/content/publications/embedpart-embedding-driven-graph-partitioning-for-scalable-graph-neural-network-training.md
index e786ce3..b6979d4 100644
--- a/content/publications/embedpart-embedding-driven-graph-partitioning-for-scalable-graph-neural-network-training.md
+++ b/content/publications/embedpart-embedding-driven-graph-partitioning-for-scalable-graph-neural-network-training.md
@@ -4,7 +4,7 @@ year = 2026
 authors = ["Nikolai Merkel", "Ruben Mayer", "Volker Markl", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning", "data-management"]
-external_url = "https://arxiv.org/abs/2604.01000"
+research = ["rca-aiops", "data-management"]
+preprint_url = "https://arxiv.org/abs/2604.01000"
 abstract = "Graph Neural Networks (GNNs) are widely used for learning on graph-structured data, but scaling GNN training to massive graphs remains challenging. To enable scalable distributed training, graphs are divided into smaller partitions that are distributed across multiple machines such that inter-machine communication is minimized and computational load is balanced. In practice, existing partitioning approaches face a fundamental trade-off between partitioning overhead and partitioning quality. We propose EmbedPart, an embedding-driven partitioning approach that achieves both speed and quality. Instead of operating directly on irregular graph structures, EmbedPart leverages node embeddings produced during the actual GNN training workload and clusters these dense embeddings to derive a partitioning. EmbedPart achieves more than 100x speedup over Metis while maintaining competitive partitioning quality and accelerating distributed GNN training. Moreover, EmbedPart naturally supports graph updates and fast repartitioning, and can be applied to graph reordering to improve data locality and accelerate single-machine GNN training. By shifting partitioning from irregular graph structures to dense embeddings, EmbedPart enables scalable and high-quality graph data optimization."
 +++
diff --git a/content/publications/energy-vs-privacy-estimating-the-ecological-impact-of-federated-learning.md b/content/publications/energy-vs-privacy-estimating-the-ecological-impact-of-federated-learning.md
index 5d90d5d..05aaa74 100644
--- a/content/publications/energy-vs-privacy-estimating-the-ecological-impact-of-federated-learning.md
+++ b/content/publications/energy-vs-privacy-estimating-the-ecological-impact-of-federated-learning.md
@@ -4,7 +4,7 @@ year = 2023
 authors = ["René Schwermer", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 14th ACM International Conference on Future Energy Systems"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3575813.3597344"
 abstract = "The increasing usage of edge devices and stricter data privacy regulations motivate the use of federated learning (FL). At the same time, more and more stakeholders are concerned about the ecological impact of machine learning and its associate network traffic. The current research in FL does not investigate the impact of different network constraints and privacy-enhancing techniques, such as differential privacy, on the network traffic and energy consumption of the clients. Most experiments run either on virtual machines or on one machine with simulated clients. In such environments, it is challenging to measure each client’s network and energy usage. Therefore, we built our \"Distributed Edge Device Testbed\" (DEDT) and evaluate a convolutional neural network trained on the MNIST data set under different network constraints on DEDT, with differential privacy and with an increasing amount of participating clients. For each experiment, we quantify the network traffic, energy consumption, and training time. The results show the importance of experiments on physically separated nodes and the need to improve software-based power monitoring. The estimated energy consumption deviates by up to 35 % from the measured ones. The accuracy of the estimated network traffic depends on the monitored network interface and gives an error of 18 % for virtual machines in combination with monitoring the Ethernet interface. The training time also increases linearly with the number of participating clients."
 +++
diff --git a/content/publications/epoch-based-optimistic-concurrency-control-in-geo-replicated-databases.md b/content/publications/epoch-based-optimistic-concurrency-control-in-geo-replicated-databases.md
index de2748f..0608714 100644
--- a/content/publications/epoch-based-optimistic-concurrency-control-in-geo-replicated-databases.md
+++ b/content/publications/epoch-based-optimistic-concurrency-control-in-geo-replicated-databases.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3802052"
 abstract = "Achieving high-performance transaction processing in geo-replicated OLTP databases is challenging due to the extensive over-coordination in distributed atomic commitment, concurrency control, and fault-tolerant replication protocols. To address this issue, we introduce Minerva, a unified distributed concurrency control protocol designed for highly scalable multi-leader replication. Minerva employs a novel epoch-based asynchronous replication protocol that decouples data propagation from the commitment process, enabling continuous transaction replication. Optimistic concurrency control is used to allow replicas to execute transactions concurrently and to commit without coordination. For conflict detection during validation, we construct a conflict graph and use a maximum weight independent set search algorithm to select the optimal subset of non-conflicting transactions for commitment, minimizing the number of invalid transactions. Finally, we deterministically re-execute conflicting transactions, ensuring serializability while eliminating aborts. Our evaluation demonstrates that Minerva outperforms state-of-the-art replicated databases, achieving over 3x higher throughput in scalability experiments and 2.8x higher throughput in a high-latency network simulation with the TPC-C benchmark."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2602.21566"
 +++
diff --git a/content/publications/escape-to-precaution-against-leader-failures.md b/content/publications/escape-to-precaution-against-leader-failures.md
index 62be767..b49d9f9 100644
--- a/content/publications/escape-to-precaution-against-leader-failures.md
+++ b/content/publications/escape-to-precaution-against-leader-failures.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = []
 external_url = "https://doi.org/10.1109/icdcs54860.2022.00066"
 abstract = "Leader-based consensus protocols must undergo a view-change phase to elect a new leader when the current leader fails. The new leader often comes from a candidate server that collects votes from a quorum of servers. However, voting-based election mechanisms intrinsically incite competition in leadership candidacy since candidates may collect only partial votes. This split-vote scenario can result in no leadership winner and thus prolongs the undesired view-change period. In this paper, we investigate a case study of Raft’s leader election and propose a new leader election protocol, called ESCAPE, that fundamentally solves split votes by prioritizing servers based on their log responsiveness. ESCAPE dynamically distributes configurations that offer different priorities to servers through periodic heartbeats. In each assignment, ESCAPE assigns configurations that are more inclined to win an election to servers that have more up-to-date log responsiveness, thereby preparing a pool of prioritized candidates. Consequently, when the next election takes place, the candidate with the highest priority can defeat its counterparts and becomes the next leader without competition. The evaluation results show that ESCAPE progressively reduces the leader election time when the cluster scales up, and the improvement becomes more significant under message loss."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2202.09434"
 +++
diff --git a/content/publications/evaluating-interwell-connectivity-in-waterflooding-reservoirs-with-graph-based-cooperation-mission-neural-networks.md b/content/publications/evaluating-interwell-connectivity-in-waterflooding-reservoirs-with-graph-based-cooperation-mission-neural-networks.md
index 93abf95..6e04fb6 100644
--- a/content/publications/evaluating-interwell-connectivity-in-waterflooding-reservoirs-with-graph-based-cooperation-mission-neural-networks.md
+++ b/content/publications/evaluating-interwell-connectivity-in-waterflooding-reservoirs-with-graph-based-cooperation-mission-neural-networks.md
@@ -4,7 +4,7 @@ year = 2022
 authors = ["Xingjie Zeng", "Weishan Zhang", "Tao Chen", "Hans-Arno Jacobsen", "Jiehan Zhou", "Bingyang Chen"]
 venue = "SPE Journal"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.2118/209607-pa"
 abstract = "Summary Interwell connectivity plays a key role in waterflooding for guiding water injection. The existing works focus on the response relationship between one injection well and one production well. No research has explored the structural information of waterflooding on a well pattern. To address this challenge, this paper proposes cooperation-mission neural networks for interwell connectivity with graph information. Specifically, we propose some assumptions based on the petroleum domain to represent the well pattern with an adjacent matrix of the graph. Then we propose two targets from the view of injection well groups and production well groups. Accordingly, we propose cooperation-mission neural networks from these two aspects to evaluate the interwell connectivity in the well pattern. We test our model from two perspectives: the accuracy of estimation with tracer and the graduality of interwell connectivity. The results demonstrate that our model makes a good performance and achieves the connectivity analysis accuracy rate of 91.4%. Moreover, this study demonstrates that it is practical to evaluate the interwell connectivity with graph."
 +++
diff --git a/content/publications/experiment-benchmark-paper-to-what-extent-does-quality-matter-the-impact-of-graph-data-quality-on-gnn-model-performance.md b/content/publications/experiment-benchmark-paper-to-what-extent-does-quality-matter-the-impact-of-graph-data-quality-on-gnn-model-performance.md
index e7fd485..50a0463 100644
--- a/content/publications/experiment-benchmark-paper-to-what-extent-does-quality-matter-the-impact-of-graph-data-quality-on-gnn-model-performance.md
+++ b/content/publications/experiment-benchmark-paper-to-what-extent-does-quality-matter-the-impact-of-graph-data-quality-on-gnn-model-performance.md
@@ -4,7 +4,7 @@ year = 2025
 authors = ["Jana Vatter", "Maurice L. Rochau", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "VLDB Workshops"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://www.vldb.org/2025/Workshops/VLDB-Workshops-2025/LSGDA/LSGDA25_05.pdf"
 abstract = "Real-world data often is noisy and error-prone which can negatively influence machine learning models. Graph Neural Networks (GNNs) introduce the additional challenge that during training, node information is iteratively passed through the graph along the edges. Consequently, errors or deviations in the graph data could highly impact the model’s predictive capability. Our work systematically investigates how quality deviations in graph datasets influence the GNN model performance. We focus on the node features and explore three dimensions: the rate of modified features, the amplitude of modification, and the feature precision. Based on our results, we give insights and recommendations for practitioners. For instance, when using highly clustered graphs, modifying around 40% of the features only results in a slight decrease of performance and the rate of modified features is more crucial than the amplitude of modification. We illustrate practical implications of our results by establishing connections to real world domains such as graph dataset acquisition and efficient GNN training."
 abstract_license_url = "https://creativecommons.org/licenses/by-nc-nd/4.0/"
diff --git a/content/publications/fabriccrdt-a-conflict-free-replicated-datatypes-approach-to-permissioned-blockchains.md b/content/publications/fabriccrdt-a-conflict-free-replicated-datatypes-approach-to-permissioned-blockchains.md
index 655432c..91c442d 100644
--- a/content/publications/fabriccrdt-a-conflict-free-replicated-datatypes-approach-to-permissioned-blockchains.md
+++ b/content/publications/fabriccrdt-a-conflict-free-replicated-datatypes-approach-to-permissioned-blockchains.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3361525.3361540"
 abstract = "With the increased adaption of blockchain technologies, permissioned blockchains such as Hyperledger Fabric provide a robust ecosystem for developing production-grade decentralized applications. However, the additional latency between executing and committing transactions, due to Fabric's three-phase transaction lifecycle of Execute-Order-Validate (EOV), is a potential scalability bottleneck. The added latency increases the probability of concurrent updates on the same keys by different transactions, leading to transaction failures caused by Fabric's concurrency control mechanism. The transaction failures increase the application development complexity and decrease Fabric's throughput. Conflict-free Replicated Datatypes (CRDTs) provide a solution for merging and resolving conflicts in the presence of concurrent updates. In this work, we introduce FabricCRDT, an approach for integrating CRDTs to Fabric. Our evaluations show that in general, FabricCRDT offers higher throughput of successful transactions than Fabric, while successfully committing and merging all conflicting transactions without any failures."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2310.15988"
 +++
diff --git a/content/publications/fame-failure-aware-mixture-of-experts-for-message-level-log-anomaly-detection.md b/content/publications/fame-failure-aware-mixture-of-experts-for-message-level-log-anomaly-detection.md
index 0058553..67f7c5a 100644
--- a/content/publications/fame-failure-aware-mixture-of-experts-for-message-level-log-anomaly-detection.md
+++ b/content/publications/fame-failure-aware-mixture-of-experts-for-message-level-log-anomaly-detection.md
@@ -2,9 +2,17 @@
 title = "FAME: Failure-Aware Mixture-of-Experts for Message-Level Log Anomaly Detection"
 year = 2026
 authors = ["Huanchi Wang", "Zihang Huang", "Yifang Tian", "Kristina Dzeparoska", "Hans-Arno Jacobsen", "Alberto Leon-Garcia"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2605.22779"
+venue = "37th IEEE International Symposium on Software Reliability Engineering (ISSRE)"
+publication_type = "Conference Paper"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2605.22779"
 abstract = "Production systems generate millions of log lines daily, yet most anomaly detectors operate at the session or window-level, flagging groups of lines rather than identifying the specific message responsible. This coarse granularity forces operators to inspect many routine lines per alert. Message-level detection offers finer granularity, but remains challenging. A single event template may correspond to both normal and anomalous messages, failures arise from heterogeneous subsystems, and line-level labeling at scale is impractical. Although large language models (LLMs) can reason over log semantics, applying them to every line is too costly for continuous monitoring. We present FAME (Failure-Aware Mixture-of-Experts), a label-efficient message-level mixture-of-experts framework that uses an LLM only once offline. We annotate at most K labeled lines per template to derive binary normal/anomaly indicators and representative examples. The LLM proposes a partition of templates into failure domains, and a certification step validates the proposal before training. FAME trains a lightweight router and domain experts that run on-premise and output anomaly predictions and failure-domain labels. On BGL, FAME achieves F1 = 98.16 at K = 100 reducing annotation effort by 76x and detects 97.7% of anomalies from unseen EventIDs. On Thunderbird, FAME reaches F1 = 99.95 with perfect recall."
+
+# Attribution: repository URL in the paper, https://arxiv.org/abs/2605.22779
+research_outcomes = [
+  { title = "FAME", url = "https://github.com/HuanchiW95/FAME-Failure-Aware-Mixture-of-Experts", summary = "Message-level log anomaly detection using offline failure-domain grouping and a lightweight router with specialized experts." },
+]
 +++
+
+Presented at [ISSRE 2026](https://cyprusconferences.org/issre2026/program/).
+The full-text link points to the arXiv version.
diff --git a/content/publications/federated-computing-in-electric-vehicles-to-predict-coolant-temperature.md b/content/publications/federated-computing-in-electric-vehicles-to-predict-coolant-temperature.md
index decb831..b25b0cf 100644
--- a/content/publications/federated-computing-in-electric-vehicles-to-predict-coolant-temperature.md
+++ b/content/publications/federated-computing-in-electric-vehicles-to-predict-coolant-temperature.md
@@ -4,7 +4,7 @@ year = 2023
 authors = ["René Schwermer", "Ekin-Alp Bicer", "Pascal A. Schirmer", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 24th International Middleware Conference: Industrial Track"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3626562.3626829"
 abstract = "Reducing greenhouse gas emissions in mobility is paramount to achieving a carbon-neutral society. However, battery-electrical vehicles (BEV) introduce unique engineering challenges to protect expensive electrical components from overheating. A centralized architecture for model-driven predictions of coolant temperatures poses privacy and legal issues. Additionally, the applications in a vehicle compete for the available resources and must use them as sparingly as possible. Therefore, we introduce a new federated computing (FC) use case to help transform the mobility sector. We evaluate the performance of two FC approaches (linear regression and machine learning) on hardware and privacy metrics by leveraging a real-world dataset from BEVs. Our findings show trade-offs between hardware utilization and model accuracy. The linear regression model yields the best performance and prediction metrics. FC with ML shows up to 761 % variances when comparing vehicle-specific models with models trained with the entire fleet and clustering the data into velocity profiles partly improves prediction performance."
 +++
diff --git a/content/publications/federated-computing-survey-on-building-blocks-extensions-and-systems.md b/content/publications/federated-computing-survey-on-building-blocks-extensions-and-systems.md
index 23866e9..f0ce711 100644
--- a/content/publications/federated-computing-survey-on-building-blocks-extensions-and-systems.md
+++ b/content/publications/federated-computing-survey-on-building-blocks-extensions-and-systems.md
@@ -4,7 +4,7 @@ year = 2024
 authors = ["René Schwermer", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2404.02779"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2404.02779"
 abstract = "In response to the increasing volume and sensitivity of data, traditional centralized computing models face challenges, such as data security breaches and regulatory hurdles. Federated Computing (FC) addresses these concerns by enabling collaborative processing without compromising individual data privacy. This is achieved through a decentralized network of devices, each retaining control over its data, while participating in collective computations. The motivation behind FC extends beyond technical considerations to encompass societal implications. As the need for responsible AI and ethical data practices intensifies, FC aligns with the principles of user empowerment and data sovereignty. FC comprises of Federated Learning (FL) and Federated Analytics (FA). FC systems became more complex over time and they currently lack a clear definition and taxonomy describing its moving pieces. Current surveys capture domain-specific FL use cases, describe individual components in an FC pipeline individually or decoupled from each other, or provide a quantitative overview of the number of published papers. This work surveys more than 150 papers to distill the underlying structure of FC systems with their basic building blocks, extensions, architecture, environment, and motivation. We capture FL and FA systems individually and point out unique difference between those two."
 +++
diff --git a/content/publications/federated-fine-tuning-of-llms-on-the-very-edge.md b/content/publications/federated-fine-tuning-of-llms-on-the-very-edge.md
index e7b8d60..58fcab6 100644
--- a/content/publications/federated-fine-tuning-of-llms-on-the-very-edge.md
+++ b/content/publications/federated-fine-tuning-of-llms-on-the-very-edge.md
@@ -4,8 +4,10 @@ year = 2024
 authors = ["Herbert Woisetschläger", "Alexander Erben", "Shiqiang Wang", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the Eighth Workshop on Data Management for End-to-End Machine Learning"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["federated-learning", "edge-systems", "llms"]
 external_url = "https://doi.org/10.1145/3650203.3663331"
 abstract = "With the emergence of AI regulations, such as the EU AI Act, requirements for simple data lineage, enforcement of low data bias, and energy efficiency have become a priority for everyone offering AI services. Being pre-trained on versatile and a vast amount of data, large language models and foundation models (FMs) offer a good basis for building high-quality deep learning pipelines. Fine-tuning can further improve model performance on a specific downstream task, which requires orders of magnitude less data than pre-training. Often, access to high-quality and low-bias data for model fine-tuning is limited due to technical or regulatory requirements. Federated learning (FL), as a distributed and privacy-preserving technique, offers a well-suited approach to significantly expanding data access for model fine-tuning. Yet, this data is often located on the network edge, where energy, computational, and communication resources are significantly more limited than in data centers."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2310.03150"
 +++
diff --git a/content/publications/federated-learning-and-ai-regulation-in-the-european-union-who-is-responsible-an-interdisciplinary-analysis.md b/content/publications/federated-learning-and-ai-regulation-in-the-european-union-who-is-responsible-an-interdisciplinary-analysis.md
index dd6cec2..e1c0f93 100644
--- a/content/publications/federated-learning-and-ai-regulation-in-the-european-union-who-is-responsible-an-interdisciplinary-analysis.md
+++ b/content/publications/federated-learning-and-ai-regulation-in-the-european-union-who-is-responsible-an-interdisciplinary-analysis.md
@@ -4,8 +4,10 @@ year = 2024
 authors = ["Herbert Woisetschläger", "Simon Mertel", "Christoph Krönke", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "GenLaw Workshop at ICML 2024"
 publication_type = "Workshop Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["federated-learning"]
 external_url = "https://blog.genlaw.org/pdfs/genlaw_icml2024/16.pdf"
 abstract = "The European Union Artificial Intelligence Act mandates clear stakeholder responsibilities in developing and deploying machine learning applications to avoid substantial fines, prioritizing private and secure data processing with data remaining at its origin. Federated Learning (FL) enables the training of generative AI Models across data siloes, sharing only model parameters while improving data security. Since FL is a cooperative learning paradigm, clients and servers naturally share legal responsibility in the FL pipeline. Our work contributes to clarifying the roles of both parties, explains strategies for shifting responsibilities to the server operator, and points out open technical challenges that we must solve to improve FL's practical applicability under the EU AI Act."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2407.08105"
 +++
diff --git a/content/publications/federated-learning-priorities-under-the-european-union-ai-act.md b/content/publications/federated-learning-priorities-under-the-european-union-ai-act.md
index 99ccd00..5b408f7 100644
--- a/content/publications/federated-learning-priorities-under-the-european-union-ai-act.md
+++ b/content/publications/federated-learning-priorities-under-the-european-union-ai-act.md
@@ -4,8 +4,8 @@ year = 2024
 authors = ["Herbert Woisetschläger", "Alexander Erben", "Bill Marino", "Shiqiang Wang", "Nicholas D. Lane", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["federated-learning", "ai-regulation", "systems-policy"]
-external_url = "https://arxiv.org/abs/2402.05968"
+preprint_url = "https://arxiv.org/abs/2402.05968"
 abstract = "The age of AI regulation is upon us, with the European Union Artificial Intelligence Act (AI Act) leading the way. Our key inquiry is how this will affect Federated Learning (FL), whose starting point of prioritizing data privacy while performing ML fundamentally differs from that of centralized learning. We believe the AI Act and future regulations could be the missing catalyst that pushes FL toward mainstream adoption. However, this can only occur if the FL community reprioritizes its research focus. In our position paper, we perform a first-of-its-kind interdisciplinary analysis (legal and ML) of the impact the AI Act may have on FL and make a series of observations supporting our primary position through quantitative and qualitative analysis. We explore data governance issues and the concern for privacy. We establish new challenges regarding performance and energy efficiency within lifecycle monitoring. Taken together, our analysis suggests there is a sizable opportunity for FL to become a crucial component of AI Act-compliant ML systems and for the new regulation to drive the adoption of FL techniques in general. Most noteworthy are the opportunities to defend against data bias and enhance private and secure computation"
 +++
diff --git a/content/publications/federated-office-plug-load-identification-for-building-management-systems.md b/content/publications/federated-office-plug-load-identification-for-building-management-systems.md
index dcc88ac..057ba9d 100644
--- a/content/publications/federated-office-plug-load-identification-for-building-management-systems.md
+++ b/content/publications/federated-office-plug-load-identification-for-building-management-systems.md
@@ -4,7 +4,7 @@ year = 2022
 authors = ["René Schwermer", "Jonas Buchberger", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the Thirteenth ACM International Conference on Future Energy Systems"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3538637.3538845"
 abstract = "Energy consumption in buildings is responsible for 40 % of the final energy consumption in the European Union and the United States of America. In addition to thermal energy, buildings require electricity for all kinds of appliances. Regulatory constraints such as energy labels aim at increasing the energy efficiency of large appliances such as fridges and washing machines. However, they only partially cover plug-loads. The amount of electricity consumption of unregulated plug-loads such as mobile phones, USB chargers and kettles is continuously increasing. For European households, their share of electricity consumption reached 25 % in 2018. Additional data about the plug-loads usage can help decrease the energy consumption of buildings by improving energy management systems, applying peak-shaving or demand-side management. People live and work in buildings, making such data privacy sensitive. Federated Learning (FL) helps to leverage these data without violating regulatory frameworks such as the General Data Protection Regulation. We use a high-frequency energy data set of office appliances (BLOND) to train four appliance classifiers (CNN, LSTM, ResNet and DenseNet). We investigate the effect of different data distributions (entire dataset, IID and non-IID) and training methods on four performance metrics (accuracy, F1 score, precision and recall). The results show that a non-IID setup decreases all performance metrics for some model architectures by 44 %. However, our LSTM model even with a non-IID labels achieves similar F1 scores compared to central training. Additionally, we show the importance of client selection in FL architectures to reduce the overall training time and we quantify the decrease in network traffic compared to a central training approach, the energy consumption and scalability."
 +++
diff --git a/content/publications/fledge.md b/content/publications/fledge.md
index b42f69a..8a6d084 100644
--- a/content/publications/fledge.md
+++ b/content/publications/fledge.md
@@ -4,8 +4,12 @@ year = 2024
 authors = ["Herbert Woisetschläger", "Alexander Erben", "Ruben Mayer", "Shiqiang Wang", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 25th International Middleware Conference"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["federated-learning", "benchmarking", "edge-systems"]
 external_url = "https://doi.org/10.1145/3652892.3700751"
 abstract = "Federated Learning (FL) has become a viable technique for realizing privacy-enhancing distributed deep learning on the network edge. Heterogeneous hardware, unreliable client devices, and energy constraints often characterize edge computing systems. In this paper, we propose FLEdge, which complements existing FL benchmarks by enabling a systematic evaluation of client capabilities. We focus on computational and communication bottlenecks, client behavior, and data security implications. Our experiments with models varying from 14K to 80M trainable parameters are carried out on dedicated hardware with emulated network characteristics and client behavior. We find that state-of-the-art embedded hardware has significant memory bottlenecks, leading to 4× longer processing times than on modern data center GPUs."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2306.05172"
 +++
+
+The preprint title includes “Federated Machine Learning Applications”; the published title uses “Federated Learning Applications”.
diff --git a/content/publications/gala-can-graph-augmented-large-language-model-agentic-workflows-elevate-root-cause-analysis.md b/content/publications/gala-can-graph-augmented-large-language-model-agentic-workflows-elevate-root-cause-analysis.md
deleted file mode 100644
index a643ed0..0000000
--- a/content/publications/gala-can-graph-augmented-large-language-model-agentic-workflows-elevate-root-cause-analysis.md
+++ /dev/null
@@ -1,10 +0,0 @@
-+++
-title = "GALA: Can Graph-Augmented Large Language Model Agentic Workflows Elevate Root Cause Analysis?"
-year = 2025
-authors = ["Yifang Tian", "Yaming Liu", "Zichun Chong", "Zihang Huang", "Hans-Arno Jacobsen"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning", "data-management"]
-external_url = "https://arxiv.org/abs/2508.12472"
-abstract = "Root cause analysis (RCA) in microservice systems is challenging, requiring on-call engineers to rapidly diagnose failures across heterogeneous telemetry such as metrics, logs, and traces. Traditional RCA methods often focus on single modalities or merely rank suspect services, falling short of providing actionable diagnostic insights with remediation guidance. This paper introduces GALA, a novel multi-modal framework that combines statistical causal inference with LLM-driven iterative reasoning for enhanced RCA. Evaluated on an open-source benchmark, GALA achieves substantial improvements over state-of-the-art methods of up to 42.22% accuracy. Our novel human-guided LLM evaluation score shows GALA generates significantly more causally sound and actionable diagnostic outputs than existing methods. Through comprehensive experiments and a case study, we show that GALA bridges the gap between automated failure diagnosis and practical incident resolution by providing both accurate root cause identification and human-interpretable remediation guidance."
-+++
diff --git a/content/publications/gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices.md b/content/publications/gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices.md
index 93bedfc..2fb8e87 100644
--- a/content/publications/gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices.md
+++ b/content/publications/gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices.md
@@ -1,10 +1,26 @@
 +++
 title = "GALA: Graph-Augmented LLM Agents for Root Cause Analysis and Incident Response in Microservices"
+aliases = ["/publications/gala-can-graph-augmented-large-language-model-agentic-workflows-elevate-root-cause-analysis/"]
 year = 2026
 authors = ["Yifang Tian", "Yaming Liu", "Zichun Chong", "Zihang Huang", "Yiran Li", "Hans-Arno Jacobsen"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning", "data-management"]
-external_url = "https://arxiv.org/abs/2608.08968"
+venue = "Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE '26), pp. 55–67"
+publication_type = "Conference Paper"
+research = ["rca-aiops", "data-management"]
+external_url = "https://doi.org/10.1145/3832783.3834335"
+tags = ["GALA", "GALA: Can Graph-Augmented Large Language Model Agentic Workflows Elevate Root Cause Analysis?"]
+# Keep the consolidated earlier manuscript title searchable.
+# ACM publisher metadata: https://api.crossref.org/works/10.1145/3832783.3834335
 abstract = "Microservice root cause analysis (RCA) requires correlating failures across heterogeneous telemetry within complex service dependency graphs. Existing methods often rely on a single telemetry modality; recent LLM-based approaches can suffer from unconstrained exploration and hallucination; and most systems stop at fault ranking without producing actionable incident response. We present GALA+, a graph-augmented LLM agentic framework centered on graph-guided investigation, which uses service dependencies to bound exploration and refine diagnosis through localized multi-modal evidence. For initial hypothesis generation, GALA+ combines complementary telemetry signals with STRIX, a novel trace- and graph-structure-aware scoring module. GALA+ then produces ranked diagnoses, incident summaries, and stratified action recommendations. We further introduce SURE-Score, a human-guided evaluation framework co-developed with industry SRE experts for assessing RCA-specific output quality beyond conventional text similarity metrics. On two microservice benchmarks, GALA+ consistently achieves the strongest overall results, surpassing the best LLM-based baseline by more than 25 percentage points in AC@1, while also receiving the highest ratings from both SURE-Score and independent human SRE evaluation."
+
+# Attribution: https://arxiv.org/abs/2608.08968; repository README describes release scope.
+research_outcomes = [
+  { title = "GALA evaluation artifacts", url = "https://github.com/EvanTian233/GALA-for-RCA", summary = "Annotated incident reports, example outputs, a supplementary technical report, and third-party baseline frameworks. The core GALA implementation is not publicly released." },
+]
+preprint_url = "https://arxiv.org/abs/2608.08968"
 +++
+
+The preprint linked above describes the GALA+ framework.
+
+This record consolidates the GALA work and its updated GALA+ framework.
+The [earlier preprint](https://arxiv.org/abs/2508.12472) circulated under the title
+*GALA: Can Graph-Augmented Large Language Model Agentic Workflows Elevate Root Cause Analysis?*.
diff --git a/content/publications/gpos-geospatially-aware-proof-of-stake.md b/content/publications/gpos-geospatially-aware-proof-of-stake.md
index e194df1..07657ed 100644
--- a/content/publications/gpos-geospatially-aware-proof-of-stake.md
+++ b/content/publications/gpos-geospatially-aware-proof-of-stake.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3774629"
 abstract = "Geospatial decentralization is essential for blockchains, ensuring regulatory resilience, robustness, and fairness. We empirically analyze five major Proof of Stake (PoS) blockchains—Aptos, Avalanche, Ethereum, Solana, and Sui—revealing that a few geographic regions dominate consensus voting power, resulting in limited geospatial decentralization. To address this, we propose Geospatially-aware Proof of Stake (GPoS), which integrates geospatial diversity with stake-based voting power. Experimental evaluation demonstrates an average 45% improvement in geospatial decentralization, as measured by the Gini coefficient of Eigenvector centrality, while incurring minimal performance overhead in BFT protocols, including HotStuff and CometBFT. These results demonstrate that GPoS can improve geospatial decentralization while, in our experiments, incurring minimal overhead to consensus performance."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2511.02034"
 +++
diff --git a/content/publications/how-can-we-train-deep-learning-models-across-clouds-and-continents-an-experimental-study.md b/content/publications/how-can-we-train-deep-learning-models-across-clouds-and-continents-an-experimental-study.md
index 611121c..0bbec21 100644
--- a/content/publications/how-can-we-train-deep-learning-models-across-clouds-and-continents-an-experimental-study.md
+++ b/content/publications/how-can-we-train-deep-learning-models-across-clouds-and-continents-an-experimental-study.md
@@ -4,7 +4,9 @@ year = 2024
 authors = ["Alexander Erben", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the VLDB Endowment"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.14778/3648160.3648165"
 abstract = "This paper aims to answer the question: Can deep learning models be cost-efficiently trained on a global market of spot VMs spanning different data centers and cloud providers? To provide guidance, we extensively evaluate the cost and throughput implications of training in different zones, continents, and clouds for representative CV, NLP and ASR models. To expand the current training options further, we compare the scalability potential for hybrid-cloud scenarios by adding cloud resources to on-premise hardware to improve training throughput. Finally, we show how leveraging spot instance pricing enables a new cost-efficient way to train models with multiple cheap VMs, trumping both more centralized and powerful hardware and even on-demand cloud offerings at competitive prices."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2306.03163"
 +++
diff --git a/content/publications/how-does-stake-distribution-influence-consensus.md b/content/publications/how-does-stake-distribution-influence-consensus.md
index 6b41c4f..9916307 100644
--- a/content/publications/how-does-stake-distribution-influence-consensus.md
+++ b/content/publications/how-does-stake-distribution-influence-consensus.md
@@ -8,4 +8,6 @@ research = ["data-management"]
 tags = ["blockchain", "consensus", "decentralization"]
 external_url = "https://doi.org/10.1109/icbc59979.2024.10634400"
 abstract = "In the PoS blockchain landscape, the challenge of achieving full decentralization is often hindered by a disproportionate concentration of staked tokens among a few validators. This study analyses this challenge by first formalizing decentralization metrics for weighted consensus mechanisms. An empirical analysis across ten permissionless blockchains uncovers significant weight concentration among validators, underscoring the need for an equitable approach. To counter this, we introduce the Square Root Stake Weight (SRSW) model, which effectively recalibrates staking weight distribution. Our examination of the SRSW model demonstrates notable improvements in the decentralization metrics: the Gini index improves by $37.16 \\%$ on average, while Nakamoto coefficients for liveness and safety see mean enhancements of $101.04 \\%$ and $80.09 \\%$, respectively. This research is a pivotal step toward a more fair and equitable distribution of staking weight, advancing the decentralization in blockchain consensus mechanisms."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2312.13938"
 +++
diff --git a/content/publications/how-to-optimize-my-blockchain-a-multi-level-recommendation-approach.md b/content/publications/how-to-optimize-my-blockchain-a-multi-level-recommendation-approach.md
index 0bdaeac..d427129 100644
--- a/content/publications/how-to-optimize-my-blockchain-a-multi-level-recommendation-approach.md
+++ b/content/publications/how-to-optimize-my-blockchain-a-multi-level-recommendation-approach.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3588704"
 abstract = "Aside from the conception of new blockchain architectures, existing blockchain optimizations in the literature primarily focus on system or data-oriented optimizations within prevailing blockchains. However, since blockchains handle multiple aspects ranging from organizational governance to smart contract design, a holistic approach that encompasses all the different layers of a given blockchain system is required to ensure that all optimization opportunities are taken into consideration. In this vein, we define a multi-level optimization recommendation approach that identifies optimization opportunities within a blockchain at the system, data, and user level. Multiple metrics and attributes are derived from a blockchain log and nine optimization recommendations are formalized. We implement an automated optimization recommendation tool, BlockOptR, based on these concepts. The system is extensively evaluated with a wide range of workloads covering multiple real-world scenarios. After implementing the recommended optimizations, we observe an average of 20% improvement in the success rate of transactions and an average of 40% improvement in latency."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2301.04719"
 +++
diff --git a/content/publications/hybrid-edge-partitioner-partitioning-large-power-law-graphs-under-memory-constraints.md b/content/publications/hybrid-edge-partitioner-partitioning-large-power-law-graphs-under-memory-constraints.md
index 34fff5a..d6ff21f 100644
--- a/content/publications/hybrid-edge-partitioner-partitioning-large-power-law-graphs-under-memory-constraints.md
+++ b/content/publications/hybrid-edge-partitioner-partitioning-large-power-law-graphs-under-memory-constraints.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3448016.3457300"
 abstract = "Distributed systems that manage and process graph-structured data internally solve a graph partitioning problem to minimize their communication overhead and query run-time. Besides computational complexity---optimal graph partitioning is NP-hard---another important consideration is the memory overhead. Real-world graphs often have an immense size, such that loading the complete graph into memory for partitioning is not economical or feasible. Currently, the common approach to reduce memory overhead is to rely on streaming partitioning algorithms. While the latest streaming algorithms lead to reasonable partitioning quality on some graphs, they are still not completely competitive to in-memory partitioners. In this paper, we propose a new system, Hybrid Edge Partitioner (HEP), that can partition graphs that fit partly into memory while yielding a high partitioning quality. HEP can flexibly adapt its memory overhead by separating the edge set of the graph into two sub-sets. One sub-set is partitioned by NE++, a novel, efficient in-memory algorithm, while the other sub-set is partitioned by a streaming approach. Our evaluations on large real-world graphs show that in many cases, HEP outperforms both in-memory partitioning and streaming partitioning at the same time. Hence, HEP is an attractive alternative to existing solutions that cannot fine-tune their memory overheads. Finally, we show that using HEP, we achieve a significant speedup of distributed graph processing jobs on Spark/GraphX compared to state-of-the-art partitioning algorithms."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2103.12594"
 +++
diff --git a/content/publications/i-know-what-you-mean-semantic-issues-in-internet-scale-publish-subscribe-systems.md b/content/publications/i-know-what-you-mean-semantic-issues-in-internet-scale-publish-subscribe-systems.md
index c308cb9..9c32ec6 100644
--- a/content/publications/i-know-what-you-mean-semantic-issues-in-internet-scale-publish-subscribe-systems.md
+++ b/content/publications/i-know-what-you-mean-semantic-issues-in-internet-scale-publish-subscribe-systems.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://dblp.org/rec/conf/semweb/BurceaPJ03"
 abstract = "In recent years, the amount of information on the Internet has increased exponentially developing great interest in selective information dissemination systems. The publish/subscribe paradigm is particularly suited for designing systems for routing information and requests according to their content throughout wide-area network of brokers. Current publish/subscribe systems use limited syntax-based content routing but since publishers and subscribers are anonymous and decoupled in time, space and location, often over wide-area network boundary, they do not necessarily speak the same language. Consequently, adding semantics to current publish/subscribe systems is important. In this paper we identify and examine the issues in developing semantic-based content routing for publish/subscribe broker networks."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/cs/0311047"
 +++
diff --git a/content/publications/i13dr-a-real-time-demand-response-infrastructure-for-integrating-renewable-energy-resources.md b/content/publications/i13dr-a-real-time-demand-response-infrastructure-for-integrating-renewable-energy-resources.md
index de59529..7a3e671 100644
--- a/content/publications/i13dr-a-real-time-demand-response-infrastructure-for-integrating-renewable-energy-resources.md
+++ b/content/publications/i13dr-a-real-time-demand-response-infrastructure-for-integrating-renewable-energy-resources.md
@@ -5,6 +5,6 @@ authors = ["Pezhman Nasirifard", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = []
-external_url = "https://arxiv.org/abs/2210.07789"
+preprint_url = "https://arxiv.org/abs/2210.07789"
 abstract = "With the ongoing integration of Renewable Energy Sources (RES), the complexity of power grids is increasing. Due to the fluctuating nature of RES, ensuring the reliability of power grids can be challenging. One possible approach for addressing these challenges is Demand Response (DR) which is described as matching the demand for electrical energy according to the changes and the availability of supply. However, implementing a DR system to monitor and control a broad set of electrical appliances in real-time introduces several new complications, including ensuring the reliability and financial feasibility of the system. In this work, we address these issues by designing and implementing a distributed real-time DR infrastructure for laptops, which estimates and controls the power consumption of a network of connected laptops in response to the fast, irregular changes of RES. Furthermore, since our approach is entirely software-based, we dramatically reduce the initial costs of the demand side participants. The result of our field experiments confirms that our system successfully schedules and executes rapid and effective DR events. However, the accuracy of the estimated power consumption of all participating laptops is relatively low, directly caused by our software-based approach."
 +++
diff --git a/content/publications/inference-of-distribution-grids-based-on-crowdsourced-grid-data-and-drone-imagery.md b/content/publications/inference-of-distribution-grids-based-on-crowdsourced-grid-data-and-drone-imagery.md
index e0ad791..2b68bbe 100644
--- a/content/publications/inference-of-distribution-grids-based-on-crowdsourced-grid-data-and-drone-imagery.md
+++ b/content/publications/inference-of-distribution-grids-based-on-crowdsourced-grid-data-and-drone-imagery.md
@@ -4,7 +4,7 @@ year = 2022
 authors = ["Hans-Arno Jacobsen", "Pezhman Nasirifard", "José Rivera", "Prerona Ray Baruah"]
 venue = "IEEE Transactions on Sustainable Computing"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1109/tsusc.2019.2961763"
 abstract = "Distribution System Operators (DSOs) face several challenges in managing comprehensive and up-to-date models of distribution grids. To address these problems, we propose a crowdsourcing framework for collecting grid devices. We also provide an inference approach for generating topological models of the distribution grids. Since distribution cables are often underground, we use spatial data analytics on the collected data in combination with other open data sources to infer the topology of the distribution grid. Additionally, to increase the quality of crowdsourced data, we propose a cost-effective approach for collecting and detecting grid elements in urban areas using commercial drones with an RGB camera. To evaluate our approach, we organized a crowdsourcing campaign to map and infer a district in Munich, Germany. The results are compared with the ground truth of the distribution system operator. Our results report a precision of up to 82 percent and a recall of up to 65 percent for the correctly crowdsourced grid devices. We also observe that the inferred models achieve a power length accuracy of 88 percent compared to the ground truth. We evaluated the detection of solar panels from aerial imagery by conducting field experiments, showing precision and recall levels of 68 and 69 percent, respectively."
 +++
diff --git a/content/publications/ksurf-attention-kalman-filter-and-principal-component-analysis-for-prediction-under-highly-variable-cloud-workloads.md b/content/publications/ksurf-attention-kalman-filter-and-principal-component-analysis-for-prediction-under-highly-variable-cloud-workloads.md
index 36555b3..9401b99 100644
--- a/content/publications/ksurf-attention-kalman-filter-and-principal-component-analysis-for-prediction-under-highly-variable-cloud-workloads.md
+++ b/content/publications/ksurf-attention-kalman-filter-and-principal-component-analysis-for-prediction-under-highly-variable-cloud-workloads.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = []
 external_url = "https://doi.org/10.1109/eecsi63442.2024.10776180"
 abstract = "Resource estimation and workload forecasting are critical in cloud data centers. Complexity in the cloud provider environment due to varying numbers of virtual machines introduces high variability in workloads and resource usage, making estimations problematic using state-of-the-art models that fail to deal with nonlinear characteristics. High measurement noise and variance affect the estimation of resource metrics of cloud systems across packet networks influenced by external dynamics. An ideal solution to these problems is the Kalman filter, a variance-minimizing estimator, ideal for highly variable data with Gaussian state space noise such as internet workloads. This work provides a new solution by making these contributions: i) it introduces a novel Kalman filter estimator using principal component analysis and an attention mechanism, ii) it evaluates the scheme on a Google Cloud benchmark comparing it to the state-of-the-art Bi-directional Grid Long Short- Term Memory network model on prediction tasks and iii) demonstrates real-time performance through a control task using a cloud-based messaging system with predictive auto-scaling. The new scheme improves prediction accuracy by 37% over state-of-the-art Kalman filters in prediction tasks, reduces the time series prediction error of the neural network model by over 40%, and improves Apache Kafka workload-based scaling stability by 58%."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2406.18801"
 +++
diff --git a/content/publications/ksurf-drone-attention-kalman-filter-for-contextual-bandit-optimization-in-cloud-resource-allocation.md b/content/publications/ksurf-drone-attention-kalman-filter-for-contextual-bandit-optimization-in-cloud-resource-allocation.md
index 680f0c5..86849ce 100644
--- a/content/publications/ksurf-drone-attention-kalman-filter-for-contextual-bandit-optimization-in-cloud-resource-allocation.md
+++ b/content/publications/ksurf-drone-attention-kalman-filter-for-contextual-bandit-optimization-in-cloud-resource-allocation.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = []
 external_url = "https://doi.org/10.1109/tcc.2026.3653558"
 abstract = "Resource orchestration and configuration parameter search are key concerns for container-based infrastructure in cloud data centers. Large configuration search space and cloud uncertainties are often mitigated using contextual bandit techniques for resource orchestration including the state-of-the-art Drone orchestrator. Complexity in the cloud provider environment due to varying numbers of virtual machines introduces variability in workloads and resource metrics, making orchestration decisions less accurate due to increased nonlinearity and noise. Ksurf, a state-of-the-art variance-minimizing estimator method ideal for highly variable cloud data, enables optimal resource estimation under conditions of high cloud variability. This work evaluates the performance of Ksurf on estimation-based resource orchestration tasks involving highly variable workloads when employed as a contextual multi-armed bandit objective function model for cloud scenarios using Drone. Ksurf enables significantly lower latency variance of over $40\\%$ at p95 and p99, demonstrates significant reduction in CPU and master node memory usage on Kubernetes, resulting in a $7\\%$ cost savings in average worker pod count on $VarBench$ Kubernetes benchmark."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2511.09766"
 +++
diff --git a/content/publications/lifting-the-fog-of-uncertainties.md b/content/publications/lifting-the-fog-of-uncertainties.md
index 5a65bd4..ce649a1 100644
--- a/content/publications/lifting-the-fog-of-uncertainties.md
+++ b/content/publications/lifting-the-fog-of-uncertainties.md
@@ -8,4 +8,6 @@ research = ["data-management"]
 tags = ["cloud-systems", "resource-orchestration", "containers"]
 external_url = "https://doi.org/10.1145/3620678.3624646"
 abstract = "The advances in virtualization technologies have sparked a growing transition from virtual machine (VM)-based to container-based infrastructure for cloud computing. From the resource orchestration perspective, containers' lightweight and highly configurable nature not only enables opportunities for more optimized strategies, but also poses greater challenges due to additional uncertainties and a larger configuration parameter search space. Towards this end, we propose Drone, a resource orchestration framework that adaptively configures resource parameters to improve application performance and reduce operational cost in the presence of cloud uncertainties. Built on Contextual Bandit techniques, Drone is able to achieve a balance between performance and resource cost on public clouds, and optimize performance on private clouds where a hard resource constraint is present. We show that our algorithms can achieve sub-linear growth in cumulative regret, a theoretically sound convergence guarantee, and our extensive experiments show that Drone achieves an up to 45% performance improvement and a 20% resource footprint reduction across batch processing jobs and microservice workloads."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2309.16962"
 +++
diff --git a/content/publications/mar-fl-a-communication-efficient-peer-to-peer-federated-learning-system.md b/content/publications/mar-fl-a-communication-efficient-peer-to-peer-federated-learning-system.md
index e2f8cd9..2c9722b 100644
--- a/content/publications/mar-fl-a-communication-efficient-peer-to-peer-federated-learning-system.md
+++ b/content/publications/mar-fl-a-communication-efficient-peer-to-peer-federated-learning-system.md
@@ -4,7 +4,9 @@ year = 2025
 authors = ["Felix Mulitze", "Herbert Woisetschläger", "Hans-Arno Jacobsen"]
 venue = "AI4NextG Workshop at NeurIPS 2025"
 publication_type = "Workshop Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://neurips.cc/virtual/2025/123210"
 abstract = "The convergence of next-generation wireless systems and distributed Machine Learning (ML) demands Federated Learning (FL) methods that remain efficient and robust with wireless connected peers and under network churn. Peer-to-peer (P2P) FL removes the bottleneck of a central coordinator, but existing approaches suffer from excessive communication complexity, limiting their scalability in practice. We introduce MAR-FL, a novel P2P FL system that leverages iterative group-based aggregation to substantially reduce communication overhead while retaining resilience to churn. MAR-FL achieves communication costs that scale as O(N log N), contrasting with the O(N^2) complexity of previously existing baselines, and thereby maintains effectiveness especially as the number of peers in an aggregation round grows. The system is robust towards unreliable FL clients and can integrate private computing."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2512.05234"
 +++
diff --git a/content/publications/mess-dynamically-learned-inference-time-llm-routing-in-model-zoos-with-service-level-guarantees.md b/content/publications/mess-dynamically-learned-inference-time-llm-routing-in-model-zoos-with-service-level-guarantees.md
index c26ff6f..55152a4 100644
--- a/content/publications/mess-dynamically-learned-inference-time-llm-routing-in-model-zoos-with-service-level-guarantees.md
+++ b/content/publications/mess-dynamically-learned-inference-time-llm-routing-in-model-zoos-with-service-level-guarantees.md
@@ -4,7 +4,9 @@ year = 2025
 authors = ["Herbert Woisetschläger", "Ryan Zhang", "Shiqiang Wang", "Hans-Arno Jacobsen"]
 venue = "Advances in Neural Information Processing Systems 38"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.52202/085713-1804"
 abstract = "Open-weight large language model (LLM) zoos provide access to numerous high-quality models, but selecting the appropriate model for specific tasks remains challenging and requires technical expertise. Most users simply want factually correct, safe, and satisfying responses without concerning themselves with model technicalities, while inference service providers prioritize minimizing operating costs. These competing interests are typically mediated through service level agreements (SLAs) that guarantee minimum service quality. We introduce MESS+, a stochastic optimization algorithm for cost-optimal LLM request routing while providing rigorous SLA compliance guarantees. MESS+ learns request satisfaction probabilities of LLMs in real-time as users interact with the system, based on which model selection decisions are made by solving a per-request optimization problem. Our algorithm includes a novel combination of virtual queues and request satisfaction prediction, along with a theoretical analysis of cost optimality and constraint satisfaction. Across a wide range of state-of-the-art LLM benchmarks, MESS+ achieves an average of $2\\times$ cost savings compared to existing LLM routing techniques."
+# Version 1 used "Dynamically Learned Test-Time Model Routing in Language Model Zoos with Service Level Guarantees"; the current preprint title matches this paper.
+preprint_url = "https://arxiv.org/abs/2505.19947"
 +++
diff --git a/content/publications/mess-energy-optimal-inferencing-in-language-model-zoos-with-service-level-guarantees.md b/content/publications/mess-energy-optimal-inferencing-in-language-model-zoos-with-service-level-guarantees.md
index 4728115..88cc7df 100644
--- a/content/publications/mess-energy-optimal-inferencing-in-language-model-zoos-with-service-level-guarantees.md
+++ b/content/publications/mess-energy-optimal-inferencing-in-language-model-zoos-with-service-level-guarantees.md
@@ -4,7 +4,9 @@ year = 2024
 authors = ["Ryan Zhang", "Herbert Woisetschläger", "Shiqiang Wang", "Hans-Arno Jacobsen"]
 venue = "Workshop on Adaptive Foundation Models at NeurIPS 2024"
 publication_type = "Workshop Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://openreview.net/forum?id=OoReeQpwmW"
 abstract = "Open-weight large language model (LLM) zoos allow users to quickly integrate state-of-the-art models into systems. Despite increasing availability, selecting the most appropriate model for a given task still largely relies on public benchmark leaderboards and educated guesses. This can be unsatisfactory for both inference service providers and end users, where the providers usually prioritize cost efficiency, while the end users usually prioritize model output quality for their inference requests. In commercial settings, these two priorities are often brought together in Service Level Agreements (SLA). We present MESS+, an online stochastic optimization algorithm for energy-optimal model selection from a model zoo, which works on a per-inference-request basis. For a given SLA that requires high accuracy, we are up to 2.5x more energy efficient with MESS+ than with randomly selecting an LLM from the zoo while maintaining SLA quality constraints."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2411.00889"
 +++
diff --git a/content/publications/on-the-effects-of-distributed-electric-vehicle-network-utility-maximization-in-low-voltage-feeders.md b/content/publications/on-the-effects-of-distributed-electric-vehicle-network-utility-maximization-in-low-voltage-feeders.md
index 42c59f7..5393fd9 100644
--- a/content/publications/on-the-effects-of-distributed-electric-vehicle-network-utility-maximization-in-low-voltage-feeders.md
+++ b/content/publications/on-the-effects-of-distributed-electric-vehicle-network-utility-maximization-in-low-voltage-feeders.md
@@ -5,6 +5,6 @@ authors = ["Jose Rivera", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = []
-external_url = "https://arxiv.org/abs/1706.10074"
+preprint_url = "https://arxiv.org/abs/1706.10074"
 abstract = "The fast charging of Electric Vehicles (EVs) in distribution networks requires real-time EV charging control to avoid the overloading of grid components. Recent studies have proposed congestion control protocols, which result from distributed optimization solutions of the Network Utility Maximization (NUM) problem. While the NUM formulation allows the definition of distributed computations with closed form solutions, its simple model does not account for many of the feeders operational constraints. This puts the resulting control algorithms effectiveness into question. In this paper, we investigate the impact of implementing such algorithms for congestion control in low voltage feeders. We review the latest NUM based algorithms for real-time EV charging control, and evaluate their behavior and impact on the comprehensive IEEE European Low Voltage Test Feeder. Our results show that the EV NUM problem can effectively capture the relevant operational constraints, as long as ampacity violations are the main bottleneck. Moreover, the results demonstrate an advantage of the primal NUM solution over the more conventional dual NUM solution in preventing a system overload."
 +++
diff --git a/content/publications/orca-observability-grounded-program-repair-for-microservice-incidents.md b/content/publications/orca-observability-grounded-program-repair-for-microservice-incidents.md
index 6ad8762..9f0f3ad 100644
--- a/content/publications/orca-observability-grounded-program-repair-for-microservice-incidents.md
+++ b/content/publications/orca-observability-grounded-program-repair-for-microservice-incidents.md
@@ -4,7 +4,7 @@ year = 2026
 authors = ["Yuanchen Gao", "Yifang Tian", "Yiran Li", "Charles Zhang", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["data-management"]
-external_url = "https://arxiv.org/abs/2608.17018"
+research = ["data-management", "rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2608.17018"
 abstract = "Microservice failures are often diagnosed from operational telemetry. However, automated program repair systems usually start from issue reports, localized code context, or failing tests. This mismatch leaves a gap between telemetry-based diagnosis and patch generation. We present ORCA, an observability-grounded APR pipeline for microservice incidents. ORCA first distills the differences in paired failure and reference telemetry into a fault signature, then uses the signature to identify candidate code and deployment-configuration locations. Repair graph agents and an Exploration agent generate unified-diff patch candidates from these locations. ORCA evaluates generated patches with a Telemetry-Grounded Patch Verifier that separates patch validity, syntactic and semantic correctness, test-oracle integrity, and telemetry replay. On a 575-case benchmark, ORCA outperforms all evaluated baselines in terms of cost-effectiveness. Results show that operational telemetry can be transformed from diagnostic evidence into actionable repair context: paired telemetry supports repair-oriented localization, while repair graph agents convert localized code and configuration evidence into constrained patch-generation context for the LLM. Telemetry-grounded verification then exposes repair outcomes that issue- or test-only evaluation would miss."
 +++
diff --git a/content/publications/orderlesschain-a-crdt-based-bft-coordination-free-blockchain-without-global-order-of-transactions.md b/content/publications/orderlesschain-a-crdt-based-bft-coordination-free-blockchain-without-global-order-of-transactions.md
index 12eba89..217dd4a 100644
--- a/content/publications/orderlesschain-a-crdt-based-bft-coordination-free-blockchain-without-global-order-of-transactions.md
+++ b/content/publications/orderlesschain-a-crdt-based-bft-coordination-free-blockchain-without-global-order-of-transactions.md
@@ -7,4 +7,9 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3590140.3629111"
 abstract = "Existing permissioned blockchains often rely on coordination-based consensus protocols to ensure the safe execution of applications in a Byzantine environment. Furthermore, the protocols serialize the transactions by ordering them in a global order. The serializ-ability preserves the correctness of the application's state stored on the blockchain. However, coordination-based protocols limit the throughput and scalability and induce high latency. In contrast, application-level correctness requirements exist that are not dependent on the order of transactions, known as invariant-confluence (I-confluence). The I-confluent applications can execute transactions in a coordination-free manner, benefiting from the improved scalability compared to the coordination-based approaches. The safety and liveness of I-confluent applications are studied in non-Byzantine environments, but the correct execution of such applications remains a challenge in Byzantine coordination-free environments. We introduce OrderlessChain, a novel permissioned blockchain based on a novel BFT coordination-free protocol for the safe and live execution of I-confluent applications in a Byzantine environment. We implemented a prototype of our system, and our evaluation results show that our coordination-free approach performs significantly better than coordination-based blockchains."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2210.01477"
 +++
+
+The preprint linked above uses the earlier title
+*OrderlessChain: Do Permissioned Blockchains Need Total Global Order of Transactions?*.
diff --git a/content/publications/out-of-core-edge-partitioning-at-linear-run-time.md b/content/publications/out-of-core-edge-partitioning-at-linear-run-time.md
index 4c9ee70..d0ce020 100644
--- a/content/publications/out-of-core-edge-partitioning-at-linear-run-time.md
+++ b/content/publications/out-of-core-edge-partitioning-at-linear-run-time.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1109/icde53745.2022.00242"
 abstract = "Graph edge partitioning is an important prepro-cessing step to optimize distributed computing jobs on graph-structured data. The edge set of a given graph is split into $k$ equally-sized partitions, such that the replication of vertices across partitions is minimized. Out-of-core edge partitioning algorithms are able to tackle the problem with low memory over-head. Existing out-of-core algorithms mainly work in a streaming manner and can be grouped into two types. While stateless streaming edge partitioning is fast and yields low partitioning quality, stateful streaming edge partitioning yields better quality, but is expensive, as it requires a scoring function to be evaluated for every edge on every partition, leading to a time complexity of O(|E| \\*k). In this paper, we propose 2PS-L, a novel out-of-core edge partitioning algorithm that builds upon the stateful streaming model, but achieves linear run-time i.e.,O(|E|)). 2PS-L consists of two phases. In the first phase, vertices are separated into clusters by a lightweight streaming clustering algorithm. In the second phase, the graph is re-streamed and vertex clustering from the first phase is exploited to reduce the search space of graph partitioning to only two target partitions for every edge. Our evaluations show that 2PS-L can achieve better partitioning quality than existing stateful streaming edge partitioners while having a much lower run-time. As a consequence, the total run-time of partitioning and subsequent distributed graph processing can be significantly reduced."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2203.12721"
 +++
diff --git a/content/publications/overcoming-data-scarcity-through-transfer-learning-in-co2-based-building-occupancy-detection.md b/content/publications/overcoming-data-scarcity-through-transfer-learning-in-co2-based-building-occupancy-detection.md
index fdbe28e..755868a 100644
--- a/content/publications/overcoming-data-scarcity-through-transfer-learning-in-co2-based-building-occupancy-detection.md
+++ b/content/publications/overcoming-data-scarcity-through-transfer-learning-in-co2-based-building-occupancy-detection.md
@@ -4,7 +4,7 @@ year = 2023
 authors = ["Manuel Weber", "Farzan Banihashemi", "Peter Mandl", "Hans-Arno Jacobsen", "Ruben Mayer"]
 venue = "Proceedings of the 10th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3600100.3623718"
 abstract = "Knowing indoor occupancy states is crucial for energy optimization in buildings. While neural networks can effectively be used to detect occupancy based on carbon dioxide measurements, their application is impeded by the need for sufficient labeled training data. In this study, we analyze the prediction performance of three different transfer learning (TL) methods leveraging target room data jointly with data from other rooms. The methods include (1) pretraining and fine-tuning, (2) layer freezing, and (3) domain-adversarial learning. Using data from five real-world rooms and one simulated room, including multiple room types, we provide the most extensive evaluation of TL in the field of occupancy prediction from environmental variables to date. This work’s contribution further includes the architecture and hyperparameters of a deep CNN-LSTM model for CO2-based occupancy detection. Our results indicate that TL effectively reduces the required amount of target room data. Moreover, while previous literature was focused on pretraining with related real-world data, we show that similar performance can be achieved by the more practical approach of leveraging simulated data."
 +++
diff --git a/content/publications/panjoin-a-partition-based-adaptive-stream-join.md b/content/publications/panjoin-a-partition-based-adaptive-stream-join.md
index d83ed0a..0b9c1de 100644
--- a/content/publications/panjoin-a-partition-based-adaptive-stream-join.md
+++ b/content/publications/panjoin-a-partition-based-adaptive-stream-join.md
@@ -5,6 +5,6 @@ authors = ["Fei Pan", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/1811.05065"
+preprint_url = "https://arxiv.org/abs/1811.05065"
 abstract = "In stream processing, stream join is one of the critical sources of performance bottlenecks. The sliding-window-based stream join provides a precise result but consumes considerable computational resources. The current solutions lack support for the join predicates on large windows. These algorithms and their hardware accelerators are either limited to equi-join or use a nested loop join to process all the requests. In this paper, we present a new algorithm called PanJoin which has high throughput on large windows and supports both equi-join and non-equi-join. PanJoin implements three new data structures to reduce computations during the probing phase of stream join. We also implement the most hardware-friendly data structure, called BI-Sort, on FPGA. Our evaluation shows that PanJoin outperforms several recently proposed stream join methods by more than 1000x, and it also adapts well to highly skewed data."
 +++
diff --git a/content/publications/parallel-index-based-stream-join-on-a-multicore-cpu.md b/content/publications/parallel-index-based-stream-join-on-a-multicore-cpu.md
index ddd25e3..f8ac460 100644
--- a/content/publications/parallel-index-based-stream-join-on-a-multicore-cpu.md
+++ b/content/publications/parallel-index-based-stream-join-on-a-multicore-cpu.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3318464.3380576"
 abstract = "Indexing sliding window content to enhance the performance of streaming queries can be greatly improved by utilizing the computational capabilities of a multicore processor. Conventional indexing data structures optimized for frequent search queries on a prestored dataset do not meet the demands of indexing highly dynamic data as in streaming environments. In this paper, we introduce an index data structure, called the partitioned in-memory merge tree, to address the challenges that arise when indexing highly dynamic data, which are common in streaming settings. Utilizing the specific pattern of streaming data and the distribution of queries, we propose a low-cost and effective concurrency control mechanism to meet the demands of high-rate update queries. To complement the index, we design an algorithm to realize a parallel index-based stream join that exploits the computational power of multicore processors. Our experiments using an octa-core processor show that our parallel stream join achieves up to 5.5 times higher throughput than a single-threaded approach."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1903.00452"
 +++
diff --git a/content/publications/parametrized-quantum-circuit-learning-for-quantum-chemical-applications.md b/content/publications/parametrized-quantum-circuit-learning-for-quantum-chemical-applications.md
index 75fddc8..2b27d52 100644
--- a/content/publications/parametrized-quantum-circuit-learning-for-quantum-chemical-applications.md
+++ b/content/publications/parametrized-quantum-circuit-learning-for-quantum-chemical-applications.md
@@ -4,7 +4,9 @@ year = 2026
 authors = ["Grier M. Jones", "Viki Kumar Prasad", "Ulrich Fekl", "Hans-Arno Jacobsen"]
 venue = "Journal of Chemical Information and Modeling"
 publication_type = "Journal Article"
-research = ["quantum-computing-systems", "distributed-machine-learning"]
+research = ["quantum-computing-systems", "rca-aiops"]
 external_url = "https://doi.org/10.1021/acs.jcim.6c00376"
 abstract = "In the field of quantum machine learning (QML), parametrized quantum circuits (PQCs)─constructed using a combination of fixed and tunable quantum gates─provide a promising hybrid framework for tackling complex machine learning problems. Despite numerous proposed applications, there remains limited exploration of data sets relevant to quantum chemistry. In this study, we investigate the potential benefits and limitations of PQCs on two chemically meaningful data sets: (1) the BSE49 data set, containing bond separation energies for 49 different classes of chemical bonds, and (2) a data set of water conformers, where coupled-cluster singles and doubles (CCSD) wave functions are predicted from lower-level electronic structure methods using the data-driven coupled-cluster (DDCC) approach. We construct a comprehensive set of 168 PQCs by combining 14 data encoding strategies with 12 variational ansätze, and evaluate their performance on circuits with 5 and 16 qubits. Our initial analysis examines the impact of circuit structure on model performance using state-vector simulations. We then explore how circuit depth and training set size influence model performance. Finally, we assess the performance of the best-performing PQCs on current quantum hardware, using both noisy simulations (\"fake\" backends) and real quantum devices. Our findings underscore the challenges of applying PQCs to chemically relevant problems that are straightforward for classical machine learning methods but remain nontrivial for quantum approaches."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2507.08183"
 +++
diff --git a/content/publications/parparaw-massively-parallel-parsing-of-delimiter-separated-raw-data.md b/content/publications/parparaw-massively-parallel-parsing-of-delimiter-separated-raw-data.md
index b68e463..0554dbe 100644
--- a/content/publications/parparaw-massively-parallel-parsing-of-delimiter-separated-raw-data.md
+++ b/content/publications/parparaw-massively-parallel-parsing-of-delimiter-separated-raw-data.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.14778/3377369.3377372"
 abstract = "Parsing is essential for a wide range of use cases, such as stream processing, bulk loading, and in-situ querying of raw data. Yet, the compute-intense step often constitutes a major bottleneck in the data ingestion pipeline, since parsing of inputs that require more involved parsing rules is challenging to parallelise. This work proposes a massively parallel algorithm for parsing delimiter-separated data formats on GPUs. Other than the state-of-the-art, the proposed approach does not require an initial sequential pass over the input to determine a thread's parsing context. That is, how a thread, beginning somewhere in the middle of the input, should interpret a certain symbol (e.g., whether to interpret a comma as a delimiter or as part of a larger string enclosed in double-quotes). Instead of tailoring the approach to a single format, we are able to perform a massively parallel finite state machine (FSM) simulation, which is more flexible and powerful, supporting more expressive parsing rules with general applicability. Achieving a parsing rate of as much as 14.2 GB/s, our experimental evaluation on a GPU with 3 584 cores shows that the presented approach is able to scale to thousands of cores and beyond. With an end-to-end streaming approach, we are able to exploit the full-duplex capabilities of the PCIe bus and hide latency from data transfers. Considering the end-to-end performance, the algorithm parses 4.8 GB in as little as 0.44 seconds, including data transfers."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1905.13415"
 +++
diff --git a/content/publications/partitioner-selection-with-ease.md b/content/publications/partitioner-selection-with-ease.md
index 26ed347..7d2b753 100644
--- a/content/publications/partitioner-selection-with-ease.md
+++ b/content/publications/partitioner-selection-with-ease.md
@@ -4,8 +4,10 @@ year = 2023
 authors = ["Nikolai Merkel", "Ruben Mayer", "Tawkir Ahmed Fakir", "Hans-Arno Jacobsen"]
 venue = "2023 IEEE 39th International Conference on Data Engineering (ICDE)"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning", "data-management"]
+research = ["rca-aiops", "data-management"]
 tags = ["graph-systems", "graph-processing", "auto-tuning"]
 external_url = "https://doi.org/10.1109/icde55515.2023.00185"
 abstract = "For distributed graph processing on massive graphs, a graph is partitioned into multiple equally-sized parts which are distributed among machines in a compute cluster. In the last decade, many partitioning algorithms have been developed which differ from each other with respect to the partitioning quality, the run-time of the partitioning and the type of graph for which they work best. The plethora of graph partitioning algorithms makes it a challenging task to select a partitioner for a given scenario. Different studies exist that provide qualitative insights into the characteristics of graph partitioning algorithms that support a selection. However, in order to enable automatic selection, a quantitative prediction of the partitioning quality, the partitioning run-time and the run-time of subsequent graph processing jobs is needed. In this paper, we propose a machine learning-based approach to provide such a quantitative prediction for different types of edge partitioning algorithms and graph processing workloads. We show that training based on generated graphs achieves high accuracy, which can be further improved when using real-world data. Based on the predictions, the automatic selection reduces the end-to-end run-time on average by 11.1% compared to a random selection, by 17.4% compared to selecting the partitioner that yields the lowest cut size, and by 29.1% compared to the worst strategy, respectively. Furthermore, in 35.7% of the cases, the best strategy was selected."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2304.04976"
 +++
diff --git a/content/publications/position-let-s-develop-data-probes-to-fundamentally-understand-how-data-affects-llm-performance.md b/content/publications/position-let-s-develop-data-probes-to-fundamentally-understand-how-data-affects-llm-performance.md
index d65cc36..797aae9 100644
--- a/content/publications/position-let-s-develop-data-probes-to-fundamentally-understand-how-data-affects-llm-performance.md
+++ b/content/publications/position-let-s-develop-data-probes-to-fundamentally-understand-how-data-affects-llm-performance.md
@@ -4,8 +4,10 @@ year = 2026
 authors = ["Shiqiang Wang", "Herbert Woisetschläger", "Hans-Arno Jacobsen", "Mingyue Ji"]
 venue = "International Conference on Machine Learning (ICML)"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://icml.cc/virtual/2026/poster/67154"
 abstract = "Data is fundamental to large language models (LLMs). However, understanding of what makes certain data useful for different stages of an LLM workflow, including training, tuning, alignment, in-context learning, etc., and why, remains an open question. Current approaches rely heavily on extensive experimentation with large public datasets to obtain empirical heuristics for data filtering and dataset construction. These approaches are compute intensive and lack a principled way of understanding the essence of how specific data characteristics drive LLM behavior. In this position paper, we advocate for the need of developing systematic methodologies for generating synthetic sequences from appropriately defined random processes, with the goal that these sequences can reveal useful characteristics when they are used in one or multiple stages of the LLM workflow. We refer to such sequences as data probes. By observing LLM behavior on data probes, researchers can systematically conduct studies on how data characteristics influence model performance, generalization, and robustness. The probing sequences exhibit statistical properties that can be viewed using theoretical concepts, such as typical sets, which are generalized to describe the behaviors of LLMs. This data-probe approach provides a pathway for uncovering foundational insights into the role of data in LLM training and inference, beyond empirical heuristics."
 abstract_license_url = "https://creativecommons.org/licenses/by/4.0/"
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2605.18801"
 +++
diff --git a/content/publications/prestigebft-revolutionizing-view-changes-in-bft-consensus-algorithms-with-reputation-mechanisms.md b/content/publications/prestigebft-revolutionizing-view-changes-in-bft-consensus-algorithms-with-reputation-mechanisms.md
index 47971c4..ab5f8df 100644
--- a/content/publications/prestigebft-revolutionizing-view-changes-in-bft-consensus-algorithms-with-reputation-mechanisms.md
+++ b/content/publications/prestigebft-revolutionizing-view-changes-in-bft-consensus-algorithms-with-reputation-mechanisms.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1109/icde60146.2024.00156"
 abstract = "Passive view-change protocols are widely employed in BFT algorithms; however, they present the risks of selecting unavailable or slow servers as leaders. To tackle these challenges, we propose PrestigeBFT, a novel BFT consensus algorithm that incorporates an active view-change protocol with reputation mechanisms. PrestigeBFT evaluates a server's reputation based on its past behavior and elects more reputable servers as leaders. Our reputation mechanism incentivizes protocol-abiding behavior while penalizing faulty servers by imposing computational work. PrestigeBFT significantly enhances system availability and efficiency by avoiding unavailable or slow servers being assigned as leaders. Under normal operation, PrestigeBFT achieves $5\\times$ higher throughput than the baseline that uses passive view-change protocols. In addition, PrestigeBFT's throughput remains unaffected under benign faults and witnesses only a 24% drop under a variety of Byzantine faults, whereas the baseline throughput drops by 62% and 69%, respectively. In the long run, while the baseline's availability struggles at 37%, PrestigeBFT progressively improves its availability to over 90%."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2307.08154"
 +++
diff --git a/content/publications/probabilistic-design-of-parametrized-quantum-circuits-through-local-gate-modifications.md b/content/publications/probabilistic-design-of-parametrized-quantum-circuits-through-local-gate-modifications.md
index 173acf1..6a8396a 100644
--- a/content/publications/probabilistic-design-of-parametrized-quantum-circuits-through-local-gate-modifications.md
+++ b/content/publications/probabilistic-design-of-parametrized-quantum-circuits-through-local-gate-modifications.md
@@ -5,6 +5,6 @@ authors = ["Grier M. Jones", "Aviraj Newatia", "Alexander Lao", "Aditya K. Rao",
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["quantum-computing-systems"]
-external_url = "https://arxiv.org/abs/2602.12465"
+preprint_url = "https://arxiv.org/abs/2602.12465"
 abstract = "Within quantum machine learning, parametrized quantum circuits provide flexible quantum models, but their performance is often highly task-dependent, making manual circuit design challenging. Alternatively, quantum architecture search algorithms have been proposed to automate the discovery of task-specific parametrized quantum circuits using systematic frameworks. In this work, we propose an evolution-inspired heuristic quantum architecture search algorithm, which we refer to as the local quantum architecture search. The goal of the local quantum architecture search algorithm is to optimize parametrized quantum circuit architectures through a local, probabilistic search over a fixed set of gate-level actions applied to existing circuits. We evaluate the local quantum architecture search algorithm on two synthetic function-fitting regression tasks and two quantum chemistry regression datasets, including the BSE49 dataset of bond separation energies for first- and second-row elements and a dataset of water conformers generated using the data-driven coupled-cluster approach. Using state-vector simulation, our results highlight the applicability of local quantum architecture search algorithm for identifying competitive circuit architectures with desirable performance metrics. Lastly, we analyze the properties of the discovered circuits and demonstrate the deployment of the best-performing model on state-of-the-art quantum hardware."
 +++
diff --git a/content/publications/prospects-of-appliance-level-load-monitoring-in-off-the-shelf-energy-monitors-a-technical-review.md b/content/publications/prospects-of-appliance-level-load-monitoring-in-off-the-shelf-energy-monitors-a-technical-review.md
index e8fb4d1..bc2181c 100644
--- a/content/publications/prospects-of-appliance-level-load-monitoring-in-off-the-shelf-energy-monitors-a-technical-review.md
+++ b/content/publications/prospects-of-appliance-level-load-monitoring-in-off-the-shelf-energy-monitors-a-technical-review.md
@@ -7,4 +7,9 @@ publication_type = "Journal Article"
 research = []
 external_url = "https://doi.org/10.3390/en11010189"
 abstract = "The smart grid initiative has encouraged utility companies worldwide to roll-out new and smarter versions of energy meters. Before an extensive roll-out, which is both labor-intensive and incurs high capital costs, consumers need to be incentivised to reap the long-term benefits of such smart meters. Off-the-shelf energy monitors (e-monitors) can provide consumers with an insight into such potential benefits. As e-monitors are owned by the consumer, the consumer has greater control over the data, which significantly reduces the privacy and data confidentiality concerns. Because only limited online technical information is available about e-monitors, we evaluate several existing e-monitors using an online technical survey directly from the vendors. Besides automated e-monitoring, the use of different off-the-shelf e-monitors can also help to demonstrate state-of-the-art techniques such as non-intrusive load monitoring (NILM), data analytics, and the predictive maintenance of appliances. Our survey indicates a trend towards the incorporation of such state-of-the-art capabilities, particularly the appliance-level e-monitoring and load disaggregation. We have also discussed some essential requirements to implement load disaggregation in the next generation e-monitors. In future, these intelligent e-monitoring techniques will encourage effective consumer participation in the demand-side management (DSM) programs."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1607.07780"
 +++
+
+The preprint linked above circulated under the earlier title
+*A Step towards Advanced Metering for the Smart Grid: A Survey of Energy Monitors*.
diff --git a/content/publications/pta-a-programmable-teaching-assistant-for-lab-courses.md b/content/publications/pta-a-programmable-teaching-assistant-for-lab-courses.md
index b9e64dd..a968e57 100644
--- a/content/publications/pta-a-programmable-teaching-assistant-for-lab-courses.md
+++ b/content/publications/pta-a-programmable-teaching-assistant-for-lab-courses.md
@@ -4,7 +4,7 @@ year = 2023
 authors = ["Jawad Tahir", "Raj Mandal", "Olha Stefanova", "Hans-Arno Jacobsen", "Christoph Doblander", "Ruben Mayer"]
 venue = "Proceedings of the 2nd International Workshop on Data Systems Education: Bridging education practice with education research"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 tags = ["education-systems", "lab-automation"]
 external_url = "https://doi.org/10.1145/3596673.3596975"
 abstract = "Lab courses play a crucial role in enabling students to gain a deeper understanding of theoretical concepts, but these courses require a significant effort from the course's organizational staff, such as instructors and teaching assistants. To address this challenge, we developed pTA, an acronym for programmable teaching assistant, which automates the functional evaluation of students' submissions for the Cloud Databases course taught at the Technical University of Munich (TUM). pTA reduces the staff workload and provides instant feedback to students, thereby enhancing their understanding of the project specifications. Additionally, pTA includes a live leaderboard that provides a gamification element that makes the course more interactive and engaging for students. It is deployed on a Kubernetes cluster that ensures scalability with evaluation requests. In this paper, we describe the course's learning milestones and provide an overview of pTA's architecture and features. The system's efficacy was evaluated at TUM and the University of Toronto, where it was deployed in two similar courses. Our findings show that pTA reduced the staff workload by at least 75%, lowered the operating cost, and increased course capacity in terms of the number of students. Furthermore, our study suggests that students exhibit more interest in courses that integrate interactive learning systems and gamification elements."
diff --git a/content/publications/q-dice-quantum-distributed-interconnect-compiler-and-emulator.md b/content/publications/q-dice-quantum-distributed-interconnect-compiler-and-emulator.md
index 9838d35..f7c5e29 100644
--- a/content/publications/q-dice-quantum-distributed-interconnect-compiler-and-emulator.md
+++ b/content/publications/q-dice-quantum-distributed-interconnect-compiler-and-emulator.md
@@ -6,6 +6,6 @@ venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["quantum-computing-systems"]
 tags = ["quantum-systems"]
-external_url = "https://arxiv.org/abs/2606.11340"
+preprint_url = "https://arxiv.org/abs/2606.11340"
 abstract = "As distributed quantum computing (DQC) offers a leading path towards scalable quantum computation, the ability to benchmark distributed algorithms under realistic conditions becomes critical for system co-design. However, without access to physical systems, researchers lack tools to evaluate distribution protocols. We introduce Q-DICE (Quantum Distributed Interconnect Compiler and Emulator), a hardware-aware emulation environment for benchmarking distributed quantum circuits on classical simulators and on NISQ-era monolithic hardware. This work provides three core contributions: (1) a programmatic scheme to construct distributed QPU backends, utilizing two novel techniques - QPU slicing and stitching - to facilitate distributed circuit mapping, (2) a methodology for modeling nonlocal link noise using physically motivated Kraus operators and stochastic error channels, and (3) a boundary-aware circuit mapping algorithm enforcing distributed QPU topology constraints during transpilation. Together, these components constitute a distribution-aware compiler and noise-modeling engine that faithfully enforces the physical limitations of distributed quantum hardware within existing execution environments. We validate Q-DICE against a multitude of experimentally demonstrated quantum circuits, including a distributed Grover's search on optically linked trapped-ion hardware, achieving a worst-case fidelity deviation of 4% between simulated and experimental results. These findings demonstrate Q-DICE's capacity to accurately reproduce real distributed quantum system behavior across platforms, streamlining experimentation with distributed quantum algorithms and architectures."
 +++
diff --git a/content/publications/quantum-hypergraph-partitioning.md b/content/publications/quantum-hypergraph-partitioning.md
index ce6a803..b62748a 100644
--- a/content/publications/quantum-hypergraph-partitioning.md
+++ b/content/publications/quantum-hypergraph-partitioning.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["quantum-computing-systems", "data-management"]
 external_url = "https://doi.org/10.1145/3811628.3811836"
 abstract = "Hypergraph partitioning is a fundamental optimization problem with applications in data management and other domains involving higher-order relations. In this paper, we study balanced hypergraph partitioning from the perspective of quantum optimization. We formalize balanced k-way hypergraph partitioning with general hyperedge cut functions, and derive corresponding binary optimization formulations targeted at quantum optimization methods in both the two-way and multi-way settings. Our discussion highlights which cut functions admit Quadratic Unconstrained Binary Optimization (QUBO) encodings and which instead lead to higher-order binary objectives or rational forms. As a preliminary empirical validation, we focus on balanced two-way partitioning with the all-or-nothing cut on 3-uniform hypergraphs, where a direct QUBO is available, and evaluate simulated Quantum Approximate Optimization Algorithm (QAOA) and Simulated Annealing (SA) on small instances against exact solutions. The results show that the formulation is effective on small hypergraphs and that the balance-penalty weight plays a critical role in trading off cut quality and balance."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2605.02635"
 +++
diff --git a/content/publications/reaching-consensus-in-the-byzantine-empire.md b/content/publications/reaching-consensus-in-the-byzantine-empire.md
index dbab320..f0f8873 100644
--- a/content/publications/reaching-consensus-in-the-byzantine-empire.md
+++ b/content/publications/reaching-consensus-in-the-byzantine-empire.md
@@ -8,4 +8,6 @@ research = ["data-management"]
 tags = ["blockchain", "consensus", "systems-survey"]
 external_url = "https://doi.org/10.1145/3636553"
 abstract = "Byzantine fault-tolerant (BFT) consensus algorithms are at the core of providing safety and liveness guarantees for distributed systems that must operate in the presence of arbitrary failures. Recently, numerous new BFT algorithms have been proposed, not least due to the traction blockchain technologies have garnered in the search for consensus solutions that offer high throughput, low latency, and robust system designs. In this article, we conduct a systematic survey of selected and distinguished BFT algorithms that have received extensive attention in academia and industry alike. We perform a qualitative comparison among all algorithms we review considering message and time complexities. Furthermore, we provide a comprehensive, step-by-step description of each surveyed algorithm by decomposing them into constituent subprotocols with intuitive figures to illustrate the message-passing pattern. We also elaborate on the strengths and weaknesses of each algorithm compared to the other state-of-the-art approaches."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2204.03181"
 +++
diff --git a/content/publications/real-time-load-prediction-with-high-velocity-smart-home-data-stream.md b/content/publications/real-time-load-prediction-with-high-velocity-smart-home-data-stream.md
index 49b385d..6a581ba 100644
--- a/content/publications/real-time-load-prediction-with-high-velocity-smart-home-data-stream.md
+++ b/content/publications/real-time-load-prediction-with-high-velocity-smart-home-data-stream.md
@@ -5,6 +5,11 @@ authors = ["Christoph Doblander", "Martin Strohbach", "Holger Ziekow", "Hans-Arn
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/1708.04613"
+preprint_url = "https://arxiv.org/abs/1708.04613"
 abstract = "This paper addresses the use of smart-home sensor streams for continuous prediction of energy loads of individual households which participate as an agent in local markets. We introduces a new device level energy consumption dataset recorded over three years wich includes high resolution energy measurements from electrical devices collected within a pilot program. Using data from that pilot, we analyze the applicability of various machine learning mechanisms for continuous load prediction. Specifically, we address short-term load prediction that is required for load balancing in electrical micro-grids. We report on the prediction performance and the computational requirements of a broad range of prediction mechanisms. Furthermore we present an architecture and experimental evaluation when this prediction is applied in the stream."
 +++
+
+Related publication: [Poster Abstract: Real-time load prediction with high velocity
+smart home data stream](https://doi.org/10.1007/s00450-017-0364-5),
+*Computer Science - Research and Development* 33(1–2), pp. 233–234 (2018).
+This two-page poster abstract cites the full manuscript linked above.
diff --git a/content/publications/remop-remote-memory-aware-operator-optimization.md b/content/publications/remop-remote-memory-aware-operator-optimization.md
index 6447fc7..d5d9c63 100644
--- a/content/publications/remop-remote-memory-aware-operator-optimization.md
+++ b/content/publications/remop-remote-memory-aware-operator-optimization.md
@@ -5,6 +5,6 @@ authors = ["Shiquan Zhang", "Yunhao Mao", "Yuqiu Zhang", "Gengrui Zhang", "Jeyhu
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/2606.19576"
+preprint_url = "https://arxiv.org/abs/2606.19576"
 abstract = "Remote and disaggregated memory tiers expand the effective memory capacity of analytical database engines, but they also reshape the cost structure of out-of-memory query processing. When an operator spills beyond local DRAM, moving pages to remote memory incurs both data-transfer time and a fixed round-trip latency per transfer. Classical operator analyses and buffer-allocation heuristics primarily target disk spilling by minimizing total I/O volume. Under remote memory, these strategies can be suboptimal because they may trigger excessive transfer rounds. We present REMOP, a remote-memory-aware operator optimization framework that uses transfer-round-aware intra-operator memory policies to improve out-of-memory execution under tight memory budgets. REMOP introduces the number of transfer rounds into the latency cost model and derives operator-specific buffer-partitioning strategies, instantiating the approach for blocked nested-loop join, external merge sort, and external hash join in DuckDB. Our evaluation on a two-node compute-memory testbed shows that REMOP reduces transfer rounds by up to 97% and operator runtime by up to 48% on spill-heavy microbenchmarks, and lowers the average runtime of spilling TPC-H and TPC-DS queries by 22.7% and 26.4% end-to-end."
 +++
diff --git a/content/publications/representation-learning-for-appliance-recognition-a-comparison-to-classical-machine-learning.md b/content/publications/representation-learning-for-appliance-recognition-a-comparison-to-classical-machine-learning.md
index 0e5b519..51ad1bc 100644
--- a/content/publications/representation-learning-for-appliance-recognition-a-comparison-to-classical-machine-learning.md
+++ b/content/publications/representation-learning-for-appliance-recognition-a-comparison-to-classical-machine-learning.md
@@ -4,7 +4,7 @@ year = 2022
 authors = ["Matthias Kahl", "Daniel Jorde", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2209.03759"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2209.03759"
 abstract = "Non-intrusive load monitoring (NILM) aims at energy consumption and appliance state information retrieval from aggregated consumption measurements, with the help of signal processing and machine learning algorithms. Representation learning with deep neural networks is successfully applied to several related disciplines. The main advantage of representation learning lies in replacing an expert-driven, hand-crafted feature extraction with hierarchical learning from many representations in raw data format. In this paper, we show how the NILM processing-chain can be improved, reduced in complexity and alternatively designed with recent deep learning algorithms. On the basis of an event-based appliance recognition approach, we evaluate seven different classification models: a classical machine learning approach that is based on a hand-crafted feature extraction, three different deep neural network architectures for automated feature extraction on raw waveform data, as well as three baseline approaches for raw data processing. We evaluate all approaches on two large-scale energy consumption datasets with more than 50,000 events of 44 appliances. We show that with the use of deep learning, we are able to reach and surpass the performance of the state-of-the-art classical machine learning approach for appliance recognition with an F-Score of 0.75 and 0.86 compared to 0.69 and 0.87 of the classical approach."
 +++
diff --git a/content/publications/reward-mechanism-for-blockchains-using-evolutionary-game-theory.md b/content/publications/reward-mechanism-for-blockchains-using-evolutionary-game-theory.md
index d711305..8beb591 100644
--- a/content/publications/reward-mechanism-for-blockchains-using-evolutionary-game-theory.md
+++ b/content/publications/reward-mechanism-for-blockchains-using-evolutionary-game-theory.md
@@ -8,4 +8,6 @@ research = ["data-management"]
 tags = ["blockchain"]
 external_url = "https://doi.org/10.1109/brains52497.2021.9569791"
 abstract = "Blockchains have witnessed widespread adoption in the past decade in various fields. The growing demand makes their scalability and sustainability challenges more evident than ever. As a result, more and more blockchains have begun to adopt proof-of-stake (PoS) consensus protocols to address those challenges. One of the fundamental characteristics of any blockchain technology is its crypto-economics and incentives. Lately, each PoS blockchain has designed a unique reward mechanism, yet, many of them are prone to free-rider and nothing-at-stake problems. To better understand the ad-hoc design of reward mechanisms, in this paper, we develop a reward mechanism framework that could apply to many PoS blockchains. We formulate the block validation game wherein the rewards are distributed for validating the blocks correctly. Using evolutionary game theory, we analyze how the participants' behaviour could potentially evolve with the reward mechanism. Also, penalties are found to play a central role in maintaining the integrity of blockchains."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2104.05849"
 +++
diff --git a/content/publications/s-topss-semantic-toronto-publish-subscribe-system.md b/content/publications/s-topss-semantic-toronto-publish-subscribe-system.md
index b35a44d..d89d56f 100644
--- a/content/publications/s-topss-semantic-toronto-publish-subscribe-system.md
+++ b/content/publications/s-topss-semantic-toronto-publish-subscribe-system.md
@@ -7,4 +7,6 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1016/b978-012722442-8/50115-4"
 abstract = "The increase in the amount of data on the Internet has led to the development of a new generation of applications based on selective information dissemination where, data is distributed only to interested clients. Such applications require a new middleware architecture that can efficiently match user interests with available information. Middleware that can satisfy this requirement include event-based architectures such as publish-subscribe systems. In this demonstration paper we address the problem of semantic matching. We investigate how current publish/subscribe systems can be extended with semantic capabilities. Our main contribution is the development and validation (through demonstration) of a semantic pub/sub system prototype S-ToPSS (Semantic Toronto Publish/Subscribe System)."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/cs/0311041"
 +++
diff --git a/content/publications/scalable-deep-learning-on-distributed-infrastructures-challenges-techniques-and-tools.md b/content/publications/scalable-deep-learning-on-distributed-infrastructures-challenges-techniques-and-tools.md
index b594d93..7770b93 100644
--- a/content/publications/scalable-deep-learning-on-distributed-infrastructures-challenges-techniques-and-tools.md
+++ b/content/publications/scalable-deep-learning-on-distributed-infrastructures-challenges-techniques-and-tools.md
@@ -4,7 +4,9 @@ year = 2021
 authors = ["Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "ACM Computing Surveys"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3363554"
 abstract = "Deep Learning (DL) has had an immense success in the recent past, leading to state-of-the-art results in various domains, such as image recognition and natural language processing. One of the reasons for this success is the increasing size of DL models and the proliferation of vast amounts of training data being available. To keep on improving the performance of DL, increasing the scalability of DL systems is necessary. In this survey, we perform a broad and thorough investigation on challenges, techniques and tools for scalable DL on distributed infrastructures. This incorporates infrastructures for DL, methods for parallel DL training, multi-tenant resource scheduling, and the management of training and model data. Further, we analyze and compare 11 current open-source DL frameworks and tools and investigate which of the techniques are commonly implemented in practice. Finally, we highlight future research trends in DL systems that deserve further research."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1903.11314"
 +++
diff --git a/content/publications/scope-fl-a-strategy-proof-chain-based-optimal-pareto-efficient-federated-learning-system.md b/content/publications/scope-fl-a-strategy-proof-chain-based-optimal-pareto-efficient-federated-learning-system.md
index c5cd7a1..7e3548f 100644
--- a/content/publications/scope-fl-a-strategy-proof-chain-based-optimal-pareto-efficient-federated-learning-system.md
+++ b/content/publications/scope-fl-a-strategy-proof-chain-based-optimal-pareto-efficient-federated-learning-system.md
@@ -4,7 +4,7 @@ year = 2026
 authors = ["Seyed Salar Ghazi", "Kaiwen Zhang", "Mehdi feizi", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2606.18384"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2606.18384"
 abstract = "Hierarchical Federated Learning (HFL) enables scalable collaborative model training across distributed devices while preserving data privacy. However, existing HFL client selection mechanisms suffer from a fundamental strategic inefficiency. By prioritizing stability over Pareto efficiency (PE), they produce suboptimal resource allocations, and without strategy proofness (SP), participants are incentivized to misrepresent their true preferences, both failures degrading system overall welfare in the Pareto sense in practice. To address it, we propose SCOPE-FL (Strategy-proof Chain-based Optimal pareto efficient Federated Learning), a synchronous HFL framework that formulates client selection as a two-sided school choice problem solved through the Top Trading Cycle (TTC) algorithm that simultaneously guarantees PE and SP. For reward distribution, SCOPE-FL employs a scalable Shapley value approximation based on One-Round Reconstruction (OR), ensuring compensation proportional to each client's contribution. The entire mechanism executes via blockchain smart contracts, providing the tamper-proof environment required for the SP guarantees to hold in practice. A comprehensive evaluation on MNIST, Fashion-MNIST, and CIFAR-10 demonstrates that SCOPE-FL outperforms state-of-the-art approaches, including DA, IAS, and other methods across model accuracy, convergence rate, and reward efficiency, while achieving communication latency comparable to DA and blockchain overhead significantly lower than DA at scale."
 +++
diff --git a/content/publications/sdn-like-the-next-generation-of-pub-sub.md b/content/publications/sdn-like-the-next-generation-of-pub-sub.md
index dbf9886..52468b7 100644
--- a/content/publications/sdn-like-the-next-generation-of-pub-sub.md
+++ b/content/publications/sdn-like-the-next-generation-of-pub-sub.md
@@ -5,6 +5,6 @@ authors = ["Kaiwen Zhang", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/1308.0056"
+preprint_url = "https://arxiv.org/abs/1308.0056"
 abstract = "Software-Defined Networking (SDN) has raised the boundaries of cloud computing by offering unparalleled levels of control and flexibility to system administrators over their virtualized environments. To properly embrace this new era of SDN-driven network architectures, the research community must not only consider the impact of SDN over the protocol stack, but also on its overlying networked applications. In this big ideas paper, we study the impact of SDN on the design of future message-oriented middleware, specifically pub/sub systems. We argue that key concepts put forth by SDN can be applied in a meaningful fashion to the next generation of pub/sub systems. First, pub/sub can adopt a logically centralized controller model for maintenance, monitoring, and control of the overlay network. We establish a parallel with existing work on centralized pub/sub routing and discuss how the logically centralized controller model can be implemented in a distributed manner. Second, we investigate the separation of the control and data plane, which is integral to SDN, which can be adopted to raise the level of decoupling in pub/sub. We introduce a new model of pub/sub which separates the traditional publisher and subscriber roles into flow regulators and producer/consumers of data. We then present use cases that benefit from this approach and study the impact of decoupling for performance."
 +++
diff --git a/content/publications/should-my-blockchain-learn-to-drive-a-study-of-hyperledger-fabric.md b/content/publications/should-my-blockchain-learn-to-drive-a-study-of-hyperledger-fabric.md
index 4a575d5..8300d75 100644
--- a/content/publications/should-my-blockchain-learn-to-drive-a-study-of-hyperledger-fabric.md
+++ b/content/publications/should-my-blockchain-learn-to-drive-a-study-of-hyperledger-fabric.md
@@ -5,6 +5,6 @@ authors = ["Jeeta Ann Chacko", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
-external_url = "https://arxiv.org/abs/2406.06318"
+preprint_url = "https://arxiv.org/abs/2406.06318"
 abstract = "Similar to other transaction processing frameworks, blockchain systems need to be dynamically reconfigured to adapt to varying workloads and changes in network conditions. However, achieving optimal reconfiguration is particularly challenging due to the complexity of the blockchain stack, which has diverse configurable parameters. This paper explores the concept of self-driving blockchains, which have the potential to predict workload changes and reconfigure themselves for optimal performance without human intervention. We compare and contrast our discussions with existing research on databases and highlight aspects unique to blockchains. We identify specific parameters and components in Hyperledger Fabric, a popular permissioned blockchain system, that are suitable for autonomous adaptation and offer potential solutions for the challenges involved. Further, we implement three demonstrative locally autonomous systems, each targeting a different layer of the blockchain stack, and conduct experiments to understand the feasibility of our findings. Our experiments indicate up to 11% improvement in success throughput and a 30% decrease in latency, making this a significant step towards implementing a fully autonomous blockchain system in the future."
 +++
diff --git a/content/publications/size-does-not-matter-sparsification-and-graph-neural-network-sampling-for-large-scale-graphs.md b/content/publications/size-does-not-matter-sparsification-and-graph-neural-network-sampling-for-large-scale-graphs.md
index e375f15..063db60 100644
--- a/content/publications/size-does-not-matter-sparsification-and-graph-neural-network-sampling-for-large-scale-graphs.md
+++ b/content/publications/size-does-not-matter-sparsification-and-graph-neural-network-sampling-for-large-scale-graphs.md
@@ -4,7 +4,7 @@ year = 2024
 authors = ["Jana Vatter", "Maurice L. Rochau", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "VLDB Workshops"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://vldb.org/workshops/2024/proceedings/LSGDA/LSGDA24.06.pdf"
 abstract = "With the ever-growing size of real-world graphs, optimizing Graph Neural Network (GNN) training has become essential. Two key methods for this are graph sparsification and GNN sampling, both aim at reducing graph size while preserving valuable information. The question arises what kind of information should be preserved and what a reasonable graph size is. We propose combining random graph sparsification with GNN sampling, showing that this approach can significantly reduce training time while maintaining accuracy. Our experiments demonstrate that sparsification to around 40% of the original graph and sampling with a fanout parameter of 4 yields the best results in terms of training time and accuracy. Beyond training time, also inference time can be decreased up to 75% which enables scalability for time-critical applications such as fraud detection. Finally, we identify open challenges and new research directions, including sampling-aware graph reduction methods, mining new graph datasets, and the prevention of bias."
 abstract_license_url = "https://creativecommons.org/licenses/by-nc-nd/4.0/"
diff --git a/content/publications/solving-big-data-challenges-for-enterprise-application-performance-management.md b/content/publications/solving-big-data-challenges-for-enterprise-application-performance-management.md
index c5f6ab9..44cfba6 100644
--- a/content/publications/solving-big-data-challenges-for-enterprise-application-performance-management.md
+++ b/content/publications/solving-big-data-challenges-for-enterprise-application-performance-management.md
@@ -7,4 +7,6 @@ publication_type = "Journal Article"
 research = ["data-management"]
 external_url = "https://doi.org/10.14778/2367502.2367512"
 abstract = "As the complexity of enterprise systems increases, the need for monitoring and analyzing such systems also grows. A number of companies have built sophisticated monitoring tools that go far beyond simple resource utilization reports. For example, based on instrumentation and specialized APIs, it is now possible to monitor single method invocations and trace individual transactions across geographically distributed systems. This high-level of detail enables more precise forms of analysis and prediction but comes at the price of high data rates (i.e., big data). To maximize the benefit of data monitoring, the data has to be stored for an extended period of time for ulterior analysis. This new wave of big data analytics imposes new challenges especially for the application performance monitoring systems. The monitoring data has to be stored in a system that can sustain the high data rates and at the same time enable an up-to-date view of the underlying infrastructure. With the advent of modern key-value stores, a variety of data storage systems have emerged that are built with a focus on scalability and high data rates as predominant in this monitoring use case. In this work, we present our experience and a comprehensive performance evaluation of six modern (open-source) data stores in the context of application performance monitoring as part of CA Technologies initiative. We evaluated these systems with data and workloads that can be found in application performance monitoring, as well as, on-line advertisement, power monitoring, and many other use cases. We present our insights not only as performance results but also as lessons learned and our experience relating to the setup and configuration complexity of these data stores in an industry setting."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/1208.4167"
 +++
diff --git a/content/publications/sre-marathon-a-continuous-change-driven-benchmark-for-autonomous-site-reliability-agents.md b/content/publications/sre-marathon-a-continuous-change-driven-benchmark-for-autonomous-site-reliability-agents.md
new file mode 100644
index 0000000..8fa7344
--- /dev/null
+++ b/content/publications/sre-marathon-a-continuous-change-driven-benchmark-for-autonomous-site-reliability-agents.md
@@ -0,0 +1,21 @@
++++
+title = "SRE-Marathon: A Continuous, Change-Driven Benchmark for Autonomous Site Reliability Agents"
+year = 2026
+authors = ["Yifang Tian", "Yingjian Bai", "Yifeng He", "Zichun Chong", "Yuanchen Gao", "Yiran Li", "Hans-Arno Jacobsen"]
+venue = "arXiv"
+publication_type = "ArXiv Preprint"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2609.33023"
+
+# Attribution: https://arxiv.org/abs/2609.33023
+# Website source and scope: https://github.com/bigbigbai/sre-marathon-leaderboard/blob/main/README.md
+research_outcomes = [
+  { title = "SRE-Marathon project website and leaderboard", url = "https://bigbigbai.github.io/sre-marathon-leaderboard/", summary = "Project website and paper-reported leaderboard for continuous, change-driven SRE evaluation. Its GitHub repository contains the website source, not the benchmark implementation." },
+]
++++
+
+This preprint studies continuous SRE operation with overlapping incidents in a
+live Kubernetes environment. Its evaluation follows the progression from
+correlation and localization to repair using recorded system evidence.
+
+[Project website and leaderboard](https://bigbigbai.github.io/sre-marathon-leaderboard/).
diff --git a/content/publications/sregym-a-live-benchmark-for-ai-sre-agents-with-high-fidelity-failure-scenarios.md b/content/publications/sregym-a-live-benchmark-for-ai-sre-agents-with-high-fidelity-failure-scenarios.md
index e953352..261d559 100644
--- a/content/publications/sregym-a-live-benchmark-for-ai-sre-agents-with-high-fidelity-failure-scenarios.md
+++ b/content/publications/sregym-a-live-benchmark-for-ai-sre-agents-with-high-fidelity-failure-scenarios.md
@@ -2,9 +2,18 @@
 title = "SREGym: A Live Benchmark for AI SRE Agents with High-Fidelity Failure Scenarios"
 year = 2026
 authors = ["Jackson Clark", "Yiming Su", "Saad Mohammad Rafid Pial", "Yifang Tian", "Lily Gniedziejko", "Hans-Arno Jacobsen", "Yinfang Chen", "Tianyin Xu"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2605.07161"
+venue = "NeurIPS Evaluations and Datasets Track (accepted; forthcoming)"
+publication_type = "Conference Paper"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2605.07161"
 abstract = "AI agents are increasingly used to diagnose and mitigate failures in production systems, known as agentic Site Reliability Engineering (SRE). Current SRE benchmarks are limited to oversimplistic SRE tasks and are unfortunately hard to extend due to bespoke designs. We present SREGym, a high-fidelity benchmark for SRE agents. SREGym exposes a live system environment built atop real-world cloud-native system stacks, where high-fidelity failure scenarios are simulated through fault injectors. SREGym models the complexity of production environments by simulating (1) a wide range of faults at different layers, (2) various ambient noises, and (3) diverse failure modes such as metastable failures and correlated failures. SREGym is architected as a modular, extensible framework that orchestrates fault and noise injectors across stacks. SREGym currently includes 90 realistic, challenging SRE problems. We use SREGym to evaluate frontier agents and show that their capabilities varies significantly in addressing different kinds of failures, with up to 40% differences in end-to-end results. SREGym is actively maintained as an open-source project and has been used by researchers and practitioners."
+
+# Attribution: https://arxiv.org/abs/2605.07161 and https://tianyin.github.io/pub.html
+research_outcomes = [
+  { title = "SREGym", url = "https://github.com/SREGym/SREGym", summary = "A live cloud-native benchmark for evaluating how SRE agents diagnose and mitigate injected operational failures." },
+]
 +++
+
+Accepted to NeurIPS 2026, as listed on the
+[co-author's publication page](https://yinfangchen.github.io/publications/).
+The full-text link points to the arXiv version; proceedings are forthcoming.
diff --git a/content/publications/the-debs-2022-grand-challenge-detecting-trading-trends-in-financial-tick-data.md b/content/publications/the-debs-2022-grand-challenge-detecting-trading-trends-in-financial-tick-data.md
index 42197d3..1828ead 100644
--- a/content/publications/the-debs-2022-grand-challenge-detecting-trading-trends-in-financial-tick-data.md
+++ b/content/publications/the-debs-2022-grand-challenge-detecting-trading-trends-in-financial-tick-data.md
@@ -9,4 +9,6 @@ research = ["data-management"]
 tags = ["benchmarking", "event-processing", "stream-processing"]
 external_url = "https://doi.org/10.1145/3524860.3539645"
 abstract = "The DEBS Grand Challenge (GC) is an annual programming competition open to practitioners from both academia and industry. The GC 2022 edition focuses on real-time complex event processing of high-volume tick data provided by Infront Financial Technology GmbH. The goal of the challenge is to efficiently compute specific trend indicators and detect patterns in these indicators like those used by real-life traders to decide on buying or selling in financial markets. The data set Trading Data used for benchmarking contains 289 million tick events from approximately 5500+ financial instruments that had been traded on the three major exchanges Amsterdam (NL), Paris (FR), and Frankfurt am Main (GER) over the course of a full week in 2021. The data set is made publicly available. In addition to correctness and performance, submissions must explicitly focus on reusability and practicability. Hence, participants must address specific nonfunctional requirements and are asked to build upon open-source platforms. This paper describes the required scenario and the data set Trading Data, defines the queries of the problem statement, and explains the enhancements made to the evaluation platform Challenger that handles data distribution, dynamic subscriptions, and remote evaluation of the submissions."
+# The author-deposited preprint reverses the title/subtitle order and names this paper's exact DOI; it is not the separate Zenodo dataset.
+preprint_url = "https://arxiv.org/abs/2206.13237"
 +++
diff --git a/content/publications/the-evolution-of-distributed-systems-for-graph-neural-networks-and-their-origin-in-graph-processing-and-deep-learning-a-survey.md b/content/publications/the-evolution-of-distributed-systems-for-graph-neural-networks-and-their-origin-in-graph-processing-and-deep-learning-a-survey.md
index 01b1483..d9682a8 100644
--- a/content/publications/the-evolution-of-distributed-systems-for-graph-neural-networks-and-their-origin-in-graph-processing-and-deep-learning-a-survey.md
+++ b/content/publications/the-evolution-of-distributed-systems-for-graph-neural-networks-and-their-origin-in-graph-processing-and-deep-learning-a-survey.md
@@ -4,7 +4,9 @@ year = 2024
 authors = ["Jana Vatter", "Ruben Mayer", "Hans-Arno Jacobsen"]
 venue = "ACM Computing Surveys"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3597428"
 abstract = "Graph neural networks (GNNs) are an emerging research field. This specialized deep neural network architecture is capable of processing graph structured data and bridges the gap between graph processing and deep learning. As graphs are everywhere, GNNs can be applied to various domains including recommendation systems, computer vision, natural language processing, biology, and chemistry. With the rapid growing size of real-world graphs, the need for efficient and scalable GNN training solutions has come. Consequently, many works proposing GNN systems have emerged throughout the past few years. However, there is an acute lack of overview, categorization, and comparison of such systems. We aim to fill this gap by summarizing and categorizing important methods and techniques for large-scale GNN solutions. Additionally, we establish connections between GNN systems, graph processing systems, and deep learning systems."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2305.13854"
 +++
diff --git a/content/publications/towards-an-optimally-distributed-quantum-fourier-transform-circuit.md b/content/publications/towards-an-optimally-distributed-quantum-fourier-transform-circuit.md
index f0df3c4..06ce23d 100644
--- a/content/publications/towards-an-optimally-distributed-quantum-fourier-transform-circuit.md
+++ b/content/publications/towards-an-optimally-distributed-quantum-fourier-transform-circuit.md
@@ -5,6 +5,6 @@ authors = ["Zachary Vernec", "Michael Silver", "Hans-Arno Jacobsen"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["quantum-computing-systems"]
-external_url = "https://arxiv.org/abs/2606.18494"
+preprint_url = "https://arxiv.org/abs/2606.18494"
 abstract = "A promising avenue for scaling quantum computing is to connect quantum processing units (QPUs) by generating entanglement between them. This requires circuit partitioning: partially rewriting quantum circuits to run on a distributed quantum system using quantum teleportation protocols, while preserving the unitary operation implemented by the circuit. The key metric to minimize when partitioning is the e-bit count, defined as the number of maximally entangled qubit pairs that must be generated between QPUs. We focus on partitioning the quantum Fourier transform (QFT) circuit, which is widely used as a subroutine in quantum algorithms such as quantum phase estimation and arithmetic circuits. Specifically, we present a partitioning scheme based on optimal gate-packing, compare it against prior analytical partitioning schemes for the QFT, and evaluate it against partitions produced by general-purpose circuit partitioning algorithms. We further validate our approach by implementing the partitioned circuit on quantum hardware."
 +++
diff --git a/content/publications/transfer-learning-with-time-series-data-a-systematic-mapping-study.md b/content/publications/transfer-learning-with-time-series-data-a-systematic-mapping-study.md
index 54d3bf9..23f18c4 100644
--- a/content/publications/transfer-learning-with-time-series-data-a-systematic-mapping-study.md
+++ b/content/publications/transfer-learning-with-time-series-data-a-systematic-mapping-study.md
@@ -4,7 +4,7 @@ year = 2021
 authors = ["Manuel Weber", "Maximilian Auch", "Christoph Doblander", "Peter Mandl", "Hans-Arno Jacobsen"]
 venue = "IEEE Access"
 publication_type = "Journal Article"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1109/access.2021.3134628"
 abstract = "Transfer Learning is a well-studied concept in machine learning, that relaxes the assumption that training and testing data need to be drawn from the same distribution. Recent success in applying transfer learning in the area of computer vision has motivated research on transfer learning also in context of time series data. This benefits learning in various time series domains, including a variety of domains based on sensor values. In this paper, we conduct a systematic mapping study of literature on transfer learning with time series data. Following the review guidelines of Kitchenham and Charters, we identify and analyze 223 relevant publications. We describe the pursued approaches and point out trends. Especially during the last two years, there has been a vast increase in the number of publications on the topic. This paper’s findings can help researchers as well as practitioners getting into the field and can help identify research gaps."
 +++
diff --git a/content/publications/v-guard.md b/content/publications/v-guard.md
index 32a8632..cc24681 100644
--- a/content/publications/v-guard.md
+++ b/content/publications/v-guard.md
@@ -6,6 +6,6 @@ venue = "arXiv"
 publication_type = "ArXiv Preprint"
 research = ["data-management"]
 tags = ["blockchain", "consensus", "v2x"]
-external_url = "https://arxiv.org/abs/2301.06210"
+preprint_url = "https://arxiv.org/abs/2301.06210"
 abstract = "This paper presents V-Guard, a new permissioned blockchain that achieves consensus for vehicular data under changing memberships, targeting the problem in V2X networks where vehicles are often intermittently connected on the roads. To achieve this goal, V-Guard integrates membership management into the consensus process for agreeing on data entries. It binds a data entry with a membership configuration profile that describes responsible vehicles for achieving consensus for the data entry. As such, V-Guard produces chained consensus results of both data entries and their residing membership profiles, which enables consensus to be achieved seamlessly under changing memberships. In addition, V-Guard separates the ordering of transactions from consensus, allowing concurrent ordering instances and periodic consensus instances to order and commit data entries. These features make V-Guard efficient for achieving consensus under dynamic memberships with high throughput and latency performance."
 +++
diff --git a/content/publications/wavegas-waveform-relaxation-for-scaling-graph-neural-networks.md b/content/publications/wavegas-waveform-relaxation-for-scaling-graph-neural-networks.md
index 5495f9a..01f696d 100644
--- a/content/publications/wavegas-waveform-relaxation-for-scaling-graph-neural-networks.md
+++ b/content/publications/wavegas-waveform-relaxation-for-scaling-graph-neural-networks.md
@@ -4,7 +4,7 @@ year = 2025
 authors = ["Jana Vatter", "Mykhaylo Zayats", "Marcos Martínez Galindo", "Vanessa López", "Ruben Mayer", "Hans-Arno Jacobsen", "Hoang Thanh Lam"]
 venue = "arXiv"
 publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2502.19986"
+research = ["rca-aiops"]
+preprint_url = "https://arxiv.org/abs/2502.19986"
 abstract = "With the ever-growing size of real-world graphs, numerous techniques to overcome resource limitations when training Graph Neural Networks (GNNs) have been developed. One such approach, GNNAutoScale (GAS), uses graph partitioning to enable training under constrained GPU memory. GAS also stores historical embedding vectors, which are retrieved from one-hop neighbors in other partitions, ensuring critical information is captured across partition boundaries. The historical embeddings which come from the previous training iteration are stale compared to the GAS estimated embeddings, resulting in approximation errors of the training algorithm. Furthermore, these errors accumulate over multiple layers, leading to suboptimal node embeddings. To address this shortcoming, we propose two enhancements: first, WaveGAS, inspired by waveform relaxation, performs multiple forward passes within GAS before the backward pass, refining the approximation of historical embeddings and gradients to improve accuracy; second, a gradient-tracking method that stores and utilizes more accurate historical gradients during training. Empirical results show that WaveGAS enhances GAS and achieves better accuracy, even outperforming methods that train on full graphs, thanks to its robust estimation of node embeddings."
 +++
diff --git a/content/publications/when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry.md b/content/publications/when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry.md
index 99e1ca3..2d09602 100644
--- a/content/publications/when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry.md
+++ b/content/publications/when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry.md
@@ -1,10 +1,23 @@
 +++
-title = "When Agentic Executions Fail: Detecting and Localizing Runtime Faults from Telemetry"
+title = "AgentChaosBench: Benchmarking External Fault Detection and Localization in Agentic Systems"
+slug = "when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry"
 year = 2026
-authors = ["Chenkai Zhang", "Yiran Li", "Yifang Tian", "Michalis Bachras", "Hans-Arno Jacobsen"]
-venue = "arXiv"
-publication_type = "ArXiv Preprint"
-research = ["distributed-machine-learning"]
-external_url = "https://arxiv.org/abs/2608.14680"
+authors = ["Chenkai Zhang", "Yiran Li", "Yifang Tian", "Michail Bachras", "Hans-Arno Jacobsen"]
+venue = "Proceedings of the 1st International Workshop on Trustworthy and Responsible aUtonomous SysTems (TRUST '26), pp. 75–82"
+publication_type = "Workshop Paper"
+research = ["rca-aiops"]
+external_url = "https://doi.org/10.1145/3843782.3844662"
+tags = ["AgentChaosBench", "When Agentic Executions Fail: Detecting and Localizing Runtime Faults from Telemetry"]
+# Keep the earlier manuscript title searchable after updating the citation.
+# ACM publisher metadata: https://api.crossref.org/works/10.1145/3843782.3844662
 abstract = "Reliability in LLM-based agentic systems is a property of the whole execution (its tool calls, model calls, guardrails, and inter-agent messages), not of the final answer alone, yet evaluating only task outcomes reveals little about how or why a run fails. We present AGENTCHAOSBENCH, a benchmark for detecting and localizing runtime faults in agentic systems from their execution telemetry. We run five heterogeneous applications that coordinate agents over the Agent-to-Agent protocol and call tools through the Model Context Protocol, and inject ten types of operational fault (unavailable or slow tools, corrupted or oversized responses, and delayed, looped, or misrouted delegations and bypassed guardrails) at their tool, model, guardrail, and inter-agent boundaries, alongside a no-fault control. The resulting dataset contains 275 sanitized traces: 250 faulty executions spanning ten fault types and 25 no-fault controls. Each faulty trace is aligned with the no-fault execution of the same input; fault-type labels and, where applicable, location labels are held out from diagnosis. On structured single-trace inputs, a first set of zero-shot LLM baselines shows the task is far from solved: local detectors up to 14B parameters reach only 13.6-19.2% top-1 fault-type accuracy and the frontier DeepSeek-v4-pro only 24.8%, while jointly identifying the fault type and its location tops out at 22%; reference-dependent faults (above all a bypassed guardrail) stay near-unsolved from a single trace. An aligned reference improves selected relative faults but does not resolve guardrail bypass. The held-out labels and compact prediction format support reproducible comparison of LLM-based and non-LLM diagnosis methods."
+
+# Attribution: https://arxiv.org/abs/2608.14680 and the authors' project links.
+research_outcomes = [
+  { title = "AgentChaosBench", url = "https://github.com/kevinzck8k/agentic-fault-diagnosis", summary = "Fault-detection and localization evaluation tools with 275 execution traces from five agentic applications." },
+]
+preprint_url = "https://arxiv.org/abs/2608.14680"
 +++
+
+The preprint linked above uses the earlier title *When Agentic Executions Fail:
+Detecting and Localizing Runtime Faults from Telemetry*.
diff --git a/content/publications/when-log-parsing-removes-what-matters-detecting-failures-in-distributed-gpu-workloads.md b/content/publications/when-log-parsing-removes-what-matters-detecting-failures-in-distributed-gpu-workloads.md
new file mode 100644
index 0000000..a85e21c
--- /dev/null
+++ b/content/publications/when-log-parsing-removes-what-matters-detecting-failures-in-distributed-gpu-workloads.md
@@ -0,0 +1,24 @@
++++
+title = "When Log Parsing Removes What Matters: Detecting Failures in Distributed GPU Workloads"
+year = 2026
+authors = ["Boyuan Cui", "Yuyang Zeng", "Yifang Tian", "Hans-Arno Jacobsen"]
+venue = "Proceedings of the 1st International Workshop on Reliable and trustworthy Automated Software Engineering (RASE '26), pp. 60–63"
+publication_type = "Workshop Paper"
+research = ["rca-aiops"]
+external_url = "https://doi.org/10.1145/3820756.3844919"
+# ACM publisher metadata: https://api.crossref.org/works/10.1145/3820756.3844919
+
+# Attribution: data availability statement, https://arxiv.org/abs/2610.14009
+# No top-level repository license was present when checked on 2026-10-22.
+research_outcomes = [
+  { title = "GF-Bench artifacts", url = "https://github.com/cuiboyuan/nccl_log_rca_bench", summary = "Tools for generating multi-GPU failures, collecting NCCL logs, and evaluating anomaly detection, with sample data and scripts to reproduce the experiments." },
+]
+preprint_url = "https://arxiv.org/abs/2610.14009"
++++
+
+The preprint linked above was revised after publication
+and differs from the RASE 2026 paper linked above.
+
+The paper introduces [GF-Bench](https://github.com/cuiboyuan/nccl_log_rca_bench)
+for generating controlled faults in distributed GPU workloads and examines how
+log parsing affects failure detection.
diff --git a/content/publications/where-is-my-training-bottleneck-hidden-trade-offs-in-deep-learning-preprocessing-pipelines.md b/content/publications/where-is-my-training-bottleneck-hidden-trade-offs-in-deep-learning-preprocessing-pipelines.md
index 3abeae3..710ce69 100644
--- a/content/publications/where-is-my-training-bottleneck-hidden-trade-offs-in-deep-learning-preprocessing-pipelines.md
+++ b/content/publications/where-is-my-training-bottleneck-hidden-trade-offs-in-deep-learning-preprocessing-pipelines.md
@@ -4,7 +4,9 @@ year = 2022
 authors = ["Alexander Isenko", "Ruben Mayer", "Jeffrey Jedele", "Hans-Arno Jacobsen"]
 venue = "Proceedings of the 2022 International Conference on Management of Data"
 publication_type = "Conference Paper"
-research = ["distributed-machine-learning"]
+research = ["rca-aiops"]
 external_url = "https://doi.org/10.1145/3514221.3517848"
 abstract = "Preprocessing pipelines in deep learning aim to provide sufficient data throughput to keep the training processes busy. Maximizing resource utilization is becoming more challenging as the throughput of training processes increases with hardware innovations (e.g., faster GPUs, TPUs, and inter-connects) and advanced parallelization techniques that yield better scalability. At the same time, the amount of training data needed in order to train increasingly complex models is growing. As a consequence of this development, data preprocessing and provisioning are becoming a severe bottleneck in end-to-end deep learning pipelines."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2202.08679"
 +++
diff --git a/content/publications/why-do-my-blockchain-transactions-fail-a-study-of-hyperledger-fabric.md b/content/publications/why-do-my-blockchain-transactions-fail-a-study-of-hyperledger-fabric.md
index eb8c1c6..58fd6be 100644
--- a/content/publications/why-do-my-blockchain-transactions-fail-a-study-of-hyperledger-fabric.md
+++ b/content/publications/why-do-my-blockchain-transactions-fail-a-study-of-hyperledger-fabric.md
@@ -7,4 +7,9 @@ publication_type = "Conference Paper"
 research = ["data-management"]
 external_url = "https://doi.org/10.1145/3448016.3452823"
 abstract = "Permissioned blockchain systems promise to provide both decentralized trust and privacy. Hyperledger Fabric is currently one of the most wide-spread permissioned blockchain systems and is heavily promoted both in industry and academia. Due to its optimistic concurrency model, the transaction failure rates in Fabric can become a bottleneck. While there is active research to reduce failures, there is a lack of understanding on their root cause and, consequently, a lack of guidelines on how to configure Fabric optimally for different scenarios. To close this gap, in this paper, we first introduce a formal definition of the different types of transaction failures in Fabric. Then, we develop a comprehensive testbed and benchmarking system, HyperLedgerLab, along with four different chaincodes that represent realistic use cases and a chaincode/workload generator. Using HyperLedgerLab, we conduct exhaustive experiments to analyze the impact of different parameters of Fabric such as block size, endorsement policies, and others, on transaction failures. We further analyze three recently proposed optimizations from the literature, Fabric++, Streamchain and FabricSharp, and evaluate under which conditions they reduce the failure rates. Finally, based on our results, we provide recommendations for Fabric practitioners on how to configure the system and also propose new research directions."
+# Preprint correspondence checked against the author-deposited manuscript and published citation.
+preprint_url = "https://arxiv.org/abs/2103.04681"
 +++
+
+The arXiv preprint linked above is an extended version of the SIGMOD 2021 paper.
+The authors request citation of the published version.
diff --git a/content/research/_index.md b/content/research/_index.md
index 8c2aea3..55a99bf 100644
--- a/content/research/_index.md
+++ b/content/research/_index.md
@@ -1,5 +1,5 @@
 +++
 title = "Research"
 layout = "overview"
-summary = "Three active research themes connect MSRG's work in systems, AI infrastructure, and quantum computing."
+summary = "Three research themes connect MSRG's work in data management, root cause analysis and AIOps, and quantum computing systems, including continuing work in distributed AI and machine learning."
 +++
diff --git a/content/research/data-management/_index.md b/content/research/data-management/_index.md
index 16b1d8d..88c67a9 100644
--- a/content/research/data-management/_index.md
+++ b/content/research/data-management/_index.md
@@ -20,6 +20,11 @@ can repeat computations and produce the same intermediate data. Our approach
 detects these overlaps online and coordinates execution so queries can share
 work and improve throughput.
 
+Our [Root Cause Analysis & AIOps](/research/rca-aiops/) research examines the
+operational reliability of cloud-native systems, using metrics, logs, traces,
+and service dependencies to support fault diagnosis, incident response, and
+automated repair.
+
 ## Blockchain, Distributed Ledger Technology & Consensus
 
 We develop consensus algorithms, fault-tolerance mechanisms, and consistency
diff --git a/content/research/distributed-machine-learning/_index.md b/content/research/distributed-machine-learning/_index.md
deleted file mode 100644
index 0047f4a..0000000
--- a/content/research/distributed-machine-learning/_index.md
+++ /dev/null
@@ -1,40 +0,0 @@
-+++
-title = "Distributed Machine Learning"
-layout = "area"
-weight = 20
-accent = "sage"
-question = "How can distributed learning systems become more efficient and scalable?"
-blurb = """At MSRG, we tackle core challenges in distributed deep learning systems. Our research focuses on optimizing training and inference across distributed infrastructure, developing novel approaches for **resource allocation** and **scheduling of training jobs**, and creating **system-level improvements** for **deep learning** and **graph learning** frameworks. We are advancing federated learning for resource-constrained and embedded devices, optimizing performance and communication efficiency in edge computing environments. Through our work, we aim to make distributed deep learning more efficient and accessible, enabling faster training and better resource utilization."""
-# Description aligned with https://msrg.org/research/distributed-machine-learning/
-+++
-
-MSRG studies performance optimization, privacy-preserving and federated
-learning, and graph processing as connected challenges in distributed machine
-learning.
-
-## Distributed Deep Learning
-
-Large training and inference workloads need substantial computing resources
-and distributed infrastructure. We investigate how resource allocation, job
-scheduling, and optimizations within learning systems can improve performance
-and reduce costs.
-
-Our federated learning research also addresses embedded devices with limited
-resources. We work across hardware-level optimizations and distributed learning
-algorithms to improve efficiency in edge deployments.
-
-## Graph Processing & Learning
-
-Graph workloads combine irregular memory access, interdependent computations,
-and large datasets. We develop partitioning and sampling methods alongside
-system optimizations to make these workloads faster.
-
-Our distributed graph systems target graphs containing billions of vertices
-and edges. We also optimize the training and inference stages of graph neural
-networks.
-
-## Current directions
-
-- Resource allocation and scheduling for distributed training and inference
-- Privacy-preserving and federated learning on embedded and edge devices
-- Graph partitioning, sampling, and scalable graph neural network execution
diff --git a/content/research/quantum-computing-systems/_index.md b/content/research/quantum-computing-systems/_index.md
index 60c8293..9c48664 100644
--- a/content/research/quantum-computing-systems/_index.md
+++ b/content/research/quantum-computing-systems/_index.md
@@ -6,6 +6,13 @@ accent = "copper"
 question = "Can quantum computers deliver reliable and scalable performance to tackle real-world challenges?"
 blurb = """MSRG works on foundational challenges of quantum computing including NISQ-era quantum algorithms, quantum error correction, and qubit noise reduction. We develop distributed quantum algorithms to accelerate electronic structure calculations, enabling more accurate modelling of molecular systems. Simultaneously, we use quantum machine learning techniques to efficiently explore chemical compound space, contributing to material discovery research. This interdisciplinary approach bridges quantum computing, machine learning, and computational chemistry to tackle fundamental challenges in molecular simulation and materials discovery."""
 # Description aligned with https://msrg.org/research/quantum-computing-systems/
+# QSC funding: https://www.qscc.ca/
+# QSC leadership: https://www.qscc.ca/people/
+# Leadership: https://cqiqc.physics.utoronto.ca/news/recent-news/cqiqc-researchers-receive-5-million-for-quantum-software-consortium-initiative/
+acknowledgement = "We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) for the Quantum Software Consortium (QSC) through Alliance Consortia Quantum grant ALLRP587590-23."
+supporters = [
+  { name = "Natural Sciences and Engineering Research Council of Canada (NSERC)", url = "https://nserc-crsng.canada.ca/", logo = "/images/support/nserc.png", width = 145, height = 70 }
+]
 +++
 
 MSRG investigates distributed quantum algorithms and quantum machine learning
@@ -18,11 +25,14 @@ catalysts that could help reduce greenhouse gases. We work with specialists
 such as synthetic inorganic chemists to connect computing research with
 chemical problems.
 
-The group participates in Canada's interdisciplinary Quantum Software
-Consortium, where Prof. Jacobsen serves as scientific director, and the
-University of Toronto's Centre for Quantum Information and Quantum Control.
-These collaborations bring our distributed algorithms and machine learning
-expertise together with university researchers and industry partners.
+Prof. Hans-Arno Jacobsen leads the interdisciplinary
+[Quantum Software Consortium (QSC)](https://www.qscc.ca/) as its scientific
+director.
+
+We also participate in the University of Toronto's Centre for Quantum
+Information and Quantum Control. These collaborations bring our distributed
+algorithms and machine learning expertise together with university researchers
+and industry partners.
 
 ## Current directions
 
diff --git a/content/research/rca-aiops/_index.md b/content/research/rca-aiops/_index.md
new file mode 100644
index 0000000..7e8cef9
--- /dev/null
+++ b/content/research/rca-aiops/_index.md
@@ -0,0 +1,102 @@
++++
+title = "Root Cause Analysis & AIOps"
+slug = "rca-aiops"
+aliases = ["/research/distributed-machine-learning/"]
+legacy_research_ids = ["distributed-machine-learning"]
+layout = "area"
+show_section_links = true
+weight = 20
+accent = "teal"
+question = "How can we turn operational telemetry into reliable diagnosis and actionable incident response?"
+summary = "Root cause analysis, anomaly detection, and telemetry-grounded repair for reliable microservices and AI agents, alongside distributed AI and machine learning."
+publication_blurb = "Recent publications in root cause analysis, AIOps, and our broader distributed AI and machine learning research."
+team_blurb = "Current and visiting MSRG members associated with our AIOps and distributed AI/ML research are shown below."
+blurb = """MSRG develops methods for **root cause analysis** and **AI for IT operations (AIOps)** in distributed systems. We combine **metrics, logs, and traces** with service dependency graphs and LLM agents to investigate failures and produce actionable diagnoses. Our work connects anomaly detection, incident response, automated repair, and realistic evaluation, with a focus on microservices and the reliability of AI agent executions."""
+# Research descriptions are based on the publication abstracts linked below.
+acknowledgement = "We acknowledge the support of Bell Mobility and the Natural Sciences and Engineering Research Council of Canada (NSERC) under grant ALLRP614128-25."
+supporters = [
+  { name = "Bell Mobility", url = "https://www.bell.ca/Mobility", logo = "/images/support/bell.svg", width = 100, height = 60 },
+  { name = "Natural Sciences and Engineering Research Council of Canada (NSERC)", url = "https://nserc-crsng.canada.ca/", logo = "/images/support/nserc.png", width = 145, height = 70 }
+]
++++
+
+## From telemetry to root causes
+
+A failure in one service can surface as errors, slowdowns, or unusual resource
+usage elsewhere. Root cause analysis (RCA) connects these symptoms to the
+underlying fault. Doing so requires reasoning across heterogeneous telemetry
+and the dependencies between services.
+
+We study how AI for IT operations (AIOps) can help engineers move from detecting
+an anomaly to understanding a failure and deciding what to do next. Our research
+brings together statistical analysis, graph structure, and LLM agents, with an
+emphasis on diagnoses and repair decisions grounded in operational evidence.
+
+<div class="research-diagrams">
+
+{{< figure src="/images/research/aiops-vision.svg" alt="AIOps research vision: observe distributed systems, detect anomalies, diagnose causes, and support incident response, with realistic benchmark evaluation." caption="Research vision: connect operational evidence to diagnosis and response, and evaluate each step in realistic systems." >}}
+
+{{< figure src="/images/research/aiops-technical.svg" alt="GALA+ workflow: telemetry and service dependencies feed initial hypotheses, followed by graph-guided investigation using local evidence, producing diagnoses and incident response recommendations evaluated with SURE-Score." caption="Technical example: GALA+ uses service dependencies and local telemetry to guide LLM investigation; SURE-Score assesses the quality of its outputs." >}}
+
+</div>
+
+## Graph-guided diagnosis and incident response
+
+[GALA](/publications/gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices/),
+whose updated framework is called GALA+, uses service dependencies to guide
+LLM agents through local metrics, logs, and traces. It produces ranked diagnoses,
+incident summaries, and action recommendations. SURE-Score, developed with
+industry site reliability engineers, assesses diagnostic quality beyond text
+similarity.
+
+## Message-level anomaly detection
+
+Effective investigation starts with finding useful signals in large volumes of
+telemetry.
+[FAME](/publications/fame-failure-aware-mixture-of-experts-for-message-level-log-anomaly-detection/)
+studies anomaly detection at the level of individual log messages. It uses an
+LLM during offline preparation to help organize failure domains, then trains
+a lightweight router and specialized experts for online detection with limited
+labeling effort.
+
+## Telemetry-grounded repair
+
+Diagnosis can also provide context for changing code and configuration.
+[ORCA](/publications/orca-observability-grounded-program-repair-for-microservice-incidents/)
+compares failure and reference telemetry to build a fault signature, locate
+candidate repair sites, and guide patch generation. Its verification process
+examines patch correctness, test integrity, and telemetry replay to connect
+automated repair with the incident that motivated it.
+
+## Reliable agents and realistic evaluation
+
+AI agents are both tools for incident response and systems whose own failures
+need diagnosis.
+[AgentChaosBench](/publications/when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry/)
+studies fault detection and localization from agent
+execution telemetry, including faults at tool, model, guardrail, and inter-agent
+boundaries.
+
+[SREGym](/publications/sregym-a-live-benchmark-for-ai-sre-agents-with-high-fidelity-failure-scenarios/)
+evaluates SRE agents in live cloud-native environments with injected faults and
+ambient noise. These complementary settings help us examine whether an agent
+can identify a fault, locate its origin, and support incident resolution under
+realistic operating conditions.
+
+[GF-Bench](/publications/when-log-parsing-removes-what-matters-detecting-failures-in-distributed-gpu-workloads/)
+extends failure detection to distributed GPU workloads, while
+[SRE-Marathon](/publications/sre-marathon-a-continuous-change-driven-benchmark-for-autonomous-site-reliability-agents/)
+studies continuous operation with overlapping incidents and code or configuration changes.
+
+## Current directions
+
+- Correlating metrics, logs, and traces across service dependencies
+- Graph-guided and evidence-grounded LLM investigation
+- Connecting root cause diagnosis to remediation and automated repair
+- Evaluating diagnostic quality, agent reliability, and operational usefulness
+
+This research connects with our work in [Data Management](/research/data-management/).
+We also continue research in **distributed AI and machine learning**, including
+resource allocation and scheduling for training and inference, privacy-preserving
+and federated learning on embedded and edge devices, and scalable graph processing
+and graph neural networks.
diff --git a/data/content_schema.json b/data/content_schema.json
index eaf7039..bc821e1 100644
--- a/data/content_schema.json
+++ b/data/content_schema.json
@@ -346,11 +346,19 @@
         },
         {
           "key": "external_url",
-          "label": "Paper URL",
+          "label": "Published / official URL",
           "type": "string",
           "default": "",
-          "required": true,
-          "format": "url"
+          "format": "url",
+          "help": "Link to the published version's DOI, publisher page, or official publication record. Leave blank if only a preprint is available, even if the paper is accepted. Provide this link, a preprint link, or both."
+        },
+        {
+          "key": "preprint_url",
+          "label": "Preprint URL",
+          "type": "string",
+          "default": "",
+          "format": "url",
+          "help": "Link to this paper's preprint on arXiv or another repository. It appears beside the published version with less emphasis, or as an explicitly labeled preprint link when no published link is available. Explain significant version differences in Additional notes."
         },
         {
           "key": "publication_type",
@@ -382,6 +390,42 @@
           "default": [],
           "options_source": "research"
         },
+        {
+          "key": "research_outcomes",
+          "label": "Research outcomes",
+          "add_label": "Add research outcome",
+          "type": "objects",
+          "default": [],
+          "help": "Optional public code, benchmarks, datasets, or project websites. Each outcome is linked automatically to this publication and its selected research areas.",
+          "fields": [
+            {
+              "key": "title",
+              "label": "Outcome name",
+              "type": "string",
+              "default": "",
+              "required": true
+            },
+            {
+              "key": "url",
+              "label": "Project or repository URL",
+              "type": "string",
+              "default": "",
+              "format": "url",
+              "https_only": true,
+              "required": true,
+              "help": "Use a public HTTPS link."
+            },
+            {
+              "key": "summary",
+              "label": "Description",
+              "type": "string",
+              "default": "",
+              "multiline": true,
+              "required": true,
+              "help": "Briefly describe what is publicly available, using plain text."
+            }
+          ]
+        },
         {
           "key": "abstract",
           "label": "Abstract",
diff --git a/docs/aiops-sources.md b/docs/aiops-sources.md
new file mode 100644
index 0000000..e64c299
--- /dev/null
+++ b/docs/aiops-sources.md
@@ -0,0 +1,64 @@
+# AIOps content sources
+
+Checked 2026-11-01. People and portraits come from `content/people/` through
+the same roster and card partials as Quantum Computing Systems. Publications
+come from `content/publications/`; the AIOps page displays up to six recent
+records, matching the other research pages, and links to the complete archive.
+The acknowledgement text and grant number were supplied by the project team.
+The consolidated page uses the shared research layout, with Outcomes followed
+by Publications and then People, as requested in Arno's email. The canonical
+page is `/research/rca-aiops/`; the former Distributed Machine Learning URL
+redirects there. The research identifier is `rca-aiops` in member, publication,
+and dataset records. It covers RCA, AIOps, and the page's broader distributed
+AI/ML work. Renaming the former `distributed-machine-learning` identifier
+preserves the same members and papers; old publication-filter links remain
+supported as aliases.
+
+Research outcomes are stored with their owning records in `content/publications/`
+and can be maintained in the publication editor. They inherit the publication's
+research areas; for example, GALA's evaluation artifacts also appear under Data
+Management. Outcomes are collected from all matching publications, independently
+of the six-reference preview. Source comments beside the outcome fields preserve
+the attribution and release-scope evidence.
+
+## Yuanchen Gao
+
+Added to the shared roster at the maintainer's request, with the PhD role
+confirmed by the maintainer. [Charles Zhang's HKUST page](https://cse.hkust.edu.hk/~charlesz/)
+lists him among current students. The [ORCA manuscript](https://arxiv.org/html/2608.17018v1)
+confirms his HKUST affiliation, public university email, and an IVGS visit to
+Toronto during the work. Visit dates are unknown and are not inferred.
+His card uses the default initials avatar, as requested. No LinkedIn, ORCID,
+or Google Scholar account was reliably matched; same-name materials researchers
+were excluded.
+
+## Publication updates
+
+The published / official URL (`external_url`) points to the published version's
+DOI or publisher page whenever available. The separate `preprint_url` appears
+beside it as a secondary link; explanatory version differences remain in notes.
+Acceptance or presentation alone does not establish an available proceedings
+version; leave the published URL empty until it can be verified. With only a
+preprint URL the website explicitly labels the link as a preprint.
+
+- GALA: [ASE programme](https://conf.researchr.org/details/ase-2026/ase-2026-research-track/5/GALA-Graph-Augmented-LLM-Agents-for-Root-Cause-Analysis-and-Incident-Response-in-Mic), [published version](https://doi.org/10.1145/3832783.3834335). At the maintainer's request, consolidate the earlier GALA preprint and the updated GALA+ work into this single record. Keep the earlier arXiv link in its notes and redirect the old website URL through an alias.
+- AgentChaosBench: [TRUST programme](https://conf.researchr.org/details/ase-2026/trust-2026-papers/12/AgentChaosBench-Benchmarking-External-Fault-Detection-and-Localization-in-Agentic-Sy), [published version](https://doi.org/10.1145/3843782.3844662). Preserve the original page URL and identify the earlier arXiv title in the publication notes.
+- FAME: [ISSRE programme](https://cyprusconferences.org/issre2026/program/). No unverified DOI or pagination added.
+- SREGym: [co-author's publication list](https://yinfangchen.github.io/publications/). Accepted to NeurIPS 2026 Evaluations and Datasets; explicitly marked forthcoming.
+- GF-Bench paper: [arXiv metadata](https://arxiv.org/abs/2610.14009), [published version](https://doi.org/10.1145/3820756.3844919). The revised arXiv version differs from the proceedings paper.
+- SRE-Marathon: [arXiv](https://arxiv.org/abs/2609.33023); retained as a preprint.
+
+## Illustrations and sponsor marks
+
+The two `static/images/research/aiops-*.svg` diagrams are original schematic
+illustrations, not reproductions of paper figures. The vision is a research
+agenda; the technical workflow summarizes the [GALA+ abstract](https://arxiv.org/abs/2608.08968).
+No experimental results are invented. Outcomes distinguish GALA evaluation
+materials from its unreleased core code and SRE-Marathon website source from
+benchmark implementation.
+
+- Bell: unmodified logo geometry and viewBox from the `icon-bell-logo` symbol
+  on [Bell's official site](https://www.bell.ca/), using Bell blue `#00549a`.
+- NSERC: unmodified [official bilingual symbol](https://nserc-crsng.canada.ca/sites/default/files/styles/gcds_full_sm/public/migrated/img/logos/img-logo2-en.png?itok=1FYj3E2S).
+
+Sponsor marks identify project support and remain their owners' property.
diff --git a/docs/publication-import.md b/docs/publication-import.md
index bed5669..8b76be4 100644
--- a/docs/publication-import.md
+++ b/docs/publication-import.md
@@ -15,8 +15,10 @@ the current or alumni roster. Existing entries outside that scope were retained.
   were excluded from the paper backfill.
 - Compared 567 [OpenAlex records](https://openalex.org/authors/A5072791865)
   to find gaps, then checked additions against publisher, conference, arXiv,
-  or institutional records. Each paper uses `external_url` as its single paper
-  link. Import sources are recorded separately in
+  or institutional records. Each paper uses `external_url` for its published /
+  official version and the optional `preprint_url` for its corresponding preprint.
+  At least one link is required. Published links take priority; preprint-only
+  records are explicitly labeled. Import sources are recorded separately in
   [publication provenance](publication-provenance.json).
 - Preserved ordered author lists and existing page paths, notes, and dataset
   associations. Added verified author-name variants to nine member profiles.
diff --git a/hugo.yaml b/hugo.yaml
index f91bcd7..dc715cc 100644
--- a/hugo.yaml
+++ b/hugo.yaml
@@ -42,7 +42,7 @@ permalinks:
   data-sets: /data-sets/:slug/
 
 params:
-  description: Systems research at the University of Toronto spanning data management, distributed machine learning, and quantum computing systems.
+  description: Systems research at the University of Toronto spanning data management, root cause analysis and AIOps, and quantum computing, with continuing work in distributed AI and machine learning.
   # Public contact: https://www.qscc.ca/open-positions/
   contact_email: jacobsen@eecg.utoronto.ca
   contact_phone: "+1 416 946 7586"
@@ -71,9 +71,9 @@ menu:
       parent: research
       pageRef: /research/quantum-computing-systems
       weight: 21
-    - name: Distributed Machine Learning
+    - name: Root Cause Analysis & AIOps
       parent: research
-      pageRef: /research/distributed-machine-learning
+      pageRef: /research/rca-aiops
       weight: 22
     - name: Data Management
       parent: research
diff --git a/layouts/alias.html b/layouts/alias.html
new file mode 100644
index 0000000..7f40f43
--- /dev/null
+++ b/layouts/alias.html
@@ -0,0 +1,25 @@
+{{- if and .Page .Page.Params.legacy_research_ids -}}
+<!doctype html>
+<html lang="{{ site.Language.Locale }}">
+  <head>
+    <meta charset="utf-8">
+    <title>{{ .Page.Title }}</title>
+    <link rel="canonical" href="{{ .Permalink }}">
+    <script>
+      window.location.replace({{ .Permalink }} + window.location.search + window.location.hash);
+    </script>
+    <noscript><meta http-equiv="refresh" content="0; url={{ .Permalink }}"></noscript>
+  </head>
+  <body><p><a href="{{ .Permalink }}">Continue to {{ .Page.Title }}</a></p></body>
+</html>
+{{- else -}}
+<!doctype html>
+<html lang="{{ site.Language.Locale }}">
+  <head>
+    <title>{{ .Permalink }}</title>
+    <link rel="canonical" href="{{ .Permalink }}">
+    <meta charset="utf-8">
+    <meta http-equiv="refresh" content="0; url={{ .Permalink }}">
+  </head>
+</html>
+{{- end -}}
diff --git a/layouts/partials/footer.html b/layouts/partials/footer.html
index b7b5fef..fb2fada 100644
--- a/layouts/partials/footer.html
+++ b/layouts/partials/footer.html
@@ -6,8 +6,8 @@
     <div>
       <img class="footer-logo" src="{{ "images/msrg-logo-white.png" | relURL }}" alt="Middleware Systems Research Group">
       <p>
-        Research on data systems, distributed learning infrastructure, and
-        quantum computing systems at the University of Toronto.
+        Research on data systems, root cause analysis and AIOps, quantum computing,
+        and distributed learning infrastructure at the University of Toronto.
       </p>
     </div>
     <div>
diff --git a/layouts/partials/publication-links.html b/layouts/partials/publication-links.html
new file mode 100644
index 0000000..fd9064c
--- /dev/null
+++ b/layouts/partials/publication-links.html
@@ -0,0 +1,37 @@
+{{/* The two URLs describe versions, independently of venue/type or acceptance. */}}
+{{ $page := .page }}
+{{ $compact := .compact | default false }}
+{{ $published := $page.Params.external_url }}
+{{ $preprint := $page.Params.preprint_url }}
+{{ $publishedHost := "" }}
+{{ with $published }}{{ $publishedHost = lower (urls.Parse .).Hostname }}{{ end }}
+{{ $publishedLabel := "Published version" }}
+{{ if eq $page.Params.publication_type "Technical Report" }}
+  {{ $publishedLabel = "Technical report" }}
+{{ else if eq $page.Params.publication_type "Dataset" }}
+  {{ $publishedLabel = "View dataset" }}
+{{ else if eq $page.Params.publication_type "Software" }}
+  {{ $publishedLabel = "View software" }}
+{{ else if in (slice "dblp.org" "www.dblp.org") $publishedHost }}
+  {{ $publishedLabel = "Publication record" }}
+{{ else if and (eq $page.Params.publication_type "Workshop Paper") (in (slice "blog.genlaw.org" "openreview.net" "neurips.cc") $publishedHost) }}
+  {{/* Official workshop copies may be non-archival; publisher/DOI links keep their label. */}}
+  {{ $publishedLabel = "Workshop version" }}
+{{ end }}
+{{ $preprintLabel := "Preprint" }}
+{{ with $preprint }}
+  {{ $host := lower (urls.Parse .).Hostname }}
+  {{ if or (eq $host "arxiv.org") (strings.HasSuffix $host ".arxiv.org") }}
+    {{ $preprintLabel = printf "%s (arXiv)" $preprintLabel }}
+  {{ end }}
+{{ end }}
+{{ if or $published $preprint }}
+  <p class="content-actions publication-links{{ if $compact }} publication-links-compact{{ end }}">
+    {{ with $published }}
+      <a{{ if not $compact }} class="button"{{ end }} href="{{ . }}" target="_blank" rel="noopener noreferrer">{{ $publishedLabel }}</a>
+    {{ end }}
+    {{ with $preprint }}
+      <a class="{{ if not $compact }}button button-secondary {{ end }}publication-preprint" href="{{ . }}" target="_blank" rel="noopener noreferrer">{{ $preprintLabel }}</a>
+    {{ end }}
+  </p>
+{{ end }}
diff --git a/layouts/partials/research-outcome.html b/layouts/partials/research-outcome.html
new file mode 100644
index 0000000..f2cbb34
--- /dev/null
+++ b/layouts/partials/research-outcome.html
@@ -0,0 +1,7 @@
+<article class="card">
+  <h3><a href="{{ .outcome.url }}">{{ .outcome.title }}</a></h3>
+  <p>{{ .outcome.summary }}</p>
+  {{ with .publication }}
+    <p class="content-actions"><a href="{{ .RelPermalink }}">Related publication</a></p>
+  {{ end }}
+</article>
diff --git a/layouts/publications/archive.html b/layouts/publications/archive.html
index 1a71193..41d9eb3 100644
--- a/layouts/publications/archive.html
+++ b/layouts/publications/archive.html
@@ -16,7 +16,7 @@ <h1>{{ .Title }}</h1>
       <div class="filter-toolbar" role="group" aria-labelledby="publication-research-label">
         <button class="filter-chip is-active" type="button" aria-pressed="true" data-publication-research="all">All research areas</button>
         {{ range (site.GetPage "/research").Pages.ByWeight }}
-          <button class="filter-chip" type="button" aria-pressed="false" data-publication-research="{{ .File.ContentBaseName }}">{{ .Title }}</button>
+          <button class="filter-chip" type="button" aria-pressed="false" data-publication-research="{{ .File.ContentBaseName }}"{{ with .Params.legacy_research_ids }} data-publication-research-aliases="{{ delimit . "," }}"{{ end }}>{{ .Title }}</button>
         {{ end }}
       </div>
       <div class="section-header-row">
diff --git a/layouts/publications/single.html b/layouts/publications/single.html
index c17b2a8..3f8d7da 100644
--- a/layouts/publications/single.html
+++ b/layouts/publications/single.html
@@ -8,9 +8,7 @@ <h1>{{ .Title }}</h1>
         <p class="page-summary">
           {{ .Params.venue }}, {{ .Params.year }} · {{ .Params.publication_type }}
         </p>
-        {{ with .Params.external_url }}
-          <p class="content-actions"><a class="button" href="{{ . }}" target="_blank" rel="noopener noreferrer">Read paper</a></p>
-        {{ end }}
+        {{ partial "publication-links.html" (dict "page" .) }}
       </div>
     </section>
 
@@ -25,9 +23,21 @@ <h1>{{ .Title }}</h1>
             {{ end }}
           {{ else with .Params.external_url }}
             <p><a href="{{ . }}" target="_blank" rel="noopener noreferrer">Read the abstract and full publication at the source.</a></p>
+          {{ else with .Params.preprint_url }}
+            <p><a href="{{ . }}" target="_blank" rel="noopener noreferrer">Read the abstract and full preprint at the source.</a></p>
           {{ else }}
             <p>The abstract has not been added yet.</p>
           {{ end }}
+          {{ with .Params.research_outcomes }}
+            <section id="research-outcomes">
+              <h2>Research outcomes</h2>
+              <div class="stack">
+                {{ range . }}
+                  {{ partial "research-outcome.html" (dict "outcome" .) }}
+                {{ end }}
+              </div>
+            </section>
+          {{ end }}
           {{ with .Content }}
             <h2>Additional notes</h2>
             {{ . }}
diff --git a/layouts/research/area.html b/layouts/research/area.html
index f7d90be..c5f191e 100644
--- a/layouts/research/area.html
+++ b/layouts/research/area.html
@@ -3,7 +3,17 @@
   {{ $roster := partialCached "people/roster.html" site }}
   {{ $people := where $roster.current "Params.research" "intersect" (slice $slug) }}
   {{ $publications := where (site.GetPage "/publications").RegularPages "Params.research" "intersect" (slice $slug) }}
+  {{ $listedPublications := first 6 (sort $publications "Params.year" "desc") }}
   {{ $datasets := where (site.GetPage "/data-sets").RegularPages "Params.research" "intersect" (slice $slug) }}
+  {{ $outcomes := slice }}
+  {{ range $publications.GroupByParam "year" "desc" }}
+    {{ range .Pages.ByTitle }}
+      {{ $publication := . }}
+      {{ range .Params.research_outcomes }}
+        {{ $outcomes = $outcomes | append (dict "outcome" . "publication" $publication) }}
+      {{ end }}
+    {{ end }}
+  {{ end }}
 
   <article class="research-area">
     <section class="page-hero scroll-section">
@@ -12,11 +22,22 @@
         <h1>{{ .Title }}</h1>
         {{ with .Params.question }}<p class="page-question">{{ . }}</p>{{ end }}
         <div class="page-intro">{{ .Params.blurb | markdownify }}</div>
+        {{ with .Params.acknowledgement }}
+          <p>{{ . }}</p>
+          <div class="research-support-logos">
+            {{ range $.Params.supporters }}
+              <a href="{{ .url }}" target="_blank" rel="noopener noreferrer"><img src="{{ .logo | relURL }}" alt="{{ .name }}" width="{{ .width }}" height="{{ .height }}"></a>
+            {{ end }}
+          </div>
+        {{ end }}
         <p class="page-summary">
-          {{ len $people }} {{ if eq (len $people) 1 }}researcher{{ else }}researchers{{ end }}
+          {{ len $people }} {{ if eq (len $people) 1 }}researcher{{ else }}researchers{{ end }} in this area
           · {{ len $publications }} {{ if eq (len $publications) 1 }}publication{{ else }}publications{{ end }}
           · {{ len $datasets }} {{ if eq (len $datasets) 1 }}data set{{ else }}data sets{{ end }}
         </p>
+        {{ if .Params.show_section_links }}
+          <p><a href="#outcomes">Outcomes</a> · <a href="#publications">Publications</a> · <a href="#people">People</a></p>
+        {{ end }}
       </div>
     </section>
 
@@ -28,26 +49,42 @@ <h1>{{ .Title }}</h1>
       </div>
     </section>
 
-    <section class="page-section page-section-alt research-team">
+    {{ if or $outcomes $datasets }}
+    <section id="outcomes" class="page-section page-section-alt scroll-section">
       <div class="site-shell">
-
-        {{ partial "people/group-cards.html" $people }}
-
+        <div class="section-heading">
+          <h2>Research outcomes</h2>
+        </div>
+        <div class="card-grid">
+          {{ range $outcomes }}
+            {{ partial "research-outcome.html" . }}
+          {{ end }}
+          {{ range sort $datasets "Title" }}
+            <article class="card">
+              <h3><a href="{{ .RelPermalink }}">{{ .Title }}</a></h3>
+              <p>{{ .Params.summary }}</p>
+            </article>
+          {{ end }}
+        </div>
       </div>
     </section>
+    {{ end }}
 
-    <section class="page-section scroll-section">
+    <section id="publications" class="page-section scroll-section">
       <div class="site-shell">
         <div class="section-heading">
-          <h2>Publications</h2>
+          <h2>{{ if lt (len $listedPublications) (len $publications) }}Recent publications{{ else }}Publications{{ end }}</h2>
+          {{ with .Params.publication_blurb }}<p>{{ . }}</p>{{ end }}
         </div>
         <div class="stack">
           {{ if $publications }}
-            {{ range first 6 (sort $publications "Params.year" "desc") }}
-              <a class="list-link-card" href="{{ .RelPermalink }}">
-                <strong>{{ .Title }}</strong>
-                <span>{{ .Params.venue }}, {{ .Params.year }}</span>
-              </a>
+            {{ range $listedPublications }}
+              <article class="list-card">
+                <p class="muted">{{ partial "publication-authors.html" (dict "authors" .Params.authors) }}</p>
+                <h3><a href="{{ .RelPermalink }}">{{ .Title }}</a></h3>
+                <p>{{ .Params.venue }}, {{ .Params.year }}</p>
+                {{ partial "publication-links.html" (dict "page" . "compact" true) }}
+              </article>
             {{ end }}
           {{ else }}
             <p class="muted">No publications are listed for this research area yet.</p>
@@ -57,22 +94,17 @@ <h2>Publications</h2>
       </div>
     </section>
 
-    {{ if $datasets }}
-    <section class="page-section page-section-alt scroll-section">
+    <section id="people" class="page-section page-section-alt research-team">
       <div class="site-shell">
         <div class="section-heading">
-          <h2>Data sets</h2>
-        </div>
-        <div class="card-grid">
-          {{ range sort $datasets "Title" }}
-            <article class="card">
-              <h3><a href="{{ .RelPermalink }}">{{ .Title }}</a></h3>
-              <p>{{ .Params.summary }}</p>
-            </article>
-          {{ end }}
+          <h2>Researchers in this area</h2>
+          <p>{{ with .Params.team_blurb }}{{ . }}{{ else }}Current and visiting MSRG members associated with {{ .Title }} are shown below.{{ end }}
+            <a href="{{ "/people/" | relURL }}">View the full MSRG team</a>.</p>
         </div>
+
+        {{ partial "people/group-cards.html" $people }}
+
       </div>
     </section>
-    {{ end }}
   </article>
 {{ end }}
diff --git a/scripts/browser_check.cjs b/scripts/browser_check.cjs
index 0c09051..4d86b4f 100644
--- a/scripts/browser_check.cjs
+++ b/scripts/browser_check.cjs
@@ -173,11 +173,20 @@ let browser;
   await search.fill('Grier 2025'); await waitForSearch(); assert.equal(await visible.count(), 1);
   await search.fill('Woisetschlager'); await waitForSearch(); assert(await visible.count() > 0);
   await search.fill('Michael Dang’ana'); await waitForSearch(); assert(await visible.count() > 0);
+  for (const [formerTitle, slug] of [
+    ['When Agentic Executions Fail: Detecting and Localizing Runtime Faults from Telemetry', 'when-agentic-executions-fail-detecting-and-localizing-runtime-faults-from-telemetry'],
+    ['GALA: Can Graph-Augmented Large Language Model Agentic Workflows Elevate Root Cause Analysis?', 'gala-graph-augmented-llm-agents-for-root-cause-analysis-and-incident-response-in-microservices'],
+  ]) {
+    await search.fill(formerTitle); await waitForSearch();
+    assert.equal(await visible.count(), 1, 'The full former title must find its consolidated publication');
+    assert.equal(await visible.locator('h2 a').getAttribute('href'), '/publications/' + slug + '/');
+  }
   await search.fill('no-result-xyz'); await waitForSearch(); assert.equal(await visible.count(), 0);
   assert(await page.locator('[data-publication-empty]').isVisible());
   await page.locator('[data-publication-clear]').click(); await waitForSearch(); assert.equal(await visible.count(), pageCount);
   const publicationIndex = JSON.parse(fs.readFileSync(path.join(root, 'publications/index.json'), 'utf8'));
   const researchButtons = page.getByRole('group', { name: 'Research area', exact: true }).getByRole('button');
+  assert.equal(await researchButtons.count(), 4, 'One All button and exactly three research areas');
   assert.equal(await page.locator('select[data-publication-research], [data-publication-tag]').count(), 0);
   for (const button of await researchButtons.all()) {
     await button.click();
@@ -213,13 +222,103 @@ let browser;
   await page.getByRole('link', { name: 'Browse all publications in this area' }).click();
   await waitForSearch();
   assert.equal(await visible.count(), quantumCount);
+  const aiopsRoute = '/research/rca-aiops/';
+  const previousAiopsRoute = '/research/distributed-machine-learning/';
+  const aiopsPublications = publicationIndex.filter(record => record.research.includes('rca-aiops'));
+  for (const route of ['/', '/research/']) {
+    await page.goto(base + route);
+    const themeLinks = page.locator('.card-grid-wide .card-readmore');
+    assert.equal(await themeLinks.count(), 3, 'Exactly three research areas');
+    const areaURLs = await themeLinks.evaluateAll(links => links.map(link => link.getAttribute('href')).sort());
+    assert(areaURLs.includes(aiopsRoute) && !areaURLs.includes(previousAiopsRoute), 'Research links must use the new canonical RCA URL');
+    const menuURLs = await page.locator('#site-nav .nav-submenu a[href^="/research/"]').evaluateAll(links => links.map(link => link.getAttribute('href')).sort());
+    assert.deepEqual(menuURLs, areaURLs, 'Every research area must be reachable from the header');
+    assert.deepEqual(await page.locator('.site-footer .footer-list').first().locator('a').evaluateAll(links => links.map(link => link.getAttribute('href')).sort()), areaURLs);
+  }
+  for (const [area, route] of [
+    ['quantum-computing-systems', '/research/quantum-computing-systems/'],
+    ['rca-aiops', aiopsRoute],
+    ['data-management', '/research/data-management/'],
+  ]) {
+    await page.goto(base + route);
+    const expected = publicationIndex.filter(record => record.research.includes(area)).slice(0, 6).map(record => record.url);
+    assert.deepEqual(await page.locator('#publications .list-card h3 a').evaluateAll(links => links.map(link => link.getAttribute('href'))),
+      expected, 'Every research area must use the same six-publication limit');
+  }
+  for (const width of [1440, 390, 320]) {
+    await page.setViewportSize({ width, height: 900 });
+    await page.goto(base + aiopsRoute);
+    assert.equal(await page.locator('h1').textContent(), 'Root Cause Analysis & AIOps');
+    assert.equal(new URL(await page.locator('link[rel="canonical"]').getAttribute('href'), base).pathname, aiopsRoute);
+    assert.deepEqual(await page.locator('.research-area > section .section-heading h2').allTextContents(),
+      ['Research outcomes', 'Recent publications', 'Researchers in this area']);
+    assert.deepEqual(await page.locator('#publications .list-card h3 a').evaluateAll(links => links.map(link => link.getAttribute('href'))),
+      aiopsPublications.slice(0, 6).map(record => record.url), 'The research page must show only its six most recent master references');
+    assert.equal(await page.locator('#outcomes .card').count(), 6);
+    assert.equal(await page.locator('.research-support-logos img').count(), 2);
+    assert.equal(await page.locator('.research-diagrams img').count(), 2);
+    assert(await page.locator('.page-hero').textContent().then(text => text.includes('grant ALLRP614128-25')));
+    assert(await page.getByRole('link', { name: 'View the full MSRG team', exact: true }).isVisible());
+    assert.equal(await page.evaluate(() => document.documentElement.scrollWidth > innerWidth + 1), false);
+    if (width !== 320) {
+      await page.screenshot({ path: path.join(output, 'aiops-' + width + '.png') });
+      for (const section of ['.research-diagrams', '#outcomes']) {
+        await page.locator(section).screenshot({ path: path.join(output, 'aiops-' + section.replace(/^[.#]/, '') + '-' + width + '.png'),
+          style: '.site-header { visibility: hidden !important; }' });
+      }
+    }
+  }
+  // Hugo aliases can use the production origin. Local previews use a relative baseURL.
+  const alias = fs.readFileSync(path.join(root, 'research/distributed-machine-learning/index.html'), 'utf8');
+  assert.match(alias, /http-equiv=["']?refresh/i);
+  const aliasTarget = alias.match(/content=["']?[^>]*?url=([^"' >]+)/i)?.[1];
+  assert.equal(new URL(aliasTarget, base).pathname, aiopsRoute, 'The former Distributed Machine Learning URL must redirect to the RCA page');
+  if (new URL(aliasTarget, base).origin === new URL(base).origin) {
+    for (const hash of ['', '#outcomes', '#publications', '#people']) {
+      await page.goto(base + previousAiopsRoute + hash);
+      await page.waitForURL(base + aiopsRoute + hash);
+      if (hash) {
+        await page.waitForFunction(hash => {
+          const section = document.querySelector(hash);
+          const top = section.getBoundingClientRect().top;
+          return scrollY > 0 && top >= 0 && top < innerHeight;
+        }, hash);
+      }
+    }
+    await page.goto(base + previousAiopsRoute + '?source=bookmark#people');
+    await page.waitForURL(base + aiopsRoute + '?source=bookmark#people');
+    const aliasNoJSContext = await createContext({ javaScriptEnabled: false });
+    const aliasNoJSPage = await aliasNoJSContext.newPage();
+    await aliasNoJSPage.goto(base + previousAiopsRoute);
+    await aliasNoJSPage.waitForURL(base + aiopsRoute);
+    assert.equal(await aliasNoJSPage.locator('h1').textContent(), 'Root Cause Analysis & AIOps');
+    await aliasNoJSContext.close();
+  }
+  for (const research of ['rca-aiops', 'distributed-machine-learning']) {
+    await page.goto(base + '/publications/?research=' + research); await waitForSearch();
+    assert.equal(new URL(page.url()).searchParams.get('research'), 'rca-aiops');
+    assert.deepEqual(await visible.locator('h2 a').evaluateAll(links => links.map(link => link.getAttribute('href'))),
+      aiopsPublications.map(record => record.url), 'Old and new archive links must select the same combined area');
+  }
+  await page.goto(base + '/publications/?research=distributed-machine-learning&q=GALA&page=1'); await waitForSearch();
+  assert.equal(new URL(page.url()).searchParams.get('research'), 'rca-aiops');
+  assert.equal(new URL(page.url()).searchParams.get('q'), 'GALA');
+  assert.equal(new URL(page.url()).searchParams.get('page'), '1');
+  const oldQueryMatches = await visible.locator('h2 a').allTextContents();
+  assert(oldQueryMatches.length > 0);
+  await page.goto(base + '/publications/?research=quantum-computing-systems'); await waitForSearch();
+  await page.goBack(); await waitForSearch();
+  assert.deepEqual(await visible.locator('h2 a').allTextContents(), oldQueryMatches, 'Back must restore the canonicalized legacy query');
+  await page.reload(); await waitForSearch();
+  assert.deepEqual(await visible.locator('h2 a').allTextContents(), oldQueryMatches);
+  await page.setViewportSize({ width: 390, height: 667 });
   await page.goto(base + '/people/');
   assert.equal(await page.locator('.person-card h3 a, .alumni-list li > span:first-child a').count(), 0, 'Member names must be plain text');
   // Snap at group starts; oversized groups must not create bottom-edge stops.
   const snapContext = await createContext();
   const snapPage = await snapContext.newPage();
   await snapPage.bringToFront();
-  const rosterRoutes = ['/people/', '/research/quantum-computing-systems/', '/research/distributed-machine-learning/', '/research/data-management/'];
+  const rosterRoutes = ['/people/', '/research/quantum-computing-systems/', '/research/rca-aiops/', '/research/data-management/'];
   for (const width of [1440, 390]) {
     await snapPage.setViewportSize({ width, height: 900 });
     for (const route of rosterRoutes) {
@@ -315,7 +414,7 @@ let browser;
   await page.keyboard.type('msrg', { delay: 90 });
   assert.equal(await page.locator('[data-msrg-rounds]').textContent(), '2', 'Hardware keyboard must complete a round');
 
-  for (const route of ['/', '/people/', '/publications/', '/research/quantum-computing-systems/', '/data-sets/begen/', '/~thomas-trenty/', '/data-sets/']) {
+  for (const route of ['/', '/people/', '/publications/', '/research/quantum-computing-systems/', '/research/rca-aiops/', '/data-sets/begen/', '/~thomas-trenty/', '/data-sets/']) {
     await page.goto(base + route);
     await page.addScriptTag({ path: require.resolve('axe-core/axe.min.js') });
     const results = await page.evaluate(async () => axe.run(document, { runOnly: { type: 'tag', values: ['wcag2a', 'wcag2aa', 'wcag21aa', 'best-practice'] } }));
diff --git a/scripts/content_model.py b/scripts/content_model.py
index 099ee4a..2fc5f02 100644
--- a/scripts/content_model.py
+++ b/scripts/content_model.py
@@ -120,7 +120,12 @@ def validate_fields(fields: list[dict], data: dict, kind: str, root: Path = ROOT
                 parsed = urlsplit(value)
                 schemes = {"https"} if field.get("https_only") else {"http", "https"}
                 host = parsed.hostname or ""
-                if parsed.scheme not in schemes or not host or any(c.isspace() for c in value):
+                if (parsed.scheme not in schemes or not host or
+                        any(c.isspace() or ord(c) < 32 or ord(c) == 127 for c in value) or "\\" in value):
+                    raise ValueError
+                # urlsplit leaves invalid ports and escaped hostnames unchecked.
+                parsed.port
+                if parsed.username is not None or parsed.password is not None or "%" in host:
                     raise ValueError
                 if field.get("host") and not (host == field["host"] or host.endswith("." + field["host"])):
                     raise ValueError
@@ -145,6 +150,21 @@ def validate_record(kind: str, data: dict, slug: str | None = None, root: Path =
         errors.append("slug: must match the profile filename")
     if kind == "personal" and slug is not None and data.get("url", f"/~{slug}/") != f"/~{slug}/":
         errors.append(f"url: expected /~{slug}/")
+    if kind == "publications":
+        published, preprint = data.get("external_url"), data.get("preprint_url")
+        if not any(isinstance(url, str) and url.strip() for url in (published, preprint)):
+            errors.append("external_url / preprint_url: provide a published / official URL, a preprint URL, or both")
+        if isinstance(published, str) and published:
+            try:
+                parsed = urlsplit(published)
+                host = (parsed.hostname or "").lower()
+                if (host == "arxiv.org" or host.endswith(".arxiv.org") or
+                        host in {"doi.org", "dx.doi.org"} and parsed.path.lower().startswith("/10.48550/arxiv.")):
+                    errors.append("external_url: put arXiv links in Preprint URL; leave Published / official URL blank until available")
+            except ValueError:
+                pass  # The URL field validation above reports malformed URLs.
+        if isinstance(published, str) and published and published == preprint:
+            errors.append("preprint_url: use a distinct preprint link, not the published / official URL")
     return errors
 
 
diff --git a/scripts/editor_check.cjs b/scripts/editor_check.cjs
index d3396fd..2006318 100644
--- a/scripts/editor_check.cjs
+++ b/scripts/editor_check.cjs
@@ -44,7 +44,9 @@ let browser;
   assert.equal(await page.locator('#slug').inputValue(), 'jane-doe');
   await page.getByLabel('Upload a portrait').setInputFiles({ name: 'portrait.png', mimeType: 'image/png', buffer: Buffer.from('iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+ip1sAAAAASUVORK5CYII=', 'base64') });
   await page.waitForFunction(() => document.querySelector('#message').textContent.startsWith('Uploaded '));
-  await page.getByRole('button', { name: 'Add awards and funding', exact: true }).click();
+  await page.getByRole('button', { name: 'Add awards and funding', exact: true }).focus();
+  await page.keyboard.press('Enter');
+  assert.equal(await page.getByLabel('Award name').evaluate(input => input === document.activeElement), true);
   await page.getByLabel('Award name').fill('Example award');
   await page.getByLabel('Award page *', { exact: true }).fill('https://example.org/award');
   async function addItem(key, value) {
@@ -71,15 +73,21 @@ let browser;
   assert.deepEqual(await items('interests'), ['Quantum Computing']);
   assert.equal(fs.existsSync(path.join(fixture, 'content/people/jane-doe.md')), false, 'Enter adds an item without submitting the form');
   async function save() {
-    await page.locator('#save').click();
-    await page.waitForFunction(() => document.querySelector('#message').textContent.startsWith('Saved ') || document.querySelector('#message').classList.contains('error'));
-    assert.match(await page.locator('#message').textContent(), /^Saved /);
+    await Promise.all([
+      page.waitForResponse(response => response.url() === base + '/api/records' && response.request().method() === 'POST'),
+      page.locator('#save').click(),
+    ]);
     await page.waitForFunction(() => !document.querySelector('#save').disabled);
+    assert.match(await page.locator('#message').textContent(), /^Saved /);
   }
   async function choose(kind) {
     await page.locator('#kind').selectOption(kind);
     await page.waitForFunction(expected => !document.querySelector('#save').disabled && document.querySelector('#kind').value === expected, kind);
   }
+  async function chooseEntry(slug) {
+    await page.locator('#entry-toggle').click();
+    await page.locator('#entry-options [data-value=' + JSON.stringify(slug) + ']').click();
+  }
   await save();
   console.log('Profile saved');
   assert.equal(await page.locator('#save').textContent(), 'Save changes');
@@ -110,6 +118,7 @@ let browser;
   await page.getByRole('button', { name: 'Add highlights', exact: true }).click();
   await page.getByLabel('Title *', { exact: true }).fill('Research highlight');
   await page.getByRole('button', { name: 'Add primary link', exact: true }).click();
+  assert.equal(await page.getByLabel('Label *', { exact: true }).evaluate(input => input === document.activeElement), true);
   await page.getByLabel('Label *', { exact: true }).fill('MSRG');
   await page.getByLabel('Link *', { exact: true }).fill('https://msrg.github.io/');
   await save();
@@ -174,6 +183,7 @@ let browser;
   await page.setViewportSize({ width: 1440, height: 900 });
   console.log('Template fields, hidden-field preservation, and timeline reordering passed');
   await page.getByRole('button', { name: 'Remove item', exact: true }).first().click();
+  assert.equal(await page.getByRole('button', { name: 'Add primary link', exact: true }).evaluate(button => button === document.activeElement), true);
   await save();
 
   await choose('publications');
@@ -213,7 +223,17 @@ let browser;
   await addItem('tags', 'quantum');
   await page.locator('#tags').fill('systems');
   await page.locator('#venue').fill('Example Conference');
-  await page.locator('#external_url').fill('https://example.org/paper');
+  const publishedURL = page.getByLabel('Published / official URL', { exact: true });
+  const preprintURL = page.getByLabel('Preprint URL', { exact: true });
+  assert.equal(await publishedURL.evaluate(input => input.checkValidity()), false, 'At least one paper link is required');
+  assert.match(await page.locator('#external_url-help').textContent(), /even if the paper is accepted/);
+  assert.match(await page.locator('#preprint_url-help').textContent(), /less emphasis/);
+  await publishedURL.fill('https://arxiv.org/abs/2601.01234');
+  assert.match(await publishedURL.evaluate(input => input.validationMessage), /Put arXiv links in Preprint URL/);
+  await preprintURL.fill('https://arxiv.org/abs/2601.01234');
+  await publishedURL.fill('');
+  assert.equal(await publishedURL.evaluate(input => input.checkValidity()), true, 'A preprint-only paper needs no published URL');
+  await publishedURL.fill('https://example.org/paper');
   await page.getByLabel('Abstract', { exact: true }).fill('An abstract with ');
   const mathTools = page.getByRole('group', { name: 'Formatting for abstract', exact: true });
   await mathTools.getByRole('button', { name: 'Inline formula', exact: true }).click();
@@ -222,24 +242,88 @@ let browser;
   const abstract = await page.locator('#abstract').inputValue();
   assert.match(abstract, /\$\$\n\\sum_\{i=1\}\^\{n\} x_i\n\$\$/);
   await page.getByLabel('Additional notes', { exact: true }).fill('Companion notes. [Member](/people/#member-jane-doe)');
+  const outcomeField = page.locator('#field-research_outcomes');
+  const outcomeRows = outcomeField.locator('.award');
+  const addOutcome = outcomeField.getByRole('button', { name: 'Add research outcome', exact: true });
+  let expectedOutcomes = [
+    { title: 'Public benchmark', url: 'https://example.org/benchmark', summary: 'Sample data and evaluation scripts.\nTwo public examples.' },
+    { title: 'Project website', url: 'https://example.org/project', summary: 'Project overview and leaderboard.' },
+  ];
+  async function outcomeValues() {
+    return outcomeRows.evaluateAll(rows => rows.map(row => Object.fromEntries(
+      ['title', 'url', 'summary'].map(key => [key, row.querySelector(`[id$="-${key}"]`).value]))));
+  }
+  async function addOutcomeWithKeyboard(outcome) {
+    await addOutcome.focus(); await page.keyboard.press('Enter');
+    const row = outcomeRows.last();
+    const title = row.getByLabel('Outcome name *', { exact: true });
+    const url = row.getByLabel('Project or repository URL *', { exact: true });
+    const summary = row.getByLabel('Description *', { exact: true });
+    assert.equal(await title.evaluate(input => input === document.activeElement), true, 'Add focuses the first new field');
+    assert.equal(await title.evaluate(input => input.checkValidity()), false, 'An added outcome requires a name');
+    await page.keyboard.insertText(outcome.title);
+    await page.keyboard.press('Tab');
+    assert.equal(await url.evaluate(input => input === document.activeElement), true);
+    await page.keyboard.insertText(outcome.url);
+    await page.keyboard.press('Tab');
+    assert.equal(await summary.evaluate(input => input === document.activeElement), true);
+    assert.equal(await summary.evaluate(input => input.tagName), 'TEXTAREA');
+    await page.keyboard.insertText(outcome.summary);
+    return row;
+  }
+  async function removeOutcomeWithKeyboard(row) {
+    await row.getByRole('button', { name: 'Remove item', exact: true }).focus();
+    await page.keyboard.press('Enter');
+    assert.equal(await addOutcome.evaluate(button => button === document.activeElement), true, 'Remove returns focus to Add');
+  }
+  assert.equal(await outcomeRows.count(), 0, 'Outcomes are optional and initially empty');
+  assert.equal(await addOutcome.getAttribute('aria-describedby'), 'research_outcomes-help');
+  assert.match(await page.locator('#research_outcomes-help').textContent(), /automatically.*publication.*research areas/);
+  await page.evaluate(() => {
+    window.outcomeFocusEvents = [];
+    document.addEventListener('focusin', event => {
+      if (event.target.closest('#field-research_outcomes')) window.outcomeFocusEvents.push(event.target.id || event.target.textContent);
+    });
+  });
+  await addOutcomeWithKeyboard(expectedOutcomes[0]);
+  await addOutcomeWithKeyboard({ title: 'Temporary outcome', url: 'https://example.org/temp', summary: 'Remove this before creating the paper.' });
+  await removeOutcomeWithKeyboard(outcomeRows.last());
+  assert.deepEqual(await outcomeValues(), expectedOutcomes.slice(0, 1));
+  await addOutcomeWithKeyboard(expectedOutcomes[1]);
+  const outcomeUrl = outcomeRows.first().getByLabel('Project or repository URL *', { exact: true });
+  await outcomeUrl.fill('http://example.org/benchmark');
+  await page.locator('#save').click();
+  await page.waitForFunction(() => document.querySelector('#message').classList.contains('error') && !document.querySelector('#save').disabled);
+  assert.match(await page.locator('#message').textContent(), /research_outcomes\[0\]\.url: use a valid HTTPS URL/);
+  assert.equal(fs.existsSync(path.join(fixture, 'content/publications/example-paper.md')), false);
+  await outcomeUrl.fill(expectedOutcomes[0].url);
+  await page.evaluate(() => { window.outcomeFocusEvents = []; });
   await save();
+  assert.deepEqual(await page.evaluate(() => window.outcomeFocusEvents), [], 'Rendering saved outcomes must not steal focus');
+  assert.deepEqual(await outcomeValues(), expectedOutcomes);
+  assert.deepEqual((await (await page.request.get(base + '/api/records/publications/example-paper')).json()).data.research_outcomes, expectedOutcomes);
+  assert.equal(await publishedURL.inputValue(), 'https://example.org/paper');
+  assert.equal(await preprintURL.inputValue(), 'https://arxiv.org/abs/2601.01234');
   assert.equal(await page.locator('#save').textContent(), 'Save changes');
   assert.equal(await publicationType.inputValue(), 'Journal Article');
   assert.deepEqual(await items('authors'), ['J. Doe', 'John Doe', 'Doe, Alex', 'External Author'], 'Saving includes unfinished input');
   assert.deepEqual(await items('tags'), ['quantum', 'systems']);
   assert.equal(await page.locator('#record-slug').inputValue(), 'example-paper');
-  await page.locator('#profile').selectOption('');
+  await chooseEntry('');
   await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#save').textContent === 'Create');
+  assert.equal(await outcomeRows.count(), 0, 'A new publication starts with no outcomes');
   assert.deepEqual(await items('authors'), [], 'New entry resets the list controls');
   assert.deepEqual(await page.locator('#tags-suggestions option').evaluateAll(options => options.map(option => option.value)),
     ['quantum', 'systems'], 'Newly saved tags are immediately available for the next entry');
   await page.locator('#title').fill('Example Paper');
   await addItem('authors', 'Another Author');
   await page.locator('#venue').fill('Another Conference');
-  await page.locator('#external_url').fill('https://example.org/another-paper');
+  assert.equal(await preprintURL.inputValue(), '', 'A new publication does not inherit another paper’s preprint');
+  await preprintURL.fill('https://example.org/another-preprint');
   await save();
   assert.equal(await page.locator('#record-slug').inputValue(), 'example-paper-2');
-  assert.equal(await page.locator('#profile').inputValue(), 'example-paper-2');
+  assert.equal(await outcomeRows.count(), 0, 'Publications can be created without outcomes or research areas');
+  assert.equal(await page.locator('#profile').inputValue(), 'Example Paper');
   assert.match(await page.locator('#preview').getAttribute('href'), /\/publications\/example-paper-2\/$/);
   assert.equal(await page.locator('#record-slug').evaluate(input => input.readOnly), true);
 
@@ -261,24 +345,64 @@ let browser;
   await choose('publications');
   console.log('Publication form loaded');
   assert.deepEqual(await page.locator('#tags-suggestions option').evaluateAll(options => options.map(option => option.value)), ['quantum', 'systems']);
-  await page.locator('#profile').selectOption('example-paper');
+  await page.evaluate(() => { window.outcomeFocusEvents = []; });
+  await chooseEntry('example-paper');
   await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#title').value === 'Example Paper');
+  assert.deepEqual(await page.evaluate(() => window.outcomeFocusEvents), [], 'Loading an existing publication must not focus its outcomes');
+  assert.deepEqual(await outcomeValues(), expectedOutcomes);
+  await page.setViewportSize({ width: 390, height: 900 });
+  const editedName = outcomeRows.first().getByLabel('Outcome name *', { exact: true });
+  await editedName.focus(); await page.keyboard.press('ControlOrMeta+A'); await page.keyboard.insertText('Updated public benchmark');
+  expectedOutcomes[0].title = 'Updated public benchmark';
+  await removeOutcomeWithKeyboard(outcomeRows.last());
+  expectedOutcomes = expectedOutcomes.slice(0, 1);
+  await save();
+  assert.deepEqual(await outcomeValues(), expectedOutcomes);
+  const thirdOutcome = { title: 'New evaluation tools', url: 'https://example.org/evaluation', summary: 'Evaluation tools added to an existing publication.' };
+  await addOutcomeWithKeyboard(thirdOutcome);
+  expectedOutcomes.push(thirdOutcome);
+  await save();
+  await page.locator('#reload').focus();
+  await page.evaluate(() => { window.outcomeFocusEvents = []; });
+  await page.keyboard.press('Enter');
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#message').textContent.startsWith('Edit the form'));
+  assert.deepEqual(await page.evaluate(() => window.outcomeFocusEvents), [], 'Reload must not focus the populated outcome rows');
+  assert.deepEqual(await outcomeValues(), expectedOutcomes);
+  await page.setViewportSize({ width: 1440, height: 900 });
+  console.log('Optional publication outcomes: keyboard create/edit/remove/reload, HTTPS validation, and focus preservation passed');
+  assert.equal(await preprintURL.inputValue(), 'https://arxiv.org/abs/2601.01234', 'Preprint survives reopening and outcome edits');
+  await preprintURL.fill('');
+  await save();
+  assert.equal(await publishedURL.inputValue(), 'https://example.org/paper', 'Removing a preprint preserves the published link');
+  await publishedURL.fill('');
+  assert.equal(await publishedURL.evaluate(input => input.checkValidity()), false, 'Clearing both links prevents saving');
+  await preprintURL.fill('https://arxiv.org/abs/2601.01234');
+  await save();
+  assert.equal(await publishedURL.inputValue(), '', 'A saved paper can retain only its preprint');
+  await publishedURL.fill('https://example.org/paper');
+  await save();
+  assert.equal(await preprintURL.inputValue(), 'https://arxiv.org/abs/2601.01234', 'Adding the published version preserves its preprint');
+  console.log('Publication versions: separate fields, preprint-only creation, published link addition, and removal checks passed');
   assert.equal(await publicationType.inputValue(), 'Journal Article', 'Publication type survives reopening');
   assert.equal(await page.locator('#abstract').inputValue(), abstract, 'LaTeX survives saving and reopening');
   assert.deepEqual(await items('authors'), ['J. Doe', 'John Doe', 'Doe, Alex', 'External Author']);
   assert.equal(await page.locator('#save').textContent(), 'Save changes');
   await page.getByLabel('Example Dataset', { exact: true }).check();
   await save();
+  assert.deepEqual(await outcomeValues(), expectedOutcomes, 'Unrelated changes preserve research outcomes');
   const publication = path.join(fixture, 'content/publications/example-paper.md');
   fs.appendFileSync(publication, '\nChanged outside the editor.\n');
+  const externalPublication = fs.readFileSync(publication, 'utf8');
+  await removeOutcomeWithKeyboard(outcomeRows.last());
   await page.locator('#title').fill('Changed title');
   await page.locator('#save').click();
   await page.waitForFunction(() => document.querySelector('#message').classList.contains('error'));
   assert.match(await page.locator('#message').textContent(), /changed since you opened/);
-  assert.match(fs.readFileSync(publication, 'utf8'), /title = "Example Paper"/);
+  assert.equal(fs.readFileSync(publication, 'utf8'), externalPublication, 'A stale removal must not discard saved outcomes');
   page.once('dialog', dialog => dialog.accept());
   await page.getByRole('button', { name: 'Reload saved entry' }).click();
   await page.waitForFunction(() => !document.querySelector('#save').disabled && !document.querySelector('#message').classList.contains('error'));
+  assert.deepEqual(await outcomeValues(), expectedOutcomes, 'Reload restores the saved outcomes after a conflict');
   const originalPreview = await page.locator('#preview').getAttribute('href');
   await page.locator('#title').fill('Revised Paper');
   await save();
@@ -287,7 +411,7 @@ let browser;
   assert.match(fs.readFileSync(publication, 'utf8'), /aliases = \["\/publications\/revised-paper\/"\]/);
 
   await choose('people');
-  await page.locator('#profile').selectOption('jane-doe');
+  await chooseEntry('jane-doe');
   await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#slug').value === 'jane-doe');
   assert.equal(await page.locator('#personal-page-status').textContent(), 'Enabled — template page');
   assert.equal(await page.locator('#personal-page-status').evaluate(element => element.tagName), 'OUTPUT');
@@ -313,7 +437,7 @@ let browser;
       await choose(kind);
       if (kind !== 'personal') {
         const slug = { people: 'jane-smith', publications: 'example-paper', 'data-sets': 'example-dataset' }[kind];
-        await page.locator('#profile').selectOption(slug);
+        await chooseEntry(slug);
         await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#save').textContent === 'Save changes');
       }
       assert.equal(await page.evaluate(() => document.documentElement.scrollWidth > innerWidth + 1), false, `Overflow at ${width}px: ${kind}`);
@@ -336,8 +460,204 @@ let browser;
   await page.waitForFunction(() => document.querySelector('#preview').getAttribute('aria-disabled') === 'true');
   assert.equal(await page.locator('#preview').getAttribute('href'), null);
   assert.equal(await page.locator('#preview-start').isVisible(), true);
+
+  // Native select popups can expand to their longest option. Exercise the
+  // replacement with a large collection and a single option taller than it.
+  const longTitle = 'Extremely long publication title: ' + 'Full title text must stay readable. '.repeat(30) + 'Unbroken'.repeat(60);
+  function pickerFixture(slug, title) {
+    fs.writeFileSync(path.join(fixture, 'content/publications', slug + '.md'),
+      '+++\ntitle = ' + JSON.stringify(title) + '\nyear = 2026\nauthors = ["Example Author"]\nvenue = "Example"\npublication_type = "Conference Paper"\nexternal_url = "https://example.org/paper"\n+++\n');
+  }
+  pickerFixture('picker-long', longTitle);
+  pickerFixture('picker-duplicate-a', 'Repeated publication title');
+  pickerFixture('picker-duplicate-b', 'Repeated publication title');
+  for (let i = 0; i < 80; i++) pickerFixture('picker-' + i, 'Picker sample ' + String(i).padStart(3, '0'));
+  await choose('publications');
+  const entry = page.locator('#profile'), entryMenu = page.locator('#entry-menu');
+  const entryOptions = page.locator('#entry-options [role="option"]');
+  const menuScreenshots = fs.mkdtempSync(path.join(os.tmpdir(), `msrg-entry-${engine}-`));
+  for (const [width, height] of [[1440, 900], [800, 600], [390, 844], [320, 568]]) {
+    await page.setViewportSize({ width, height });
+    await entry.scrollIntoViewIfNeeded();
+    await entry.fill('Extremely long');
+    assert.equal(await entryOptions.count(), 1);
+    assert.equal(await entryOptions.first().textContent(), longTitle, 'Retain the full option label');
+    assert.equal(await entryOptions.first().getAttribute('aria-selected'), 'true', 'Filtered active suggestion is selected for assistive technology');
+    const bounds = await page.evaluate(() => {
+      const menu = document.querySelector('#entry-menu'), list = document.querySelector('#entry-options'), control = document.querySelector('.entry-control');
+      const rect = menu.getBoundingClientRect(), field = control.getBoundingClientRect();
+      const row = menu.querySelector('[role=option]').getBoundingClientRect();
+      return { left: rect.left, right: rect.right, top: rect.top, bottom: rect.bottom,
+        width: rect.width, fieldWidth: field.width, height: rect.height, rowHeight: row.height,
+        rowTop: row.top, scrolling: list.scrollHeight > list.clientHeight,
+        overflow: document.documentElement.scrollWidth > innerWidth + 1,
+        menuOverflow: list.scrollWidth > list.clientWidth + 1 };
+    });
+    assert(bounds.left >= 0 && bounds.right <= width + 1 && bounds.top >= 0 && bounds.bottom <= height + 1, 'Menu must stay within viewport');
+    assert(bounds.width <= bounds.fieldWidth + 1 && bounds.height <= Math.min(320, height * .45) + 1);
+    assert(bounds.scrolling && bounds.rowHeight > bounds.height, 'Exercise an option taller than the menu');
+    assert(Math.abs(bounds.rowTop - bounds.top) < 8, 'A tall option initially shows the start of its title');
+    assert.equal(bounds.overflow || bounds.menuOverflow, false);
+    const beforeScroll = await page.evaluate(() => scrollY);
+    await entry.press('PageDown');
+    assert(await page.locator('#entry-options').evaluate(list => list.scrollTop > 50), 'Keyboard users can read the rest of a tall option');
+    assert.equal(await page.evaluate(() => scrollY), beforeScroll);
+    await entry.press('PageUp');
+    assert(await page.locator('#entry-options').evaluate(list => list.scrollTop < 10));
+    const menuBox = await entryMenu.boundingBox();
+    await page.mouse.move(menuBox.x + 20, menuBox.y + menuBox.height / 2);
+    await page.mouse.wheel(0, 150);
+    await page.waitForFunction(() => document.querySelector('#entry-options').scrollTop > 50);
+    assert.equal(await page.evaluate(() => scrollY), beforeScroll, 'Wheel scrolling stays inside the menu');
+    await entry.press('Escape');
+    assert.equal(await entry.inputValue(), 'New entry');
+    assert.equal(await page.locator('#title').inputValue(), '');
+    await page.locator('#entry-toggle').click();
+    assert.equal(await entryOptions.count(), 86);
+    await page.screenshot({ path: path.join(menuScreenshots, `entries-${width}.png`) });
+    await page.addScriptTag({ path: require.resolve('axe-core/axe.min.js') });
+    assert.deepEqual(await page.evaluate(async () => (await axe.run()).violations.map(v => v.id)), [], 'Open entry menu accessibility');
+    await entry.press('Escape');
+  }
+  await page.setViewportSize({ width: 1440, height: 900 });
+  await page.evaluate(() => scrollTo(0, 0));
+  await entry.press('ArrowDown');
+  assert.equal(await entryMenu.isVisible(), true);
+  await page.mouse.move(20, 250); await page.mouse.wheel(0, 500);
+  await page.waitForFunction(() => document.querySelector('#entry-menu').hidden);
+  assert.equal(await entry.getAttribute('aria-expanded'), 'false', 'Dismiss when document scrolling hides the control');
+  assert.equal(await entry.getAttribute('aria-activedescendant'), null);
+  assert.equal(await entry.inputValue(), 'New entry');
+  assert.equal(await page.locator('#record-slug').inputValue(), '');
+  await page.setViewportSize({ width: 390, height: 844 });
+  await page.evaluate(() => scrollTo(0, 0));
+  await entry.press('ArrowDown');
+  assert.equal(await entryMenu.isVisible(), true);
+  await page.setViewportSize({ width: 390, height: 360 });
+  await page.evaluate(() => scrollTo(0, 0));
+  await page.waitForFunction(() => document.querySelector('#entry-menu').hidden);
+  assert.equal(await entry.getAttribute('aria-expanded'), 'false', 'Dismiss when resizing hides the control');
+  assert.equal(await entry.getAttribute('aria-activedescendant'), null);
+  assert.equal(await entry.inputValue(), 'New entry');
+  assert.equal(await page.locator('#record-slug').inputValue(), '');
+  await page.setViewportSize({ width: 1440, height: 900 });
+  await page.evaluate(() => scrollTo(0, 0));
+  await entry.fill('No matching entry xyz');
+  assert.equal(await entryOptions.count(), 0);
+  assert.equal(await page.locator('#entry-empty').isVisible(), true);
+  assert.equal(await entry.getAttribute('aria-activedescendant'), null);
+  await entry.press('Escape');
+  assert.equal(await entry.inputValue(), 'New entry');
+  await entry.press('ArrowDown');
+  await page.mouse.move(0, 0);
+  const beforeKeys = await page.evaluate(() => scrollY);
+  for (let i = 0; i < 45; i++) await entry.press('ArrowDown');
+  assert.equal(await page.locator('#entry-options [aria-selected="true"]').count(), 1);
+  assert.equal(await page.locator('#' + await entry.getAttribute('aria-activedescendant')).getAttribute('aria-selected'), 'true');
+  assert.equal(await page.evaluate(() => scrollY), beforeKeys, 'Keyboard navigation scrolls only the menu');
+  assert.equal(await page.locator('#record-slug').inputValue(), '', 'Highlighting does not select a record');
+  await entry.press('Escape');
+  await entry.fill('Picker sample 037');
+  await entry.press('Enter');
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'picker-37');
+  assert.equal(await entry.evaluate(input => input === document.activeElement), true, 'Focus returns after loading the chosen record');
+  assert.equal(await entryMenu.isVisible(), false);
+  await entry.fill('Uncommitted search');
+  await entry.press('Tab');
+  assert.equal(await entry.inputValue(), 'Picker sample 037');
+  assert.equal(await entryMenu.isVisible(), false);
+  await entry.fill('Picker sample 038');
+  await page.locator('h1').click();
+  assert.equal(await entry.inputValue(), 'Picker sample 037', 'Outside clicks dismiss without changing the record');
+  assert.equal(await page.locator('#record-slug').inputValue(), 'picker-37');
+  await entry.fill('Repeated publication title');
+  assert.equal(await entryOptions.count(), 2);
+  await page.locator('#entry-options [data-value="picker-duplicate-b"]').click();
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'picker-duplicate-b');
+  assert.equal(await entry.inputValue(), 'Repeated publication title');
+
+  await chooseEntry('example-paper');
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'example-paper');
+  const unsavedOutcome = page.locator('#field-research_outcomes input[id$="-title"]').first();
+  await unsavedOutcome.fill('Keep this unsaved outcome');
+  await entry.fill('Picker sample 037');
+  let discardPrompted = false;
+  page.once('dialog', dialog => { discardPrompted = true; dialog.dismiss(); });
+  await entry.press('Enter');
+  assert.equal(discardPrompted, true, 'Selecting another entry must prompt before discarding edits');
+  assert.equal(await entry.inputValue(), 'Revised Paper');
+  assert.equal(await page.locator('#record-slug').inputValue(), 'example-paper');
+  assert.equal(await unsavedOutcome.inputValue(), 'Keep this unsaved outcome');
+  await entry.fill('Picker sample 037');
+  discardPrompted = false;
+  page.once('dialog', dialog => { discardPrompted = true; dialog.accept(); });
+  await entry.press('Enter');
+  assert.equal(discardPrompted, true);
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'picker-37');
+
+  const failedRecordURL = base + '/api/records/publications/picker-38';
+  await page.route(failedRecordURL, route => route.fulfill({ status: 500, json: { error: 'Could not load test entry' } }));
+  await entry.fill('Picker sample 038'); await entry.press('Enter');
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#message').classList.contains('error'));
+  assert.equal(await entry.inputValue(), 'Picker sample 037', 'A failed load restores the current selection');
+  assert.equal(await page.locator('#record-slug').inputValue(), 'picker-37');
+  await page.unroute(failedRecordURL);
+  let releaseLoad;
+  const pausedLoad = new Promise(resolve => { releaseLoad = resolve; });
+  const delayedRecordURL = base + '/api/records/publications/picker-39';
+  await page.route(delayedRecordURL, async route => { await pausedLoad; await route.continue(); });
+  await entry.fill('Picker sample 039'); await entry.press('Enter');
+  try {
+    await page.waitForFunction(() => document.querySelector('#profile').disabled);
+    assert.equal(await page.locator('#entry-toggle').isDisabled(), true);
+    assert.equal(await entryMenu.isVisible(), false);
+  } finally { releaseLoad(); }
+  await page.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'picker-39');
+  await page.unroute(delayedRecordURL);
+  if (engine === 'chromium') {
+    // Use the real visual viewport at mobile pinch zoom, not a mocked object.
+    const mobile = await browser.newContext({ viewport: { width: 390, height: 844 }, isMobile: true, hasTouch: true });
+    try {
+      const phone = await mobile.newPage();
+      phone.on('pageerror', error => errors.push(error.message));
+      await phone.route(base + '/api/preview', route => route.fulfill({ json: previewState }));
+      await phone.goto(base);
+      await phone.waitForFunction(() => !document.querySelector('#save').disabled);
+      await phone.locator('#kind').selectOption('publications');
+      await phone.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#title'));
+      const input = phone.locator('#profile');
+      await input.fill('Picker sample');
+      const cdp = await mobile.newCDPSession(phone);
+      await cdp.send('Emulation.setPageScaleFactor', { pageScaleFactor: 2 });
+      await phone.waitForFunction(() => visualViewport.scale === 2);
+      await phone.evaluate(() => scrollTo(0, document.querySelector('.entry-control').getBoundingClientRect().top + scrollY - 50));
+      await input.fill('Picker sample 037');
+      assert.equal(await input.inputValue(), 'Picker sample 037', 'Zoom must not discard the typed query');
+      assert.equal(await phone.locator('#entry-menu').isVisible(), true);
+      const zoomBounds = await phone.evaluate(() => {
+        const menu = document.querySelector('#entry-menu').getBoundingClientRect();
+        const control = document.querySelector('.entry-control').getBoundingClientRect();
+        const viewport = visualViewport;
+        return { left: menu.left, right: menu.right, top: menu.top, bottom: menu.bottom,
+          visibleLeft: viewport.offsetLeft, visibleTop: viewport.offsetTop,
+          visibleRight: viewport.offsetLeft + viewport.width, visibleBottom: viewport.offsetTop + viewport.height,
+          controlWidth: control.width, visibleWidth: viewport.width };
+      });
+      assert(zoomBounds.controlWidth > zoomBounds.visibleWidth, 'Exercise a partly visible control at 2× zoom');
+      assert(zoomBounds.left >= zoomBounds.visibleLeft && zoomBounds.right <= zoomBounds.visibleRight + 1 &&
+        zoomBounds.top >= zoomBounds.visibleTop && zoomBounds.bottom <= zoomBounds.visibleBottom + 1, 'Zoomed menu stays inside the visual viewport');
+      assert.equal(await phone.locator('#record-slug').inputValue(), '', 'Zoomed search does not commit by itself');
+      await phone.screenshot({ path: path.join(menuScreenshots, 'entries-mobile-2x.png') });
+      await input.press('Enter');
+      await phone.waitForFunction(() => !document.querySelector('#save').disabled && document.querySelector('#record-slug').value === 'picker-37');
+      assert.equal(await input.inputValue(), 'Picker sample 037');
+      assert.equal(await phone.locator('#title').inputValue(), 'Picker sample 037');
+      console.log('chromium: mobile 2× zoom preserves search, bounds the popup, and selects the intended record');
+    } finally { await mobile.close(); }
+  }
+  console.log(`${engine}: bounded long-title entry picker, search, keyboard, dismissal, duplicate titles, dirty cancellation, and failed/busy loads passed. Screenshots: ${menuScreenshots}`);
   assert.deepEqual(errors, []);
-  console.log(`${engine}: profile photo, awards, personal templates, publications, dataset uploads, conflicts, accessibility and desktop/mobile layouts passed.`);
+  console.log(`${engine}: profile photo, awards, personal templates, publications and outcomes, dataset uploads, conflicts, accessibility and desktop/mobile layouts passed.`);
 })().catch(error => { console.error(error); process.exitCode = 1; }).finally(async () => {
   if (browser) await browser.close();
   if (server.exitCode === null) { const stopped = new Promise(resolve => server.once('exit', resolve)); server.kill(); await stopped; }
diff --git a/scripts/profile_editor/editor.css b/scripts/profile_editor/editor.css
index a6561f0..edfa669 100644
--- a/scripts/profile_editor/editor.css
+++ b/scripts/profile_editor/editor.css
@@ -8,7 +8,18 @@ h2, legend { font-size: 1.2rem; font-weight: 700; }
 .eyebrow { font-size: .8rem; text-transform: uppercase; letter-spacing: .1em; }
 main { padding: 1.5rem 0 3rem; }
 .picker, .actions { display: flex; gap: 1rem; align-items: end; flex-wrap: wrap; }
+.picker { align-items: start; }
 .picker > div { flex: 1; min-width: min(16rem, 100%); }
+.entry-picker { position: relative; }
+.entry-control { display: flex; }
+.entry-control input { min-width: 0; border-top-right-radius: 0; border-bottom-right-radius: 0; }
+.entry-control button { flex: 0 0 44px; padding: .6rem; border-color: #8a98ac; border-left: 0; border-top-left-radius: 0; border-bottom-left-radius: 0; }
+.entry-menu { position: absolute; left: 0; right: 0; z-index: 10; display: flex; flex-direction: column; max-height: min(20rem, 45vh); overflow: hidden; background: white; border: 1px solid #8a98ac; border-radius: .4rem; box-shadow: 0 .4rem 1.2rem #14284b26; }
+#entry-options { list-style: none; margin: 0; padding: .25rem; min-height: 0; overflow: auto; overscroll-behavior: contain; }
+#entry-options [role=option] { min-height: 44px; margin: 0; padding: .6rem; overflow-wrap: anywhere; white-space: normal; border-radius: .25rem; cursor: pointer; }
+#entry-options [aria-selected=true] { font-weight: 700; }
+#entry-options .active { background: #eaf0f9; outline: 2px solid #3274c9; outline-offset: -2px; }
+#entry-empty { margin: 0; padding: .75rem; }
 label { display: block; font-weight: 600; margin-bottom: .3rem; }
 input, select, textarea, button { font: inherit; }
 input:not([type=checkbox]), select, textarea { width: 100%; border: 1px solid #8a98ac; border-radius: .4rem; padding: .6rem; background: white; color: #14284b; min-height: 44px; }
diff --git a/scripts/profile_editor/editor.js b/scripts/profile_editor/editor.js
index 68fa0dd..7c931f5 100644
--- a/scripts/profile_editor/editor.js
+++ b/scripts/profile_editor/editor.js
@@ -1,6 +1,9 @@
 'use strict';
 const form = document.querySelector('#editor');
-const picker = document.querySelector('#profile');
+const picker = createEntryPicker(async value => {
+  if (canLeave()) await run(() => loadRecord(value));
+  else picker.value = current.slug;
+});
 const kindPicker = document.querySelector('#kind');
 const message = document.querySelector('#message');
 const saveButton = document.querySelector('#save');
@@ -19,6 +22,145 @@ function node(tag, text, className) {
   if (className) element.className = className;
   return element;
 }
+function createEntryPicker(select) {
+  const input = document.querySelector('#profile');
+  const toggle = document.querySelector('#entry-toggle');
+  const wrapper = input.closest('.entry-picker');
+  const control = wrapper.querySelector('.entry-control');
+  const menu = document.querySelector('#entry-menu');
+  const list = document.querySelector('#entry-options');
+  const empty = document.querySelector('#entry-empty');
+  let options = [], matches = [], value = '', active = -1;
+  const normalize = text => text.normalize('NFKC').toLocaleLowerCase().trim();
+  function resetText() {
+    input.value = options.find(option => option.value === value)?.label || '';
+    input.title = input.value;
+  }
+  function close() {
+    menu.hidden = true;
+    input.setAttribute('aria-expanded', 'false'); toggle.setAttribute('aria-expanded', 'false');
+    toggle.setAttribute('aria-label', 'Show entries');
+    input.removeAttribute('aria-activedescendant');
+    resetText();
+  }
+  function position() {
+    if (menu.hidden) return;
+    const rect = control.getBoundingClientRect();
+    const parent = wrapper.getBoundingClientRect();
+    const viewport = window.visualViewport;
+    const top = viewport?.offsetTop || 0;
+    const height = viewport?.height || window.innerHeight;
+    const left = viewport?.offsetLeft || 0;
+    const width = viewport?.width || window.innerWidth;
+    if (rect.bottom <= top || rect.top >= top + height || rect.right <= left || rect.left >= left + width) {
+      close(); return;
+    }
+    // Pinch zoom can make the visible viewport narrower than the control.
+    // Keep the popup inside it while the control is at least partly visible.
+    const menuWidth = Math.min(rect.width, Math.max(0, width - 16));
+    const menuLeft = Math.max(left + 8, Math.min(rect.left, left + width - 8 - menuWidth));
+    menu.style.left = menuLeft - parent.left + 'px';
+    menu.style.right = 'auto';
+    menu.style.width = menuWidth + 'px';
+    const below = top + height - rect.bottom - 8;
+    const above = rect.top - top - 8;
+    const limit = Math.min(320, height * .45);
+    const upward = below < limit && above > below;
+    menu.style.maxHeight = Math.max(0, Math.min(limit, upward ? above : below)) + 'px';
+    menu.style.top = upward ? 'auto' : rect.bottom - parent.top + 4 + 'px';
+    menu.style.bottom = upward ? parent.bottom - rect.top + 4 + 'px' : 'auto';
+  }
+  function activate(index, scroll = true) {
+    active = index;
+    for (const [i, row] of Array.from(list.children).entries()) {
+      row.classList.toggle('active', i === active);
+      row.setAttribute('aria-selected', String(i === active));
+    }
+    const row = list.children[active];
+    if (!row) { input.removeAttribute('aria-activedescendant'); return; }
+    input.setAttribute('aria-activedescendant', row.id);
+    if (!scroll) return;
+    const bounds = list.getBoundingClientRect(), item = row.getBoundingClientRect();
+    // Scroll the list only; selecting an option must not jump the whole page.
+    if (item.height > bounds.height || item.top < bounds.top) list.scrollTop += item.top - bounds.top;
+    else if (item.bottom > bounds.bottom) list.scrollTop += item.bottom - bounds.bottom;
+  }
+  async function choose(index) {
+    const option = matches[index];
+    if (!option || input.disabled) return;
+    close(); input.focus();
+    if (option.value !== value) await select(option.value);
+    if (document.activeElement === document.body) input.focus();
+  }
+  function open(query = '') {
+    if (input.disabled) return;
+    const words = normalize(query).split(/\s+/).filter(Boolean);
+    matches = options.filter(option => words.every(word => normalize(option.label).includes(word)));
+    list.replaceChildren();
+    matches.forEach((option, index) => {
+      const row = node('li', option.label); row.id = 'entry-option-' + index;
+      row.setAttribute('role', 'option'); row.setAttribute('aria-selected', 'false');
+      row.dataset.value = option.value;
+      row.addEventListener('pointerdown', event => { if (event.button === 0) event.preventDefault(); });
+      row.addEventListener('click', () => choose(index));
+      row.addEventListener('pointerenter', () => activate(index, false));
+      list.append(row);
+    });
+    empty.hidden = matches.length !== 0;
+    menu.hidden = false; list.scrollTop = 0;
+    input.setAttribute('aria-expanded', 'true'); toggle.setAttribute('aria-expanded', 'true');
+    toggle.setAttribute('aria-label', 'Hide entries');
+    position();
+    if (menu.hidden) return;
+    const selected = query ? 0 : matches.findIndex(option => option.value === value);
+    activate(matches.length ? Math.max(0, selected) : -1);
+  }
+  input.addEventListener('focus', () => input.select());
+  input.addEventListener('click', () => { if (menu.hidden) { open(); input.select(); } });
+  input.addEventListener('input', () => {
+    input.title = '';
+    // Focus scrolling may finish after input in Firefox; anchor typed results
+    // before positioning them, especially when returning from a long form.
+    control.scrollIntoView({ block: 'nearest', inline: 'nearest' });
+    open(input.value);
+  });
+  input.addEventListener('keydown', event => {
+    if (event.isComposing) return;
+    if (event.key === 'Escape') {
+      if (!menu.hidden) { event.preventDefault(); event.stopPropagation(); }
+      close();
+    } else if (event.key === 'Tab') close();
+    else if (event.key === 'ArrowDown' || event.key === 'ArrowUp') {
+      event.preventDefault();
+      if (event.altKey && event.key === 'ArrowUp') close();
+      else if (menu.hidden) open();
+      else if (matches.length) activate(Math.max(0, Math.min(matches.length - 1, active + (event.key === 'ArrowDown' ? 1 : -1))));
+    } else if ((event.key === 'PageDown' || event.key === 'PageUp') && !menu.hidden) {
+      event.preventDefault();
+      list.scrollTop += list.clientHeight * .8 * (event.key === 'PageDown' ? 1 : -1);
+      const top = list.getBoundingClientRect().top;
+      activate(Array.from(list.children).findIndex(row => row.getBoundingClientRect().bottom > top + 4), false);
+    } else if (event.key === 'Enter') {
+      event.preventDefault();
+      if (menu.hidden) open(); else choose(active);
+    }
+  });
+  toggle.addEventListener('pointerdown', event => { if (event.button === 0) event.preventDefault(); });
+  toggle.addEventListener('click', () => { input.focus(); if (menu.hidden) open(); else close(); });
+  wrapper.addEventListener('focusout', event => { if (!wrapper.contains(event.relatedTarget)) close(); });
+  document.addEventListener('pointerdown', event => { if (!wrapper.contains(event.target)) close(); });
+  window.addEventListener('resize', position);
+  window.addEventListener('scroll', position, { capture: true, passive: true });
+  window.visualViewport?.addEventListener('resize', position);
+  window.visualViewport?.addEventListener('scroll', position);
+  return {
+    get value() { return value; },
+    set value(next) { value = next; close(); },
+    get options() { return options; },
+    setOptions(next) { options = next; close(); },
+    set disabled(disabled) { input.disabled = disabled; toggle.disabled = disabled; if (disabled) close(); },
+  };
+}
 function notify(text, error = false) {
   message.textContent = text;
   message.classList.toggle('error', error);
@@ -149,7 +291,8 @@ function fieldControl(field, value, prefix = '') {
   if (field.type === 'objects' || field.type === 'object') {
     label.removeAttribute('for');
     const rows = node('div');
-    const add = node('button', 'Add ' + field.label.toLowerCase(), 'secondary'); add.type = 'button';
+    const add = node('button', field.add_label || 'Add ' + field.label.toLowerCase(), 'secondary'); add.type = 'button';
+    if (helpText) add.setAttribute('aria-describedby', help.id);
     function updateOrderControls() {
       Array.from(rows.children).forEach((row, index) => {
         if (!field.reorderable) return;
@@ -182,12 +325,18 @@ function fieldControl(field, value, prefix = '') {
         }
       }
       const remove = node('button', 'Remove item', 'secondary'); remove.type = 'button';
-      remove.addEventListener('click', () => { row.remove(); add.hidden = false; updateOrderControls(); markDirty(); });
+      remove.addEventListener('click', () => {
+        row.remove(); add.hidden = false; updateOrderControls(); markDirty(); add.focus();
+      });
       actions.append(remove); row.append(actions); rows.append(row); updateOrderControls();
       if (field.type === 'object') add.hidden = true;
+      return row;
     }
     (field.type === 'object' ? (value && Object.keys(value).length ? [value] : []) : value || []).forEach(addRow);
-    add.addEventListener('click', () => { addRow({}); markDirty(); });
+    add.addEventListener('click', () => {
+      const row = addRow({}); markDirty();
+      row.querySelector('input:not([type="hidden"]):enabled, select:enabled, textarea:enabled, button:enabled:not([hidden])')?.focus();
+    });
     wrapper.append(rows, add);
     read = () => field.type === 'object' ? (rows.firstElementChild?.readValue() || {}) : Array.from(rows.children, row => row.readValue());
   } else if (field.type === 'strings' && field.choices) {
@@ -381,6 +530,27 @@ function render() {
     document.querySelector('#layout').addEventListener('change', updateTemplateFields);
     updateTemplateFields();
   }
+  if (kind === 'publications') {
+    const published = document.querySelector('#external_url');
+    const preprint = document.querySelector('#preprint_url');
+    function validatePaperLinks() {
+      let error = !published.value.trim() && !preprint.value.trim()
+        ? 'Provide a published / official URL, a preprint URL, or both.' : '';
+      try {
+        const url = new URL(published.value);
+        if (url.hostname === 'arxiv.org' || url.hostname.endsWith('.arxiv.org') ||
+            (['doi.org', 'dx.doi.org'].includes(url.hostname) && url.pathname.toLowerCase().startsWith('/10.48550/arxiv.'))) {
+          error = 'Put arXiv links in Preprint URL. Leave Published / official URL blank until available.';
+        }
+      } catch { /* The URL input checks malformed links. */ }
+      published.setCustomValidity(error);
+      preprint.setCustomValidity(published.value && preprint.value === published.value
+        ? 'Use a distinct preprint link, not the published / official URL.' : '');
+    }
+    published.addEventListener('input', validatePaperLinks);
+    preprint.addEventListener('input', validatePaperLinks);
+    validatePaperLinks();
+  }
   if (kind === 'people' || kind === 'publications' || kind === 'data-sets') {
     const slug = kind === 'people' ? document.querySelector('#slug') : document.querySelector('#record-slug');
     slug.readOnly = kind === 'publications' || (kind !== 'people' && Boolean(current.slug));
@@ -409,13 +579,11 @@ function render() {
 }
 async function refreshList() {
   const records = await api('/api/records/' + kind);
-  picker.replaceChildren();
   if (kind === 'personal') {
     const members = await api('/api/profiles');
-    members.forEach(member => picker.add(new Option(member.name + (records.some(record => record.slug === member.slug) ? '' : ' — create page'), member.slug)));
+    picker.setOptions(members.map(member => ({ label: member.name + (records.some(record => record.slug === member.slug) ? '' : ' — create page'), value: member.slug })));
   } else {
-    picker.add(new Option('New entry', ''));
-    records.forEach(record => picker.add(new Option(record.name, record.slug)));
+    picker.setOptions([{ label: 'New entry', value: '' }, ...records.map(record => ({ label: record.name, value: record.slug }))]);
   }
   picker.value = current.slug;
 }
@@ -446,7 +614,6 @@ async function run(action) {
   try { await action(); } catch (error) { notify(error.message, true); }
   finally { setBusy(false); kindPicker.value = kind; picker.value = current.slug; }
 }
-picker.addEventListener('change', () => { if (canLeave()) run(() => loadRecord(picker.value)); else picker.value = current.slug; });
 kindPicker.addEventListener('change', () => { if (canLeave()) run(() => changeKind(kindPicker.value)); else kindPicker.value = kind; });
 document.querySelector('#reload').addEventListener('click', () => { if (canLeave()) run(() => loadRecord(current.slug)); });
 document.querySelector('#preview-start').addEventListener('click', async event => {
diff --git a/scripts/profile_editor/index.html b/scripts/profile_editor/index.html
index c941acb..2f47f1d 100644
--- a/scripts/profile_editor/index.html
+++ b/scripts/profile_editor/index.html
@@ -13,7 +13,18 @@
   <main class="shell">
     <section class="picker" aria-label="Choose content">
       <div><label for="kind">What would you like to edit?</label><select id="kind"></select></div>
-      <div><label for="profile">Entry</label><select id="profile"></select></div>
+      <div class="entry-picker">
+        <label for="profile">Entry</label>
+        <div class="entry-control">
+          <input id="profile" role="combobox" aria-autocomplete="list" aria-expanded="false" aria-controls="entry-options" aria-describedby="entry-help" autocomplete="off" spellcheck="false" placeholder="Search entries…" disabled>
+          <button id="entry-toggle" type="button" class="secondary" aria-label="Show entries" aria-controls="entry-options" aria-expanded="false" tabindex="-1" disabled><span aria-hidden="true">▾</span></button>
+        </div>
+        <p id="entry-help" class="help">Type to search. Use the arrow keys and Enter to choose.</p>
+        <div id="entry-menu" class="entry-menu" hidden>
+          <ul id="entry-options" role="listbox" aria-label="Entries"></ul>
+          <p id="entry-empty" role="status" hidden>No matching entries.</p>
+        </div>
+      </div>
     </section>
     <p class="note">Changes save locally. They appear online after your pull request is reviewed and merged. Fields marked * are required.</p>
     <p id="message" role="status" aria-live="polite"></p>
diff --git a/static/images/people/ruiwu-liu.jpg b/static/images/people/ruiwu-liu.jpg
new file mode 100644
index 0000000..7a83232
Binary files /dev/null and b/static/images/people/ruiwu-liu.jpg differ
diff --git a/static/images/research/aiops-technical.svg b/static/images/research/aiops-technical.svg
new file mode 100644
index 0000000..50bc2d3
--- /dev/null
+++ b/static/images/research/aiops-technical.svg
@@ -0,0 +1,30 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 400 616" role="img" aria-labelledby="title desc">
+<title id="title">GALA+: technical workflow</title><desc id="desc">Telemetry and service dependencies seed hypotheses. Graph-guided LLM investigation gathers local evidence and refines the diagnosis. Outputs are assessed with SURE-Score.</desc>
+<defs><marker id="arrow" markerWidth="8" markerHeight="8" refX="7" refY="4" orient="auto"><path d="M0 0L8 4L0 8Z" fill="#426875"/></marker></defs>
+<rect width="100%" height="100%" rx="16" fill="#f5f8fa"/>
+<text x="200" y="32" text-anchor="middle" font-family="Arial,sans-serif" font-size="21" font-weight="700" fill="#1e3765">GALA+: technical workflow</text>
+<rect x="20" y="52" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="79" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Operational evidence</text>
+<text x="200" y="103" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Metrics, logs, and traces</text>
+<text x="200" y="126" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">plus service dependencies</text>
+<path d="M200 141V162" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="164" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="191" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Initial hypotheses</text>
+<text x="200" y="215" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Combine telemetry signals</text>
+<text x="200" y="238" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">with STRIX structural scoring</text>
+<path d="M200 253V274" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="276" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="303" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Graph-guided investigation</text>
+<text x="200" y="327" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Gather local evidence</text>
+<text x="200" y="350" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Refine hypotheses with LLM agents</text>
+<path d="M200 365V386" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="388" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="415" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Incident response outputs</text>
+<text x="200" y="439" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Ranked causes and summaries</text>
+<text x="200" y="462" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Stratified action recommendations</text>
+<path d="M200 477V498" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="500" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="527" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Quality assessment</text>
+<text x="200" y="551" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">SURE-Score and human review</text>
+<text x="200" y="574" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Beyond text similarity</text>
+</svg>
diff --git a/static/images/research/aiops-vision.svg b/static/images/research/aiops-vision.svg
new file mode 100644
index 0000000..ef428b8
--- /dev/null
+++ b/static/images/research/aiops-vision.svg
@@ -0,0 +1,30 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 400 616" role="img" aria-labelledby="title desc">
+<title id="title">AIOps research vision</title><desc id="desc">Observe distributed services, detect anomalies, diagnose causes, support response, and evaluate with realistic benchmarks. This is a research agenda, not a single deployed system.</desc>
+<defs><marker id="arrow" markerWidth="8" markerHeight="8" refX="7" refY="4" orient="auto"><path d="M0 0L8 4L0 8Z" fill="#426875"/></marker></defs>
+<rect width="100%" height="100%" rx="16" fill="#f5f8fa"/>
+<text x="200" y="32" text-anchor="middle" font-family="Arial,sans-serif" font-size="21" font-weight="700" fill="#1e3765">AIOps research vision</text>
+<rect x="20" y="52" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="79" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Observe</text>
+<text x="200" y="103" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Metrics, logs, and traces</text>
+<text x="200" y="126" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">from distributed services</text>
+<path d="M200 141V162" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="164" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="191" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Detect</text>
+<text x="200" y="215" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Identify anomalous signals</text>
+<text x="200" y="238" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">and unreliable agent behaviour</text>
+<path d="M200 253V274" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="276" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="303" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Diagnose</text>
+<text x="200" y="327" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Connect symptoms to causes</text>
+<text x="200" y="350" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">using dependencies and evidence</text>
+<path d="M200 365V386" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="388" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="415" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Respond</text>
+<text x="200" y="439" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Support incident decisions</text>
+<text x="200" y="462" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">and investigate automated repair</text>
+<path d="M200 477V498" stroke="#426875" stroke-width="2" marker-end="url(#arrow)"/>
+<rect x="20" y="500" width="360" height="88" rx="10" fill="white" stroke="#b9cbd1"/>
+<text x="200" y="527" text-anchor="middle" font-family="Arial,sans-serif" font-size="20" font-weight="700" fill="#183443">Evaluate</text>
+<text x="200" y="551" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Realistic faults and benchmarks</text>
+<text x="200" y="574" text-anchor="middle" font-family="Arial,sans-serif" font-size="18" font-weight="400" fill="#183443">Diagnostic and repair quality</text>
+</svg>
diff --git a/static/images/support/bell.svg b/static/images/support/bell.svg
new file mode 100644
index 0000000..ef4dc07
--- /dev/null
+++ b/static/images/support/bell.svg
@@ -0,0 +1,3 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 65 37.2" fill="#00549a" role="img" aria-label="Bell">
+                <path d="M48.3 21.4c-.3-1.6-1-3-1.9-4.2-1.9-2.6-5.1-4.1-9.3-4.1-6.6 0-10.6 4.5-11.8 9.2-1-3.3-4-5.6-7-6.5 2.3-1.7 3.5-4.1 3.5-6.8 0-2.5-1-4.8-2.8-6.3C16.4.4 12.8 0 8.5 0H0v36.7h7.3c2.2 0 4.2 0 6.2-.1 6.2-.2 10.4-3.4 11.8-8 .5 2.1 1.3 3.8 2.5 5.1 2.4 2.6 5.8 3.7 9.3 3.7 4.3 0 7.4-1.2 9.5-3.1l-3-3.5c-1.5 1.2-3.5 1.8-5.7 1.8-4.3 0-6.2-2.7-6.4-5.4h16.9v9.6h6.3V.1h-6.3l-.1 21.3zM6.5 4.9h1.7c1.7 0 3 .2 4 .6 1.9.7 2.8 2.3 2.8 4.1 0 1.6-.6 4.1-4 4.8-.6.1-1.4.2-2.5.2h-2V4.9zm7.1 26.4c-.8.1-1.7.2-2.6.2H6.5v-12h4.1c1.3 0 2.7.1 3.6.3 2.9.8 4.5 2.8 4.5 5.5 0 2.9-1.5 5.3-5.1 6zm18-8.9c.6-2.7 2.5-4.6 5.4-4.6 2.9 0 4.8 1.7 5.4 4.6H31.6zM58.8.1H65v36.5h-6.2V.1z" class="svg-bell-logo-path"></path>
+            </svg>
diff --git a/static/images/support/nserc.png b/static/images/support/nserc.png
new file mode 100644
index 0000000..9a52776
Binary files /dev/null and b/static/images/support/nserc.png differ
diff --git a/tests/test_content_contracts.py b/tests/test_content_contracts.py
index 4839571..a6458b9 100644
--- a/tests/test_content_contracts.py
+++ b/tests/test_content_contracts.py
@@ -74,6 +74,8 @@ def test_research_and_related_content_references_exist(self) -> None:
             with self.subTest(path=path):
                 for area in data.get("research", []):
                     self.assertIn(area, areas)
+                self.assertEqual(len(data.get("research", [])), len(set(data.get("research", []))),
+                                 "A record must not repeat the consolidated research area")
                 for dataset in data.get("related_datasets", []):
                     self.assertIn(dataset, datasets)
                 if data.get("related_publication"):
@@ -90,6 +92,13 @@ def test_research_and_related_content_references_exist(self) -> None:
                         self.assertIn(parsed.scheme, {"http", "https"})
                         self.assertTrue(parsed.netloc)
 
+    def test_legacy_research_filters_are_unique_and_not_separate_areas(self) -> None:
+        areas = {path.parent.name: load_toml_front_matter(path)
+                 for path in (ROOT / "content/research").glob("*/_index.md")}
+        legacy_ids = [alias for data in areas.values() for alias in data.get("legacy_research_ids", [])]
+        self.assertEqual(len(legacy_ids), len(set(legacy_ids)))
+        self.assertFalse(set(legacy_ids).intersection(areas), "A legacy filter must not be a separate research area")
+
 
 if __name__ == "__main__":
     unittest.main()
diff --git a/tests/test_publication_outcomes_editor.py b/tests/test_publication_outcomes_editor.py
new file mode 100644
index 0000000..f8527ed
--- /dev/null
+++ b/tests/test_publication_outcomes_editor.py
@@ -0,0 +1,178 @@
+from __future__ import annotations
+
+from pathlib import Path
+import sys
+import tempfile
+import unittest
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / 'scripts'))
+from edit_profiles import Conflict, ProfileStore
+
+
+class PublicationOutcomesEditorTests(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.temp.cleanup)
+        self.root = Path(self.temp.name)
+        self.store = ProfileStore(self.root)
+
+    def paper(self, title='Example Paper'):
+        return {'title': title, 'authors': ['Example Author'], 'year': 2026,
+                'venue': 'Example Conference', 'publication_type': 'Conference Paper',
+                'external_url': 'https://example.org/paper'}
+
+    def outcome(self, **changes):
+        return {'title': 'Public benchmark', 'url': 'https://example.org/benchmark',
+                'summary': 'Sample data and evaluation scripts.\nIncludes "quoted" labels and Unicode: α.'} | changes
+
+    def create(self, **changes):
+        return self.store.save({'kind': 'publications', 'data': self.paper() | changes})
+
+    def reopen(self):
+        return self.store.get('example-paper', 'publications', include_preview=False)
+
+    def save(self, changes, revision=None):
+        return self.store.save({'kind': 'publications', 'slug': 'example-paper',
+                                'revision': revision or self.reopen()['revision'], 'data': changes})
+
+    def files(self):
+        return {path.relative_to(self.root): path.read_bytes()
+                for path in self.root.rglob('*') if path.is_file()}
+
+    def legacy_paper(self):
+        self.create()
+        path = self.store.path('example-paper', 'publications')
+        source = path.read_text().replace('+++\n', '+++\n# Verified citation source.\n', 1)
+        source = source.replace('\n+++\n', '\ncustom_metadata = { owner = "Example", reviewed = true }\n+++\n')
+        source += '\n## Additional notes\nKeep this publication body and [source](https://example.org/source).\n'
+        path.write_text(source)
+        return path, source
+
+    def test_schema_exposes_optional_repeatable_outcomes_without_requiring_research(self):
+        schema = self.store.schema('publications')
+        fields = {field['key']: field for field in schema['schema']['fields']}
+        outcomes = fields['research_outcomes']
+        self.assertEqual(outcomes['type'], 'objects')
+        self.assertEqual(schema['defaults']['research_outcomes'], [])
+        self.assertEqual(outcomes['add_label'], 'Add research outcome')
+        self.assertFalse(outcomes.get('required', False))
+        self.assertFalse(fields['research'].get('required', False))
+        self.assertEqual({child['key'] for child in outcomes['fields']}, {'title', 'url', 'summary'})
+        self.assertTrue(all(child['required'] for child in outcomes['fields']))
+        self.assertIn('this publication', outcomes['help'])
+        self.assertIn('research areas', outcomes['help'])
+
+    def test_create_reload_edit_and_remove_multiple_outcomes(self):
+        outcomes = [self.outcome(), self.outcome(title='Project website', url='https://example.org/project')]
+        self.create(research_outcomes=outcomes)
+        self.assertEqual(self.reopen()['data']['research_outcomes'], outcomes)
+        self.assertNotIn('research', self.reopen()['data'])
+        self.assertEqual(list(self.files()), [Path('content/publications/example-paper.md')])
+        edited = [outcomes[0] | {'title': 'Updated benchmark', 'summary': 'Revised public release.'}, outcomes[1]]
+        self.save({'research_outcomes': edited})
+        self.assertEqual(self.reopen()['data']['research_outcomes'], edited)
+        self.save({'research_outcomes': edited[1:]})
+        self.assertEqual(self.reopen()['data']['research_outcomes'], edited[1:])
+        self.save({'research_outcomes': []})
+        self.assertEqual(self.reopen()['data']['research_outcomes'], [])
+
+    def test_legacy_publication_can_save_without_adding_optional_fields(self):
+        path, source = self.legacy_paper()
+        self.save({'venue': 'Revised Conference'})
+        self.assertEqual(path.read_text(), source.replace('venue = "Example Conference"', 'venue = "Revised Conference"'))
+        self.assertNotIn('research_outcomes', self.reopen()['data'])
+        self.assertNotIn('research', self.reopen()['data'])
+
+    def test_adding_and_removing_outcomes_preserves_body_comments_and_unknown_metadata(self):
+        path, source = self.legacy_paper()
+        original = self.reopen()
+        self.save({'research_outcomes': [self.outcome()]})
+        self.assertEqual(self.reopen()['data'], original['data'] | {'research_outcomes': [self.outcome()]})
+        self.assertEqual(self.reopen()['body'], original['body'])
+        self.assertIn('# Verified citation source.\n', path.read_text())
+        provenance = '# Public artifact provenance: https://example.org/source\n'
+        path.write_text(path.read_text().replace('research_outcomes = ', provenance + 'research_outcomes = '))
+        revised = self.outcome(title='Revised public benchmark', summary='Updated release scope.')
+        self.save({'research_outcomes': [revised]})
+        self.assertEqual(self.reopen()['data'], original['data'] | {'research_outcomes': [revised]})
+        self.assertEqual(self.reopen()['body'], original['body'])
+        self.assertIn(provenance + 'research_outcomes = ', path.read_text())
+        self.assertIn('# Verified citation source.\n', path.read_text())
+        self.save({'research_outcomes': []})
+        self.assertEqual(path.read_text(), source.replace('\n+++\n', '\n' + provenance + 'research_outcomes = []\n+++\n'))
+
+    def test_unrelated_save_preserves_outcomes_and_their_source_comments_exactly(self):
+        path, _ = self.legacy_paper()
+        self.save({'research_outcomes': [self.outcome()]})
+        source = path.read_text().replace('research_outcomes = [', '# Artifact provenance.\nresearch_outcomes = [\n  ')
+        source = source.replace(' }]', ' }, # Keep this nested source note.\n]')
+        path.write_text(source)
+        outcomes = self.reopen()['data']['research_outcomes']
+        self.save({'venue': 'Revised Conference'})
+        expected = source.replace('venue = "Example Conference"', 'venue = "Revised Conference"')
+        self.assertEqual(path.read_text(), expected)
+        # The editor sends present fields even when their values have not changed.
+        self.save({'venue': 'Another Conference', 'research_outcomes': outcomes})
+        self.assertEqual(path.read_text(), expected.replace('Revised Conference', 'Another Conference'))
+
+    def test_stale_revision_cannot_remove_or_overwrite_newer_outcomes(self):
+        self.create(research_outcomes=[self.outcome()])
+        stale = self.reopen()
+        newer = [self.outcome(title='New public release', url='https://example.org/new')]
+        self.save({'research_outcomes': newer})
+        before = self.files()
+        for changes in ({'research_outcomes': []}, {'research_outcomes': stale['data']['research_outcomes']}, {'venue': 'Stale edit'}):
+            with self.subTest(changes=changes), self.assertRaises(Conflict):
+                self.save(changes, revision=stale['revision'])
+            self.assertEqual(self.files(), before)
+        self.assertEqual(self.reopen()['data']['research_outcomes'], newer)
+
+    def assert_rejected_without_write(self, outcomes):
+        before = self.files()
+        with self.assertRaises(ValueError):
+            self.store.save({'kind': 'publications', 'data': self.paper('Invalid Paper') | {'research_outcomes': outcomes}})
+        self.assertEqual(self.files(), before, 'An invalid new publication must not create a file')
+        with self.assertRaises(ValueError):
+            self.save({'research_outcomes': outcomes})
+        self.assertEqual(self.files(), before, 'An invalid edit must not replace any existing file')
+
+    def test_partial_and_invalid_nested_values_rejected_without_file_write(self):
+        self.create(research_outcomes=[self.outcome()])
+        invalid = [None, {}, 'not a list', [None], ['not an object'], [1], [{}],
+                   [{'title': 'Missing fields'}],
+                   [{'title': 'Missing URL', 'summary': 'Description'}],
+                   [self.outcome(title='')], [self.outcome(title='  ')], [self.outcome(title=1)],
+                   [self.outcome(url=None)], [self.outcome(url=[])],
+                   [self.outcome(summary='')], [self.outcome(summary=' \n ')],
+                   [self.outcome(summary=False)], [self.outcome(summary={})],
+                   [self.outcome(), {'title': 'Incomplete second outcome'}]]
+        for outcomes in invalid:
+            with self.subTest(outcomes=outcomes):
+                self.assert_rejected_without_write(outcomes)
+
+    def test_unsafe_malformed_and_http_outcome_urls_rejected_without_file_write(self):
+        self.create(research_outcomes=[self.outcome()])
+        urls = ['', 'http://example.org/project', 'javascript:alert(1)', 'data:text/html,test',
+                'file:///tmp/project', '//example.org/project', '/project/', 'not a URL',
+                'https://', 'https:///project', 'https://example.org/a b',
+                'https://example.org/a\nb', 'https://example.org/\x00project',
+                'https://example.org\\project', 'https://exa%6dple.org/project',
+                'https://example.org:invalid/project', 'https://example.org:70000/project',
+                'https://[broken/project', 'https://user:password@example.org/project']
+        for url in urls:
+            with self.subTest(url=url):
+                self.assert_rejected_without_write([self.outcome(url=url)])
+
+    def test_valid_https_urls_and_plain_text_descriptions_round_trip(self):
+        outcomes = [self.outcome(url=url) for url in (
+            'https://github.com/example/benchmark',
+            'https://example.org:8443/project?version=2&download=sample%20data#usage',
+            'https://example.org/数据', 'https://[2001:db8::1]/project')]
+        outcomes[0]['summary'] = 'Literal <tag>, **asterisks**, and $x^2$.\nPlain text stays intact.'
+        self.create(research_outcomes=outcomes)
+        self.assertEqual(self.reopen()['data']['research_outcomes'], outcomes)
+
+
+if __name__ == '__main__':
+    unittest.main()
diff --git a/tests/test_publication_versions.py b/tests/test_publication_versions.py
new file mode 100644
index 0000000..20be44a
--- /dev/null
+++ b/tests/test_publication_versions.py
@@ -0,0 +1,181 @@
+"""Published and preprint links remain distinct through editing and rendering."""
+from __future__ import annotations
+
+from html.parser import HTMLParser
+import os
+from pathlib import Path
+import shutil
+import subprocess
+import sys
+import tempfile
+import unittest
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / "scripts"))
+from edit_profiles import ProfileStore
+
+HUGO = os.environ.get("HUGO_BIN") or shutil.which("hugo")
+PUBLISHED = "https://doi.org/10.1234/example"
+PREPRINT = "https://arxiv.org/abs/2601.01234"
+
+
+class PublicationVersionEditorTests(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.temp.cleanup)
+        self.root = Path(self.temp.name)
+        self.store = ProfileStore(self.root)
+
+    def create(self, **links):
+        return self.store.save({"kind": "publications", "data": {
+            "title": "Versioned Paper", "authors": ["Example Author"], "year": 2026,
+            "venue": "Example Conference (accepted)", "publication_type": "Conference Paper",
+            **links}})
+
+    def read(self):
+        return self.store.get("versioned-paper", "publications", include_preview=False)
+
+    def edit(self, **data):
+        return self.store.save({"kind": "publications", "slug": "versioned-paper",
+                                "revision": self.read()["revision"], "data": data})
+
+    def test_preprint_can_gain_a_published_version_and_either_link_can_be_removed(self):
+        self.create(preprint_url=PREPRINT)
+        self.assertNotIn("external_url", self.read()["data"])
+        self.assertEqual(self.read()["data"]["publication_type"], "Conference Paper")
+        self.edit(external_url=PUBLISHED)
+        self.assertEqual(self.read()["data"]["preprint_url"], PREPRINT)
+        self.edit(preprint_url="")
+        self.assertEqual(self.read()["data"]["external_url"], PUBLISHED)
+        self.edit(external_url="", preprint_url=PREPRINT)
+        self.assertEqual(self.read()["data"]["preprint_url"], PREPRINT)
+        before = self.store.path("versioned-paper", "publications").read_bytes()
+        with self.assertRaisesRegex(ValueError, "provide a published / official URL"):
+            self.edit(preprint_url="")
+        self.assertEqual(self.store.path("versioned-paper", "publications").read_bytes(), before)
+
+    def test_no_link_or_whitespace_cannot_create_a_record(self):
+        for links in ({}, {"external_url": "", "preprint_url": ""},
+                      {"external_url": "  ", "preprint_url": "\n"}):
+            with self.subTest(links=links), self.assertRaises(ValueError):
+                self.create(**links)
+            self.assertFalse(self.store.path("versioned-paper", "publications").exists())
+
+    def test_unrelated_edit_preserves_absent_preprint_field_and_existing_provenance(self):
+        self.create(external_url=PUBLISHED)
+        path = self.store.path("versioned-paper", "publications")
+        original = path.read_text().replace("external_url =", "# Published source provenance.\nexternal_url =")
+        original += "\nKeep these explanatory notes.\n"
+        path.write_text(original)
+        self.edit(venue="Published Conference")
+        self.assertEqual(path.read_text(), original.replace("Example Conference (accepted)", "Published Conference"))
+        self.assertNotIn("preprint_url", self.read()["data"])
+        self.edit(preprint_url=PREPRINT)
+        self.assertIn("# Published source provenance.\n", path.read_text())
+        self.assertTrue(path.read_text().endswith("Keep these explanatory notes.\n"))
+
+    def test_bad_preprint_urls_and_duplicate_links_are_rejected_without_writing(self):
+        self.create(external_url=PUBLISHED)
+        path = self.store.path("versioned-paper", "publications")
+        before = path.read_bytes()
+        for url in ("javascript:alert(1)", "//arxiv.org/abs/2601.01234", "https://",
+                    "https://example.org:broken/paper", "https://example.org/a b", PUBLISHED):
+            with self.subTest(url=url), self.assertRaises(ValueError):
+                self.edit(preprint_url=url)
+            self.assertEqual(path.read_bytes(), before)
+
+    def test_arxiv_links_belong_in_the_preprint_field_including_arxiv_dois(self):
+        for url in (PREPRINT, "https://export.arxiv.org/pdf/2601.01234",
+                    "https://ARXIV.ORG/abs/2601.01234", "https://doi.org/10.48550/arXiv.2601.01234"):
+            with self.subTest(url=url), self.assertRaisesRegex(ValueError, "put arXiv links in Preprint URL"):
+                self.create(external_url=url)
+            self.assertFalse(self.store.path("versioned-paper", "publications").exists())
+        self.create(preprint_url="https://example.org/manuscript.pdf")
+        self.assertEqual(self.read()["data"]["preprint_url"], "https://example.org/manuscript.pdf")
+
+
+class ActionLinks(HTMLParser):
+    def __init__(self, html):
+        super().__init__()
+        self.active = False
+        self.links = []
+        self.link = None
+        self.feed(html)
+
+    def handle_starttag(self, tag, attrs):
+        attrs = dict(attrs)
+        if tag == "p" and "publication-links" in attrs.get("class", "").split():
+            self.active = True
+        if tag == "a" and self.active:
+            self.link = {"url": attrs["href"], "classes": attrs.get("class", "").split(), "text": ""}
+
+    def handle_data(self, text):
+        if self.link is not None:
+            self.link["text"] += text
+
+    def handle_endtag(self, tag):
+        if tag == "a" and self.link is not None:
+            self.links.append(self.link)
+            self.link = None
+        if tag == "p":
+            self.active = False
+
+
+@unittest.skipUnless(HUGO, "Hugo is required for publication version rendering")
+class PublicationVersionRenderingTests(unittest.TestCase):
+    def test_versions_and_source_fallback_are_labeled_on_paper_and_research_pages(self):
+        with tempfile.TemporaryDirectory(prefix="msrg-versions-") as tmp:
+            root = Path(tmp)
+            for directory in ("content", "layouts", "assets", "static", "data", "archetypes"):
+                shutil.copytree(ROOT / directory, root / directory)
+            shutil.copy2(ROOT / "hugo.yaml", root / "hugo.yaml")
+            store = ProfileStore(root)
+            examples = [
+                ("Both versions", {"external_url": PUBLISHED, "preprint_url": PREPRINT},
+                 [(PUBLISHED, "Published version"), (PREPRINT, "Preprint (arXiv)")]),
+                ("Published only", {"external_url": "https://openreview.net/forum?id=conference"},
+                 [("https://openreview.net/forum?id=conference", "Published version")]),
+                ("Accepted preprint only", {"preprint_url": PREPRINT}, [(PREPRINT, "Preprint (arXiv)")]),
+                ("Official workshop copy", {"external_url": "https://openreview.net/forum?id=workshop",
+                                            "publication_type": "Workshop Paper"},
+                 [("https://openreview.net/forum?id=workshop", "Workshop version")]),
+                ("Similar hostname", {"preprint_url": "https://arxiv.org.example.com/paper"},
+                 [("https://arxiv.org.example.com/paper", "Preprint")]),
+                ("Official record", {"external_url": "https://dblp.org/rec/example"},
+                 [("https://dblp.org/rec/example", "Publication record")]),
+            ]
+            expected = []
+            for title, links, labels in examples:
+                saved = store.save({"kind": "publications", "data": {
+                    "title": title, "authors": ["Example Author"], "year": 2030,
+                    "venue": "Example Conference (accepted)", "publication_type": "Conference Paper",
+                    "research": ["rca-aiops"], **links}})
+                expected.append((saved["slug"], links, labels))
+            build = subprocess.run([HUGO, "--source", str(root), "--destination", str(root / "public"),
+                                    "--cacheDir", str(root / "cache"), "--panicOnWarning"],
+                                   capture_output=True, text=True)
+            self.assertEqual(build.returncode, 0, build.stdout + build.stderr)
+            research = (root / "public/research/rca-aiops/index.html").read_text()
+            research_links = ActionLinks(research).links
+            self.assertEqual(len(research_links), 7, "Six recent papers: five single links and one pair")
+            for slug, links, labels in expected:
+                html = (root / "public/publications" / slug / "index.html").read_text()
+                rendered = ActionLinks(html).links
+                self.assertEqual([(link["url"], link["text"]) for link in rendered], labels)
+                for url, label in labels:
+                    self.assertTrue(any(link["url"] == url and link["text"] == label for link in research_links))
+                if links.get("preprint_url"):
+                    self.assertIn("button-secondary", rendered[-1]["classes"])
+                    self.assertIn("publication-preprint", rendered[-1]["classes"])
+                    for link in research_links:
+                        if link["url"] == links["preprint_url"]:
+                            self.assertIn("publication-preprint", link["classes"])
+                if links.get("external_url") and links.get("preprint_url"):
+                    self.assertNotIn("button-secondary", rendered[0]["classes"])
+                if not links.get("external_url"):
+                    self.assertIn("Read the abstract and full preprint at the source.", html)
+                    self.assertNotIn("Read the abstract and full publication at the source.", html)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/tests/test_research_outcomes.py b/tests/test_research_outcomes.py
new file mode 100644
index 0000000..a8af240
--- /dev/null
+++ b/tests/test_research_outcomes.py
@@ -0,0 +1,150 @@
+"""Exercise publication-owned outcomes through the editor store and real templates."""
+from __future__ import annotations
+
+from html import unescape
+from html.parser import HTMLParser
+import os
+from pathlib import Path
+import re
+import shutil
+import subprocess
+import sys
+import tempfile
+import unittest
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / "scripts"))
+from edit_profiles import ProfileStore
+
+HUGO = os.environ.get("HUGO_BIN") or shutil.which("hugo")
+AIOPS = "rca-aiops"
+QUANTUM = "quantum-computing-systems"
+
+
+class Links(HTMLParser):
+    def __init__(self, html):
+        super().__init__()
+        self.hrefs = []
+        self.feed(html)
+
+    def handle_starttag(self, tag, attrs):
+        if tag == "a":
+            self.hrefs.append(dict(attrs).get("href"))
+
+
+@unittest.skipUnless(HUGO, "Hugo is required for outcome rendering tests")
+class ResearchOutcomeTests(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory(prefix="msrg-outcome-test-")
+        self.addCleanup(self.temp.cleanup)
+        self.root = Path(self.temp.name)
+        for directory in ("content", "layouts", "assets", "static", "data", "archetypes"):
+            shutil.copytree(ROOT / directory, self.root / directory)
+        shutil.copy2(ROOT / "hugo.yaml", self.root / "hugo.yaml")
+        self.store = ProfileStore(self.root)
+
+    def outcome(self, name):
+        return {"title": name, "url": f"https://example.org/outcomes/{name}",
+                "summary": "Public evaluation scripts and sample data."}
+
+    def paper(self, title, year, areas, outcomes=None):
+        data = {"title": title, "year": year, "authors": ["Example Author"],
+                "venue": "Example Conference", "publication_type": "Conference Paper",
+                "external_url": "https://example.org/paper", "research": areas}
+        if outcomes is not None:
+            data["research_outcomes"] = outcomes
+        result = self.store.save({"kind": "publications", "data": data})
+        return result["slug"]
+
+    def edit(self, slug, **changes):
+        record = self.store.get(slug, "publications", include_preview=False)
+        self.store.save({"kind": "publications", "slug": slug,
+                         "revision": record["revision"], "data": changes})
+
+    def build(self):
+        result = subprocess.run(
+            [HUGO, "--source", str(self.root), "--destination", str(self.root / "public"),
+             "--cacheDir", str(self.root / "cache"), "--panicOnWarning"],
+            capture_output=True, text=True,
+        )
+        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
+
+    def page(self, route):
+        return (self.root / "public" / route.strip("/") / "index.html").read_text()
+
+    def section(self, html, identifier):
+        match = re.search(rf'<section\b[^>]*\bid="{identifier}"[^>]*>(.*?)</section>', html, re.S)
+        self.assertIsNotNone(match, f"Missing section {identifier}")
+        return match[1]
+
+    def assert_preview_limit(self, html):
+        section = self.section(html, "publications")
+        self.assertEqual(section.count('class="list-card"'), 6)
+        return Links(section).hrefs
+
+    def test_all_papers_supply_ordered_outcomes_independently_of_recent_citations(self):
+        first = self.outcome("first") | {
+            "title": '<b>Benchmark</b> & "examples"',
+            "summary": '<script>alert("example")</script> is plain text, not executable markup.',
+        }
+        second = self.outcome("second")
+        oldest = self.paper("Zebra old outcome paper", 2000, [AIOPS], [first, second])
+        self.paper("Alpha old outcome paper", 2000, [AIOPS], [self.outcome("alphabetical")])
+        self.paper("Newer outcome paper", 2001, [AIOPS], [self.outcome("newer")])
+        unscoped = self.paper("Outcome without an area", 2000, [], [self.outcome("unscoped")])
+        absent = self.paper("Paper without outcomes", 2000, [AIOPS])
+        empty = self.paper("Paper with empty outcomes", 2000, [AIOPS], [])
+        self.build()
+
+        research = self.page("research/rca-aiops")
+        outcomes = self.section(research, "outcomes")
+        links = Links(outcomes).hrefs
+        fixture_links = [url for url in links if url.startswith("https://example.org/outcomes/")]
+        self.assertEqual(fixture_links, [self.outcome(name)["url"]
+                                      for name in ("newer", "alphabetical", "first", "second")])
+        owner_url = f"/publications/{oldest}/"
+        self.assertEqual(links.count(owner_url), 2, "Both outputs link to their owning publication")
+        self.assertNotIn(owner_url, self.assert_preview_limit(research), "The old paper must be outside the preview")
+        self.assertNotIn(first["url"], self.page("research/quantum-computing-systems"))
+
+        detail = self.section(self.page(f"publications/{oldest}"), "research-outcomes")
+        for rendered in (outcomes, detail):
+            self.assertNotIn("<script>", rendered)
+            self.assertNotIn("<b>Benchmark</b>", rendered)
+            self.assertIn(first["title"], unescape(rendered))
+            self.assertIn(first["summary"], unescape(rendered))
+        self.assertEqual(Links(detail).hrefs, [first["url"], second["url"]])
+        self.assertIn(self.outcome("unscoped")["url"], self.section(
+            self.page(f"publications/{unscoped}"), "research-outcomes"))
+        for slug in (absent, empty):
+            self.assertNotIn('id="research-outcomes"', self.page(f"publications/{slug}"))
+
+    def test_editing_publication_keeps_association_and_moves_or_removes_outputs(self):
+        first, second = self.outcome("keep"), self.outcome("remove")
+        slug = self.paper("Original outcome paper", 2000, [AIOPS], [first, second])
+        self.build()
+        owner_url = f"/publications/{slug}/"
+        self.assertIn(first["url"], self.section(self.page("research/rca-aiops"), "outcomes"))
+
+        updated = first | {"title": "Renamed public benchmark", "url": "https://example.org/renamed"}
+        self.edit(slug, title="Revised paper title", research=[QUANTUM], research_outcomes=[updated])
+        self.build()
+        self.assertNotIn(updated["url"], self.page("research/rca-aiops"))
+        quantum = self.page("research/quantum-computing-systems")
+        links = Links(self.section(quantum, "outcomes")).hrefs
+        self.assertIn(updated["url"], links)
+        self.assertIn(owner_url, links, "Title and research edits must not detach the related publication")
+        self.assertNotIn(second["url"], quantum)
+        self.assertNotIn(owner_url, self.assert_preview_limit(quantum))
+        detail = self.page(f"publications/{slug}")
+        self.assertIn("Revised paper title", detail)
+        self.assertEqual(Links(self.section(detail, "research-outcomes")).hrefs, [updated["url"]])
+
+        self.edit(slug, research_outcomes=[])
+        self.build()
+        self.assertNotIn('id="research-outcomes"', self.page(f"publications/{slug}"))
+        self.assertNotIn(updated["url"], self.page("research/quantum-computing-systems"))
+
+
+if __name__ == "__main__":
+    unittest.main()