docs: record the 2026-10-06 evening rulings and the overnight queue

CLAUDE.md

@@ -1138,10 +1138,13 @@ section headers).
    as exhaustive, and this step wasn't in the checklist — which is
    how the README drifted 24 page bullets (plus the Practice counts:
    8 quizzes and 3 drills unlisted) behind by 2026-07-01.
-6. Consider bumping the home-page hero's `Latest: <name>` badge
-   to point at the new tool — `html/index.html`, the
-   `<p class="hero-latest">` paragraph (~L382). Editorial pick; skip
-   on small revisions.
+6. Consider pointing the home-page hero's `Featured · <name>` badge
+   at the new tool — `html/index.html`, the `<p class="hero-latest">`
+   paragraph (~L400; relabelled from `Latest:` 2026-07-19, owner
+   decision). Editorial pick; skip on small revisions. As of 2026-10-06
+   it is a **seasonal pick** — the Coil Freeze Protection lesson for
+   freezestat season, with the DDC Workbench anchor preserved in an
+   HTML comment beside it for the spring revert.
 7. Bump `package.json.version` when shipping something notable; the
    footer reads it via `html/_data/site.js`. A new tool is a minor
    bump (`1.X.0`); a bug fix is a patch bump (`1.X.Y`). Bump with
@@ -1266,7 +1269,7 @@ opt-out.
 7. Add the quiz/drill to `README.md`'s Practice groups and bump its
    count sentence (same reason as tools step 5b — the tour is
    exhaustive).
-8. Same `Latest:` badge + `package.json.version` rules as tools.
+8. Same `Featured` badge + `package.json.version` rules as tools.
 
 ## Git conventions
 
@@ -1480,17 +1483,25 @@ re-submit); add `--dry-run` to print the URL list without POSTing.
 opens against its parent's branch (no run), and when the parent merges
 GitHub retargets the child to `main` with an `edited` event (still no
 run) — so it shows **no `test` check at all**, `mergeStateStatus` reads
-CLEAN, and `gh pr merge` goes through untested. Three chain PRs merged
+CLEAN, and `gh pr merge` goes through untested. Two chain PRs merged
 that way on 2026-10-06 before anyone noticed (a full local suite on
-`main` came back clean, 1223/0/1). **Rule:** after a parent merges,
-rebase the child onto `origin/main` and force-push (the push is a
+`main` came back clean, 1223/0/1). **Fixed on the workflow side
+2026-10-06 (PR #645, codebase-issues #331):** `pull_request` now also
+listens for `edited`, gated so only a base change runs the job, and
+`push: branches: [main]` tests `main` itself after every merge. Two
+things to know: a title or body edit still fires `edited`, the gate
+skips the job, and the rollup then shows a `test` check with conclusion
+SKIPPED **beside** the real run on the same head — read SUCCESS on the
+head SHA and ignore the SKIPPED twin. And the automatic-retarget arm
+was unproven when this was written (no stacked child had merged since),
+so the belt-and-braces **rule stands:** after a parent merges, rebase
+the child onto `origin/main` and force-push (the push is a
 `synchronize`, which runs), then merge only when
 `gh pr view N --json statusCheckRollup` shows a `test` check with
-`SUCCESS` — a green parent is not evidence about the child. The
-workflow-side fix is codebase-issues #331. Before a merge session, a
-pairwise `git merge-tree --write-tree A B` over the whole queue finds
-the adjacent-hunk ledger conflicts GitHub only reports once a parent
-has merged.
+`SUCCESS` — a green parent is not evidence about the child. Before a
+merge session, a pairwise `git merge-tree --write-tree A B` over the
+whole queue finds the adjacent-hunk ledger conflicts GitHub only
+reports once a parent has merged.
 
 ## Local preview & tests
 
@@ -1501,13 +1512,17 @@ has merged.
   home server's hub docroot, where a rootless Caddy serves it at
   `https://cfdev.home.arpa/`. **Owner's box only** — it is a home-lab
   convenience, not part of the deploy path, and it no-ops nowhere
-  else (the destination guard refuses any path not ending
-  `/caddy/dashboard/cfdev`). **From a worktree outside the repo's
-  parent directory, pass `CF_PREVIEW_DIR=$HOME/caddy/dashboard/cfdev`**
-  — the default resolves `../caddy/dashboard/cfdev` relative to the
-  checkout, and the suffix guard accepts the sibling folder that
-  creates, so the script reports "live" while the hub serves the old
-  build (codebase-issues #330). It publishes a **snapshot, not a
+  else. **The default destination is home-relative**
+  (`~/caddy/dashboard/cfdev`, since PR #646 / codebase-issues #330), so
+  it works from any checkout or worktree; `CF_PREVIEW_DIR` overrides it.
+  The guard has three checks and refuses on any of them: the path must
+  end `/caddy/dashboard/cfdev`, sit under `$HOME`, and its
+  `caddy/dashboard` parent must already exist (the script no longer
+  creates it — that third check is what catches an in-home lane
+  worktree, where the old repo-relative default passed the other two
+  and reported "live" while the hub served the old build).
+  `npm run publish:preview -- --dry-run` runs every guard and prints
+  the plan without touching the docroot. It publishes a **snapshot, not a
   server**: nothing watches, so every build you want to see needs
   another publish. It also can't exercise the Worker — clean-URL
   301s, the legacy tool redirects and `POST /api/contact` are all

docs/audits/2026-10-reentry/dashboard/dashboard-extra.json

@@ -1,6 +1,6 @@
 {
- "mainSha": "a7d0d46",
- "lastMergeDate": "2026-10-06",
+ "mainSha": "a2ad39f",
+ "lastMergeDate": "2026-10-07",
  "banksAtOrAbove": [
   {
    "slug": "bacnet-basics",
@@ -44,33 +44,23 @@
    "to": "4.147.0"
   }
  ],
- "suiteBaseline": "**1223 passed, 0 failed, 1 skipped** in 8.3 min on `:9489`, 2026-10-06, against `main` @ `1bf23ce` (after the three chain A merges that had no CI run of their own); chain B and the lesson merged after it, each CI-green on `main`.",
+ "suiteBaseline": "**1242 passed, 0 failed, 1 skipped in 8.2 min on `:9661`, 2026-10-07, against the scratch merge `a3cef82` of all fifteen open PRs on `main` @ `a2ad39f`** (the owner-walk queue); the 2026-10-06 baseline on `main` @ `1bf23ce` was 1223/0/1.",
  "rulings": [
   "**All sixteen agenda items ruled 2026-10-05**, one by one with pros and cons; the owner reached the same call as the recommendation on every item before reading it. Record: `docs/audits/2026-10-reentry/rulings.md` (PR #611, merged).",
   "**The arc is picked: freeze season + quizzes.** Coil-freeze-protection lesson first, quiz wave 2 alongside, #228 in the background; F4 case-file mockup at the arc's end; §7.2 after that.",
   "**The no-coming-soon convention is amended:** the ban is on prominence. A low-key follow-up clause inside a scope note is acceptable — narrow it, don't delete it (CLAUDE.md carries the amendment).",
   "**The status grammar gained a `ruled` state:** decided-but-unshipped entries no longer count as decision-class. Decision-class fell from 6 to 2 (#228, #283).",
   "**PR #602** (Korean localization) closed courteously with the fork route; **#318** Speed Brain off; **#319** AI Crawl Control verified safe.",
   "**Review-day rulings (2026-10-06, rulings.md §18):** no Td on reheat coils as a teaching default (bank question replaced, D callout tempered); coil valves are two-way *or* three-way; the PICV sentence scoped to *modulating* valves; the ASHRAE 135 paraphrases verified (\"Clauses check out\"); the #228 design-note defaults accepted (\"I like your defaults\"); filter racks are drawn as pleated media site-wide.",
-  "**Owner on the format:** *\"Forcing both of us to take a look at each PR in sequence may be the most balanced workflow we've used.\"*"
+  "**Owner on the format:** *\"Forcing both of us to take a look at each PR in sequence may be the most balanced workflow we've used.\"*",
+  "**Evening rulings before the overnight lanes (2026-10-06, rulings.md §19):** the lesson adopts 35–38 °F (#99); #331 both triggers, merge on green; #328 narrow to scope statements; #324 reword as field advice; #91/#92/#94/#95/#96 fix, #93 no page change; hero badge → the Coil Freeze Protection lesson for the season; quiz wave 3 = six banks in two triples; F4 mockup on hold; #228 step 0 + step 1 as stacked PRs; combined LAN preview for the walk; the Workbench filter rack redrawn pleated."
  ],
  "today": [
-  "`main` pulled to `e33a825` (PR #601, 2026-09-02) — it had been two commits behind — and then to `178af5c` after the four PRs below merged.",
-  "Eight merged local branches deleted (four `worktree-agent-*` leftovers, four shipped feature branches); every lane worktree removed afterwards.",
-  "PR #603 **merged** — 12 codebase-issues headings and all 66 content-audit headings now carry the ledger's own status marker, so status reads from the heading line.",
-  "PR #604 **merged** — the state map, per-entry appendix and PR #602 assessment live under `docs/audits/2026-10-reentry/`; ledger #322 records the PR #602 ruling.",
-  "PR #605 **merged** — `npm run status`, the readout this page is rendered from.",
-  "Full Playwright suite run on a free high port as the session baseline: 1215 passed, 0 failed.",
-  "Cloudflare dashboard checked by the owner: Speed Brain turned off (#318); AI Crawl Control confirmed on with zero refused requests (#319).",
-  "PR #606 **merged** — six more ledger headings trued up to the marker grammar, and the two Cloudflare rulings recorded in the ledger.",
-  "PR #611 **merged** — every ruling written into its ledger entry; the durable record is `docs/audits/2026-10-reentry/rulings.md`.",
-  "PR #607 opened — the three dependency bumps plus the matching CI Playwright image, green; owner merges (Worker toolchain).",
-  "Night 1: sixteen cleanup lanes, three quiz banks, the coil-freeze lesson, the two BACnet softenings, the PID aside and the #228 design note all ran as Opus/Fable lanes with adversarial verification; everything that reaches a live page is a PR for the owner, listed under **Pick up here**.",
-  "Three merge chains were pre-built so nothing conflicts on merge: the ledger-tail chain #619 → #632 → #626 → #630 → #634 (each appends a new ledger entry), and the version/bank chain ending in the lesson PR #620 (see each PR’s *Stacked on*). Merge top-down; GitHub retargets the child when the parent lands.",
-  "**Day 2 (2026-10-06): the whole queue merged, PR by PR, with the owner.** 25 PRs landed — the 23 from night 1 plus #635 (two lesson sentences from the review) and #636 (Air Handlers filter racks). **The Coil Freeze Protection lesson is live** at forced-air step 3; `main` is v3.91.0; zero open PRs.",
-  "Before the walk: two PRs had gone CONFLICTING overnight (adjacent-hunk ledger collisions) and chain B would have on first merge — all three fixed by rebase; a stray duplicate ledger entry on #625 was dropped.",
-  "Process gap caught mid-session: a stacked PR gets **no CI run** when GitHub retargets it to `main`. Three chain A PRs merged that way; a full suite on `main` was clean (1223/0/1), and the chain B children were rebase-pushed before merging. Logged as codebase-issues #331 (owner picks the trigger fix) and as a CLAUDE.md Workflow rule.",
-  "Wrap-up docs PR #637 (rulings §18, #228 accepted, #90 closed, new #99 / #330 / #331, CLAUDE.md notes) merges on green. LAN preview back on `main`."
+  "**Overnight 2026-10-06 → 07 (autonomous):** the handoff was verified as a workflow first (140 claims: 127 verified, 12 corrected, 1 unverifiable — counts and framings carried forward), then five sequential workflows ran: nine small-fix lanes, two quiz triples, #228 steps 0–1, and the wrap-up.",
+  "**Merged on the owner's grant:** #645 (CI: `edited` gate + `push: main`, #331) and #646 (preview-script guard, #330). The first two `push: main` suite runs both passed — `main` had never been tested directly before.",
+  "**Fifteen PRs wait for the walk:** #639–#644 and #647 (small content fixes, the hero badge, the Workbench filter rack), #648/#649/#650 and #653/#654/#655 (wave 3, all six banks to 15), #651 → #652 (#228 steps 0 and 1, stacked). Zero conflicting pairs in the merge-tree matrix; the scratch merge of all fifteen passed the full suite (see *Repo and toolchain*) and is published at cfdev.home.arpa.",
+  "Walk notes with review pointers: `docs/audits/2026-10-reentry/walk-2026-10-07.md`. Ledger by-catch from the night: codebase-issues #332–#337, content-audit #100–#104 (docs PR, merged on green).",
+  "Every lane reported at least one drifted line cite in the entry it closed; CLAUDE.md's Workflow and LAN-preview bullets trued up for #645/#646, the hero-badge pointer relabelled `Featured`."
  ],
  "agenda": [
   {
@@ -282,4 +272,4 @@
   "**#633** — arc pick recorded (rulings.md §17), the two skipped annotations, coming-soon scan widened to the \"its own page\" shapes.",
   "**Day 2, merged with the owner (2026-10-06):** #607 deps · #608 #609 #610 quiz banks · #613 author/breadcrumb · #615 BACnet softenings · #616 PID aside · #617 FAQ rationale · #618 AHU hedge · #619 → #632 → #626 → #630 → #634 (ledger-tail chain) · #621 balancing · #622 narrowed clauses · #625 subhead ids · #631 design note · #612 → #623 → #624 → #627 → #620 (version chain, the lesson) · #635 lesson sentences · #636 filter racks."
  ]
-}
+}

docs/audits/2026-10-reentry/rulings.md

@@ -265,3 +265,84 @@ sequence may be the most balanced workflow we've used."*
 **Merged 2026-10-06:** #607 #608 #609 #610 #612 #613 #615 #616 #617
 #618 #619 #620 #621 #622 #623 #624 #625 #626 #627 #630 #631 #632 #634
 #635 #636. Open PRs after: none. `main` at 3.91.0.
+
+## 19. Evening rulings before the overnight lanes *(2026-10-06 — frontloaded in the first two hours, then the assistant worked the queue alone)*
+
+The session opened with `/verify-handoff` on `docs/next-session-handoff.md`
+(140 claims: 127 verified, 12 corrected, 1 unverifiable — the corrections
+are in the handoff's *Corrections* block) and then put every open call to
+the owner in four rounds. He answered each one, then slept; the lanes ran
+overnight and their PRs wait for the morning walk. Where a
+recommendation was offered, he took it on every item but two (marked).
+
+1. **content-audit #99 — the lesson adopts 35–38 °F** as the practical
+   freezestat band (*What Actually Freezes* and protection-stack item 1 in
+   `coil-freeze-protection.html`); the tool's `bandTxt` stays. *Against
+   the recommendation*, which had been to relabel the tool. PR #639.
+2. **codebase-issues #331 — both triggers**: `pull_request` listens for
+   `edited` with a base-change gate, AND `push: branches: [main]` runs the
+   suite on `main` after every merge. **Merge on green granted** (the
+   recommendation had been to hold it for the walk). PR #645, merged
+   2026-10-07 01:14Z; the first `push` runs on `main` followed immediately.
+3. **codebase-issues #328 — narrow to scope statements.** All five
+   *What this page didn't cover* bullets on `bacnet-networking.html`
+   (segmentation, BACnet/SC, capture-driven troubleshooting, alarms and
+   events, vendor profiles) plus `bacnet-basics.html`'s "their own future
+   pages" line. The topics stay listed; the promise goes. PR #643.
+4. **codebase-issues #324 — reword as field advice.** The tuner's
+   cheat-sheet row now says to flip direct ↔ reverse on the real
+   controller, not on this page. PR #641.
+5. **content-audit #91 + #92 — fix, two sentences.** One prose pointer
+   near the sensor-strip widget to the freeze lesson's 38 °F figure, and
+   one sentence giving the MA-T-reads-the-coil point in prose (the story
+   itself stays in the widget). PR #640.
+6. **content-audit #94 + #95 — fix.** `pid-basics.html` softens "fast,
+   clean loop" (duct static is noisy: D has little lag to anticipate and
+   plenty of noise to amplify) and names the measured variable; the
+   tuner's Fast row names duct static pressure and mixed-air temperature
+   as the PVs, with the outputs they drive. PR #641 (with #324 and #329).
+7. **content-audit #96 — fix, one disclosure sentence.** "Lockout" is used
+   at both ends of changeover — high-limit lockout of free cooling, and a
+   low-ambient / low-limit lockout at the cold end. Name both, legislate
+   neither (the deadband-convention register). PR #642.
+8. **content-audit #93 — no page change.** The damper-stroke caveat is a
+   candidate question for a later air-handlers wave; the entry stays open
+   as that pointer.
+9. **Home hero badge — feature the Coil Freeze Protection lesson** for
+   freezestat season; revert to the DDC Workbench in spring (the old
+   anchor is preserved in an HTML comment beside the badge). PR #644.
+10. **Quiz wave 3 — six banks, two batches of three.** `building-pressure`,
+    `duct-static-control`, `vav-systems`, then `psychrometrics-basics`,
+    `hydronic-loops`, `load-piping` — freeze-season neighbours of the new
+    lesson first. Owner merges each bank PR.
+11. **F4 case-file mockup — hold** until wave 3 is further along.
+12. **#228 — step 0 + step 1 as stacked PRs.** Engine helpers
+    (`Psychro.mixAir` / `mixFraction`) + `tests/psychro-mixair.spec.js` +
+    the minor bump first; the psychrometric-chart page with its
+    before/after fixture stacked on it. Owner merges both; the child is
+    rebase-pushed after the parent lands (or rides #331's fix).
+13. **Morning walk — combined LAN preview.** Every live-page PR merged
+    into one scratch tree, published to `cfdev.home.arpa`, with full-page
+    screenshots (dark + light) in the walk notes.
+14. **DDC Workbench filter rack — redraw it tonight** with the pleated
+    idiom (the handoff verification found the Workbench's `#ahu-filter`
+    still drew the old parallel-line rack; §18.6's "site-wide" had
+    covered the three lesson racks only). PR #647.
+15. **Session shape.** Owner: *"use Sonnet and Opus agents for mechanical
+    tasks, you judge if Opus is needed based on complexity. But use other
+    Fable agents for design tasks … let's try to frontload that into the
+    first 2 hours or so and then I'll want you to work after I go to
+    sleep."* Merge authority overnight: #330, #331 and the wrap-up docs PR
+    only; everything that reaches a live page waits for the walk.
+
+**Overnight output (for the morning walk, in order):** #639 freeze band · #640 air-handlers pointer · #641 pid loops · #642 economizer lockout · #643 bacnet scope · #644 hero badge · #647 workbench filter rack · #650 building-pressure · #648 duct-static-control · #649 vav-systems · #651 psychro step 0 · #652 psych chart step 1 (stacked on #651) · #655 psychrometrics-basics · #654 hydronic-loops · #653 load-piping. Merged on the grant: #645 (#331), #646 (#330). Walk notes: `docs/audits/2026-10-reentry/walk-2026-10-07.md`.
+
+**Process notes from the night.** (a) PR #645's `push: main` arm ran on
+its own merge commit and on #646's and passed both times — the first
+suite runs `main` has ever had. (b) A PR body edit now leaves a SKIPPED
+`test` check beside the real one (codebase-issues #331 resolution). (c)
+The quiz workflow's second refute round can leave an open item that the
+Opus fixer then applies without a third refute; both wave-3 batch-1
+open items were applied verbatim and are called out in the walk notes.
+(d) Every lane reported at least one drifted line cite in the entry it
+was fixing (the 2026-08-12 lesson, again).