docs: refresh stale notes (cache sizes measured, pqxx reconcile done, 05 re-run recorded)

.github/workflows/examples.yml

@@ -185,18 +185,18 @@ jobs:
   #     1.6 GB -> 163 MB), budgeted here at 20% = 135 MB for the largest. Per-pom Maven entries
   #     were 56-350 MB and per-go.sum Go entries 22-248 MB (avg 111 MB), so those two are
   #     what outgrew the quota, not Conan. Steady state on main:
-  #       Conan   22 x <=135 MB (typical 82)     <= 3.0 GB
-  #       Maven   1 shared union                 <= 1.0 GB   (estimate)
-  #       Go      1 shared module cache          <= 0.6 GB   (estimate)
+  #       Conan   22 x <=135 MB (typical 80)     <= 3.0 GB
+  #       Maven   1 shared union                 0.45 GB  (measured: 446 MB)
+  #       Go      1 shared module cache          0.22 GB  (measured: 216 MB)
   #       layers  27 blobs = 200 MB measured     0.25 GB
   #       Pages   bundler cache                  <= 0.05 GB  (estimate)
-  #       typical 22 x 82 MB + 1.0 + 0.6 + 0.25 + 0.05 = ~3.7 GB; bound ~4.9 GB.
-  #       Peak ~8 GB: a Conan key bump or bulk lock change leaves the superseded per-lock
+  #       typical 22 x 80 MB + 0.45 + 0.22 + 0.25 + 0.05 = ~2.7 GB; bound ~3.9 GB.
+  #       Peak ~7 GB: a Conan key bump or bulk lock change leaves the superseded per-lock
   #       entries in place until they age out, and layer blobs stay 7 days unused.
-  #     Measured: unpruned Conan entries (API), BuildKit layers. Partly measured: the prune
-  #     ratio (one project, 25-caching/cpp). Estimated: Maven and Go unions, Pages. After the
-  #     first merge, check the real saved Maven and Go entry sizes (`gh cache list --key`)
-  #     and correct these numbers. PR refs hold nothing.
+  #     Measured on main 2026-10-15 (`gh cache list --key`): Conan 27 entries = 2.16 GB
+  #     (5 superseded after the S14 lock changes), Maven 446 MB, Go 216 MB; 3.46 GB in all.
+  #     Partly measured: the prune ratio (one project, 25-caching/cpp). Estimated: Pages.
+  #     PR refs hold nothing.
   #   - Hygiene: .github/workflows/cache-cleanup.yml deletes the caches of a PR ref when the
   #     PR closes; CLAUDE.md ("CI cache budget") has the commands for inspecting and clearing.
   #

CLAUDE.md

@@ -104,11 +104,11 @@ The repo gets 10 GB of Actions cache. Policy in `examples.yml`: every run restor
 `warm-go` jobs, which resolve every project and delete the superseded entry. Conan is one entry per
 distinct `conan.lock` (22), pruned of sources, build trees and downloads before it is saved; on a main miss the
 builder stage is forced to run so the cache mount is never empty. BuildKit layers are `type=gha,mode=min`.
-Steady state on main: typical about 3.7 GB (22 x 82 MB Conan + Maven 1.0 + Go 0.6 + layers 0.25 + Pages 0.05),
-bound 4.9 GB, peak about 8 GB when a Conan key bump or bulk lock change leaves superseded per-lock entries and
-layer blobs age out after 7 days unused. Measured: unpruned Conan entries, BuildKit layers (27 blobs = 200 MB).
-Partly measured: prune ratio (one project). Estimated: Maven <=1.0 GB, Go <=0.6 GB, Pages. After the first merge,
-check the real saved Maven and Go entry sizes and correct the numbers (arithmetic in the `examples.yml` comment).
+Steady state on main, measured 2026-10-15: Conan 27 entries = 2.16 GB (avg 80 MB; 5 superseded entries
+from the S14 lock changes age out after 7 days unused), Maven 446 MB, Go 216 MB, 3.46 GB in all. Typical about
+2.7 GB (22 x 80 MB Conan + Maven 0.45 + Go 0.22 + layers 0.25 + Pages 0.05), bound 3.9 GB (22 x 135 MB Conan),
+peak about 7 GB when a Conan key bump or bulk lock change leaves superseded per-lock entries and layer blobs
+age out after 7 days unused. Arithmetic in the `examples.yml` comment.
 `cache-cleanup.yml` drops a PR ref's caches when the PR closes (same-repo PRs only).
 
 Check usage with `gh api repos/patterncatalyst/cloud-native-design-patterns/actions/cache/usage`. Drop one PR's
@@ -153,8 +153,6 @@ Do not use `gh cache delete --all`: it also removes the entries the current poli
     (cppgraphqlgen needs 3.x), lua 5.4.8 (sol2), librdkafka 2.14.2 and opentelemetry-cpp 1.26.0 (ConanCenter
     behind upstream). flex and bison get `MAKEINFO=true` through a scoped Conan conf, not a fake binary.
     The Conan cache mount id is `cndp-conan2`; CI persists it with buildkit-cache-dance.
-  - ⚠️ Reconcile: the r76 outbox codetab in `_docs/05-event-driven.md` used `pqxx` for
-    brevity; it now shows libpq, as the example does.
 - **JVM / .NET / Python** (as used in the site codetabs): Spring Boot (Spring Kafka,
   `spring-boot-starter-websocket`, Kafka Streams); Quarkus (`websockets-next`, SmallRye
   Reactive Messaging, Kafka Streams); .NET 10.0.12 (**Confluent.Kafka** 2.16.0 for Kafka —

_plans/s13-idiomatic-plan.md

@@ -137,8 +137,8 @@ and Kafka payload drift (DTOs only, compare with kcat); idempotent consumers und
 and start time against compose `--wait` and the 05 k8s limits; deck `expect` counts (measure on committed decks);
 Python lock drift (diff `uv export` against the old pins); long matrix runs.
 
-## Not re-run
+## Re-run after S14
 
-Example 05 on minikube (Appendix R) and OpenShift Local (Appendix S) was not repeated after S13 (D7). Its
-Quarkus, Spring Boot and .NET services changed (ORMs, health endpoints), so those two footers describe the
-pre-S13 code; the deploy scripts accept the new `"UP"` health bodies.
+Example 05 on minikube (Appendix R) and OpenShift Local (Appendix S) was not repeated during S13 (D7). It was
+re-run on the S13/S14 code on 2026-10-13 (PR #84): minikube 6/6 and OpenShift Local 5/5 languages, 10/10 checks
+each, and both footers now describe the current code.