test(scripts): batch 10f — six-banner battery roster for pm/bare-root-worklist's self-test

[baozhoutao]

Part of #13799

Batch 10f of the assertion-floor programme: scripts/pm/bare-root-worklist.mjs
takes the Tier B multi-battery roster over the SIX named section banners of its
selfTest().

failures.length === 0 was this self-test's only success condition, so "every
case held" and "the cases never ran" printed the same verdict line. The roster
closes that: what is pinned is the registered NAMES, not a number.

Shape

• SELF_TEST_BATTERIES — six keys, each byte-equal to the box-drawing
  banner it stands for (checked mechanically, all six match, six openers, one
  per banner).
• SELF_TEST_BATTERY_FLOOR = 6 — deleting an entry silences that battery's
  floor as effectively as zeroing it, so the roster's own size is pinned.
• UNATTRIBUTED_BATTERY — a case registered with no battery open reds by set
  difference rather than inflating whichever battery ran last.
• battery(name) opens a section; registerCase() is called from the block
  body of the existing t(); the floor is evaluated at this body's own verdict
  site, immediately above the success line, into the same failures sink.
• The seven cases that run BEFORE the first banner have no section head of
  their own, so the FIRST battery's opener is hoisted to the top of the body and
  they are attributed to it — stated at the banner as well. No comment was
  promoted to a section head.

The landed recipe refuses a floor placed before a plain return. That hole is
not open here and the reason is mechanical: this body's only exit a caller
accepts is the SELF_TEST_VERDICT sentinel returned BELOW the success line, and
the dispatch refuses anything else — so an early return reds at the HANDSHAKE
while a section that runs to the bottom with its cases gone reds at the FLOOR.
Ablation (e) below measures exactly that split.

Floors: MEASURED on a run, never planned from static sites

All six were pinned at 9999 first; one run produced exactly six breach lines
(counted before any number was read), and those numbers are the floors.

#	battery (= banner)	static t( sites	runtime cases	why they differ
1	The FOLD: one row per literal, however many invocations reach it	7 + 7 pre-banner	14	—
2	The triage coupling, both directions	6	6	—
3	The MECHANISM a repaired reason turns on, held mechanically	4	4	—
4	The RECORDED SPELLINGS, pinned: LIVENESS and PRECISION	9	65	4 sites inside the loop over 15 distinct spellings: 5 + 4x15
5	The DECLARATION a row describes, read from the gate's own SOURCE	7	43	4 sites inside the loop over the 10 DECLARED-NARROWER rows: 3 + 4x10
6	CENSUS_REFUSE_WIDE, pinned on its OWN terms (#14695)	5	5	—
	total	45	137

Both loop-driven counts are driven by this file's OWN tables (TRIAGE,
SPELLINGS), not by the live tree, so re-measuring a floor is the same edit
that changes a row — the discipline the file already states for its rows.

No behaviour change

• No assertion condition, message or argument changed. Deletions are the sink
  and verdict lines only: t() gains a registerCase() call in its block body,
  and the FAILURE line gains " (cases and floor)", the landed spelling.
• The green line is byte-identical to the installed base, stdout and stderr
  both, cmp-clean.
• Normal mode (which reads live PM state) is byte-identical too, stdout and
  stderr, over 134 lines of output — with a base-vs-base control run first to
  prove the comparison is a measurement and not two empty streams: there is no
  varying line in this tool's normal-mode output.

Ablations

Every mutation proven on disk (anchor count before and after, plus the blob
hash) before its run was read; every restore by git checkout HEAD -- ABSOLUTE PATH proven by blob-hash equality with the HEAD blob AND an empty
git diff HEAD; the driver carries trap ... EXIT INT TERM. No build or dist
is involved — the script is run directly by node.

#	mutation	result
a	one battery opener renamed	BOTH halves: the declared battery "DID NOT RUN — 0 cases registered, 6 pinned" and the renamed one "registered 6 case(s) but is not declared" — 3 failures
b	registerCase() removed from t()	ALL SIX report DID NOT RUN with 0 registered; every case still green (7 failures = 6 breaches + the remedy line, no case failure)
c	one banner's section skipped by a forced if (false) guard	ONLY that battery: CENSUS_REFUSE_WIDE DID NOT RUN, 0 of 5; the other five silent
d1	SELF_TEST_BATTERY_FLOOR pinned at 9999	"SELF_TEST_BATTERIES declares 6 batteries, below the pinned 9999"
d2	one roster entry DELETED	both halves of the size pin: "declares 5 batteries, below the pinned 6" AND the orphaned battery's 4 cases reported as not declared
e	early return after the first case	NO floor line at all — the run reds at the #13798 HANDSHAKE ("selfTest() returned without reaching its verdict"), exit 1. Handshake and floor stay orthogonal: the handshake owns early exits, the floor owns sections that stop registering
f	the floor deleted, with (b) applied	EXIT 0 and the full green line printed with ZERO cases registered — the vacuous green this batch exists to make impossible. Restored

Gates

Exit codes captured by redirect before any pipe.

Derived family, node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (re-derived after merging current main, so the
answer is not from a stale tree): 22 commands, ALL GREEN.

check:pm-dispatch-gates ran in the foreground under the shared verify lock
(scripts/pm/os-verify-lock.sh), five attempts across a busy queue, then
VERDICT command-exit 0 · held the lock 375s (6m15s) · waited 127s, with the
gate's own line: ✓ dispatch-gates self-test: 1392 cases pass. The four earlier
attempts returned the lock's exit 99, which is NOT MEASURED — neither a red nor
part of any green count.

Always-runs tail: check-self-test-workflow-commands.mjs and its --self-test,
check:declared-population-live, check-self-test-wired.mjs (+ --self-test),
check:ratchet-remedy-authority, check:nul-bytes, check:watch-hint-literal
— all exit 0. Verdict lines, quoted:

• check-nul-bytes: OK (scanned 7465 text file(s) -- 7465 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes).
• OK check-ratchet-remedy-authority: 202 scripts swept ...
• check:declared-population-live — 201 of 252 famil(ies) declare a path population, and every one of them reaches this tree's 7472 tracked file(s).

The file's own --self-test: exit 0, green line byte-identical to base.

Instrument, before and after (node scripts/measure-self-test-floor.mjs --json): 170 records before, 170 after; exactly ONE record moved —
scripts/pm/bare-root-worklist.mjs floor: NONE to floor: ROSTER. Nothing
entered the population and nothing else left it.

extractWatchHints over this file: [] before, [] after — identical, with a
positive control in the same session (check-cli-command-ids.mjs returns four
hints, check-nul-bytes.mjs two), so the empty answer is a measurement rather
than a silent extractor. This file still declares no population of its own.

Whole-repo pnpm lint (eslint . --no-inline-config): exit 0 in 88s. No
narrowing was needed or claimed.

Gate union run at 7f1242569.

skip-changeset: this PR publishes nothing from any package — the only changed
path is scripts/pm/bare-root-worklist.mjs.

🤖 Generated with Claude Code

https://claude.ai/code/session_012zGPuVVX3deAx9LdjK8jCk

---

Generated by Claude Code