Restore authenticated replay of the historical complex-ceiling control

docs/research/agent-procedure.md

@@ -139,6 +139,26 @@ historical catalogue, picker and baseline bytes remain intact. The replacement
 authenticates selected sources and checks metadata consistency; it is not a cold
 replay or supplier admission.
 
+**Keep the complex-ceiling control in its historical context.** Four cards
+(`complex-rank-shape-decoder`, `complex-new-helper-inventory`,
+`centre-path-sharing` and `coverage-law-change`) name
+`research/complex-local-word-ceiling/verify.py`. That frozen checker reads the
+#1094 certificate schema and fails on selected #4470 with
+`KeyError: 'complex_profile'`. Run the current selected orientation above, then
+use the [authenticated historical replay](../../research/historical-complex-ceiling-replay/README.md)
+for the card's original positive control:
+
+```sh
+python3 -B scripts/research_package.py --package research/historical-complex-ceiling-replay
+```
+
+This executes the unchanged scientific checker and its controls against exact
+#1094/#1042 inputs in a temporary tree. Its ceiling and `levers.py` budgets
+belong to that historical word, coverage law and outer guard. Before screening
+a changed current supplier, derive and source-bind the applicable complete cost
+law; a historical pass does not make the old ceiling a cap on #4470. The card,
+its pinned sources and the live selected record retain their bytes.
+
 In a sparse checkout, the helper can validate omitted catalogue references when
 they are regular files in `HEAD` with matching skip-worktree index entries. It
 reports these as `sparse_checkout.omitted_catalogue_paths`. For absent paths,

research/historical-complex-ceiling-replay/README.md

@@ -0,0 +1,149 @@
+# Reproduce the historical complex-ceiling control without changing selection
+
+**Tooling repair; no new mathematical bound or supplier.** This package runs
+Douglas Colkitt's unchanged PR1240 ceiling checker against its exact #1094
+selected record and #1042 comparison. It restores a runnable scientific
+positive control for four research cards while preserving all predecessor
+bytes and the current selected frontier.
+
+At base `7c472563474e60185e6ca101a17622133d047e56`, priority revision 7,
+`complex-rank-shape-decoder`, `complex-new-helper-inventory`,
+`centre-path-sharing` and `coverage-law-change` still start with
+`research/complex-local-word-ceiling/verify.py`. With its inputs materialized,
+that command raises `KeyError: 'complex_profile'`: selected #4470's certificate
+has a productive-helper schema. The [existing handoff](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19132013)
+already identified this mismatch. This contribution adds the actual historical
+replay route; it does not claim to discover the mismatch.
+
+The original checker reproduces exactly:
+
+```text
+PASS_COMPLEX_LOCAL_WORD_CEILING a_ceiling 0.00169005410908900006 kappa<= 0.0016872026453254019
+```
+
+Those values describe **the #1094 model**. The selected conditional #4470 value
+at this base is 0.00302015, under a different construction and accounting.
+For current orientation run `python3 -B scripts/research_selected_baseline.py`
+after materializing its source closure. A historical ceiling or lever budget
+must not be applied to that current contract without a new applicable cost law.
+
+## Exact source and execution boundary
+
+The nine-file executable closure comprises the original `verify.py`,
+`ceiling.py`, `RESULT.json`, `SOURCES.json`, two #1094 audit inputs, the #1042
+checkpoint and assembly, and the historical selected record. Every file is
+SHA256-authenticated before any predecessor code runs. The three live input
+pins in the original `SOURCES.json` are then independently reconciled with
+that payload. The added #1042 pins also bind the actual predecessor comparison.
+
+The retained [historical-selected.json](historical-selected.json) is the exact
+`certificates/selected-result.json` blob `ae6a99f28aa468649eb66a361e764d769363bf11`
+at original base `e3bb23b4c6bae16ba277f9443e11257a0a37780a`, with SHA256
+`10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32`.
+Its original Git path is a provenance pin in [SOURCES.json](SOURCES.json);
+the package-owned copy is authenticated by the wrapper and exported with it.
+The current `certificates/checkpoints/frontier1094.json` has SHA256
+`bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129`.
+It is byte-distinct and is not silently substituted for the old pin.
+
+Only the retained selected record is relocated, to its original path **inside
+a fresh temporary tree**. All other inputs keep their paths and bytes. The
+wrapper executes the unchanged checker as a separate isolated Python process,
+preserving an explicitly requested `-O` level. It copies no bytecode or unrelated
+repository files, uses a 90-second child timeout and never passes `--write`.
+The original checker retains its exact `RESULT.json` comparison, #1042 check,
+coverage and moment calculations and four original failure/dominance controls.
+
+The wrapper requires successful exit and the complete expected stdout, compares
+all temporary input bytes after the positive replay, reauthenticates original
+inputs at return, and checks the live selected record is unchanged when present.
+The standalone verifier requires no Git, network or third-party dependencies.
+The generic metadata command does use Git to authenticate provenance. In a
+shallow checkout, fetch the public original base explicitly if it is absent:
+
+```sh
+git fetch --depth 1 --no-tags https://github.com/CrocSwap/integer-mult-bounds.git e3bb23b4c6bae16ba277f9443e11257a0a37780a
+```
+
+This fetch is not a full-history ancestry audit. A committed Git-free export
+contains everything needed for the standalone replay and avoids that lookup.
+
+## Reproduce and rejection coverage
+
+From the repository root, CPython 3.12 and the standard library:
+
+```sh
+python3 -B scripts/research_package.py --package research/historical-complex-ceiling-replay --check-only
+python3 -B scripts/research_package.py --package research/historical-complex-ceiling-replay
+```
+
+The registered commands run both normal and optimized modes. Direct commands
+also work from an exported tree:
+
+```sh
+python3 -I -W error -B research/historical-complex-ceiling-replay/verify.py
+python3 -I -W error -O -B research/historical-complex-ceiling-replay/verify.py
+```
+
+Each mode includes:
+
+- The entire unchanged historical positive replay and its four original controls.
+- Eighteen source-integrity controls: individually corrupt and remove every one
+  of the nine inputs, requiring the specific input's rejection.
+- Three scientific-data controls in separate disposable trees. A changed owner
+  profile rejects with `profile mismatch`; an impossible #1042 endpoint rejects
+  with `previous endpoint above the ceiling`; a changed expected receipt rejects
+  with `RESULT.json differs`. These intentionally bypass only the wrapper's
+  earlier hash gate to test the unchanged scientific checker itself.
+- Exact preservation of all source inputs and the live selected record.
+
+The final line is
+`PASS_HISTORICAL_COMPLEX_CEILING_REPLAY context=PR1094 current_supplier_bound=false inputs=9 input_controls=18 scientific_controls=3 unchanged=true`.
+The PR and [canonical claim](https://github.com/CrocSwap/integer-mult-bounds/discussions/4458#discussioncomment-19152822)
+record measured runs, committed/export checks and hosted status. No full selected
+cold replay, Lean, Docker, new cost law, independent human review or proof-kernel
+acceptance is claimed. The underlying historical theorem keeps its original
+scope and proof dependencies.
+
+## Task choice, overlap and credit
+
+The picker selected `complex-rank-shape-decoder` with nonce
+`3d916505ccc35c2632298bca8333380c`. [TASK.json](TASK.json) explicitly changes the
+scope from decoder synthesis to this first-step reproducibility repair. Its
+original card and originating catalogue remain pinned. The standing procedure
+links this route without changing the catalogue, allocation, frozen checkers
+or scientific receipts. This complements #4906's generic-baseline navigation;
+that change deliberately left named scientific replay commands alone.
+
+Entry overlap covered 304 open PR bodies and 4,625 Discussion records with
+complete remote pagination at 18:55/18:57 UTC on October 23, 2026. Local note
+coverage was sparse (9 of 7,267 tracked paths); directly relevant packages,
+guidance, #4458/#1166 and targeted all-state PR searches were inspected.
+Arbitrary diffs, omitted unrelated notes and unpublished work were not covered.
+No claim of global novelty follows from these searches.
+
+Credit Douglas Colkitt and the original credited Claude assistance for PR1240,
+the PR1012 coverage identity, and the #1094/#1042 construction and audit authors
+retained in the original package. Mihir Srivastava's #4463/#4906 supply the
+current-schema orientation and generic startup routing; lee-phys's #4631 and
+the retained #4936/#5050 contributors supply sparse and source-loading follow-ups.
+Dennis's existing historical-check replay and all repository workflow tools
+retain their original credit. This focused copy needs only nine inputs; it
+does not change those shared tools or their historical configurations.
+
+Prepared by Mihir Srivastava (@mihir-s-05) with substantial OpenAI Codex assistance
+in inspection, implementation, documentation and verification; no subagents.
+Apache-2.0. This is author-checked tooling evidence, not scientific approval.
+
+## Open next questions
+
+1. **Bind a current helper-inventory cost law.** Authenticate the selected closure
+   with `research_selected_baseline.py`, then derive the actual paid completion,
+   compiler and outer guard for a named changed invoice. This matters because
+   the historical cap is inapplicable to #4470. The missing complete law and
+   consumer bill are blockers; this replay does not supply them.
+2. **Migrate the four emitted historical first steps.** Extend the exact-path
+   dependency inventory in [the startup compatibility note](../../docs/research/task-startup-compatibility.md)
+   before editing the pinned catalogue. Users who skip the standing procedure
+   still see the legacy command. Preserving historical task/source receipts
+   requires a separate migration decision; no catalogue repin is included here.

research/historical-complex-ceiling-replay/SOURCES.json

@@ -0,0 +1,74 @@
+{
+  "schema_version": 1,
+  "base_commit": "7c472563474e60185e6ca101a17622133d047e56",
+  "files": [
+    {
+      "path": "certificates/checkpoints/frontier1042.json",
+      "usage": "input",
+      "git_blob": "f451ced6626a1f03aba6915ce23a136b01ebe51d",
+      "sha256": "207d1f49f113c585024b9d3c17eadbb8b1be17c1321e3d190aabe2f18ce1c5b7"
+    },
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "ae6a99f28aa468649eb66a361e764d769363bf11",
+      "sha256": "10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32",
+      "pin": "head:e3bb23b4c6bae16ba277f9443e11257a0a37780a:certificates/selected-result.json"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
+      "sha256": "f7a48bfa2659b88378282a8eee7d9fe52cdf505594124c4d1686b58510fffd72"
+    },
+    {
+      "path": "research/community-frontier1042-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "4e308dc23da97373d66e5d8d4a68b0661caea8e9",
+      "sha256": "093f35445812ea96fe1bfe6a3af7d769cb7f274e08f4fb30f71a17ed1161e54e"
+    },
+    {
+      "path": "research/community-frontier1094-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "5fd676df060f3b450782fdf5ff5f6beb63a5eeb0",
+      "sha256": "1122cb3a575f8c788345fc1bf4c108ac7c3250fb81a5988671f5eb04b3ab3045"
+    },
+    {
+      "path": "research/community-frontier1094-audit/INDEPENDENT1094.json",
+      "usage": "input",
+      "git_blob": "009a385e26858058a18188364078d162aa6f9519",
+      "sha256": "301fb8a0e4ab7e3c383ff974c54c08178283b2a3453c5282757f692a3c7a4bfd"
+    },
+    {
+      "path": "research/complex-local-word-ceiling/RESULT.json",
+      "usage": "input",
+      "git_blob": "77f55043a679dd722330f1e2499863e36376d9f9",
+      "sha256": "d91d6c46509875abf1f61de67c2609656d3dc685451299f83caa6f922493692d"
+    },
+    {
+      "path": "research/complex-local-word-ceiling/SOURCES.json",
+      "usage": "input",
+      "git_blob": "c45a9995fe68f9b42c88c02e237269c60533ef96",
+      "sha256": "df587fbef4bb7c00f827f4a4eaddf2f4910ef19f6decf32984f1d32ac036de6a"
+    },
+    {
+      "path": "research/complex-local-word-ceiling/ceiling.py",
+      "usage": "input",
+      "git_blob": "6c0f54527d879396f689523bfde19928a9b6157c",
+      "sha256": "86a1432870b3b21d8454106ec9162fddc0471472d9214c590a1420f2426cedfc"
+    },
+    {
+      "path": "research/complex-local-word-ceiling/verify.py",
+      "usage": "input",
+      "git_blob": "069f09de381a52ac91ec3edb372bd4536a051d91",
+      "sha256": "06b59955c69a2e7f49aa031d2ebaac2f39dbe107f4087122704f2a37ce9806d5"
+    }
+  ],
+  "retained_copy": {
+    "path": "research/historical-complex-ceiling-replay/historical-selected.json",
+    "source": "head:e3bb23b4c6bae16ba277f9443e11257a0a37780a:certificates/selected-result.json",
+    "git_blob": "ae6a99f28aa468649eb66a361e764d769363bf11",
+    "sha256": "10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32",
+    "verification": "The standalone verifier authenticates this package-owned input and reconciles it with the unchanged historical SOURCES.json. The standard provenance row above authenticates its original Git object; the package export retains the copy without a runtime Git dependency."
+  }
+}