fix(scripts): give 24 more self-tests a verdict handshake at their dispatch (#13798 batch 2)

[claude]

Part of #13798 (batch 2). Batch 1 is PR #14479, merged.

node scripts/measure-self-test-floor.mjs --probe injects return; as the first
statement of the function each --self-test dispatch calls, runs the gate and reads
the exit code before any pipe. Exit 0 is the defect: a self-test that never finished,
reported as one that passed.

What this batch does

The same boundary-only handshake batch 1 landed, transplanted to 24 more files. Two
shapes, one per dispatch spelling, both taken verbatim from the PR #13797 precedent:

• sentinel (13 files) — selfTest() returns SELF_TEST_VERDICT only after its
  verdict line, and the dispatch refuses anything else. Used where the dispatch discards
  the return value. 10 files take it for the first time; the other 3 are the
  production-path additions below, on files that already carry the sentinel from batch 1.
• module-level selfTestReachedVerdict flag (11 files) — used where the
  self-test's own exit code is load-bearing (process.exit(selfTest()),
  return selfTest(), the ternary form) or where the self-test already ends in
  process.exit and an appended sentinel would be dead code.

Nothing inside a self-test body changes beyond the verdict marker on the line after
the success line. The census's hole-1 floor classification is identical for all 165
files before and after — the mechanical proof the remedy is boundary-only.

The probe, before and after

Measured on this branch's own merge base, not re-derived by hand.

	population	DEFEATED	HELD	NOT MEASURED
before (2d40f9146)	165	16	143	6
after	165	2	157	6

The 2 that remain are asserted equal as a set to the two files held by an open PR
and deliberately excluded here: scripts/check-type-check-coverage.mjs (PR #14805)
and scripts/pm/bare-root-worklist.mjs (PR #14800). Nothing else is left defeated.

14 of the 24 flipped DEFEATED -> HELD in the census. The other 10 are shapes the
census reads HELD or NOT MEASURED for reasons that are not a handshake, and each
carries its own evidence below.

The six that held only by accident

The card's own note: "5 gates hold today only by luck — a downstream TypeError or a
usage error, not a handshake." The ACCEPT ruled them into this batch (Q1 → A). The
census verdict does not move for these (HELD before, HELD after), so the evidence is
the first line the mutated run prints — the accident replaced by the named refusal:

file	mutated head, before	mutated head, after
check-osv-exemptions.mjs	file:///… (TypeError stack)	the named refusal
typecheck-configs.mjs	file:///… (TypeError stack)	the named refusal
check-exported-any-returns.mts	/home/… (TypeError stack)	the named refusal
checklist-select.mjs	usage: node scripts/checklist-select.mjs … (exit 2)	the named refusal
run-with-stall-guard.mjs	run-with-stall-guard: unknown option --self-test	the named refusal
check-page-declaration-shape.mjs	(printed nothing, exit 1)	the named refusal

check-osv-exemptions.mjs needed one extra line beyond the transplant, in its own
commit: its dispatch destructures the result (const { passed, lines } = selfTest()),
so on an early return the destructuring threw before the handshake could run — the
accident survived the first attempt, and the census said so. The result is now read
into a local first and destructured after the handshake.

Three traps this surface carries

1. Unbraced branch bodies. } else if (argv.includes('--self-test')) selfTest();
   followed by else run(); needs braces, or the trailing else re-binds.
2. The temporal dead zone. scripts/run-with-stall-guard.mjs dispatches at the top
   of the file, above the self-test definition; the flag is declared above the
   dispatch, not next to the function.
3. A never return type. check-exported-any-returns.mts declares
   function selfTest(): never. It takes the flag rather than the sentinel precisely so
   the annotation does not have to move — check-test-typecheck.mts in batch 1 is the
   precedent for how a sentinel return drifts a shrink-only DEBT ledger, and raising a
   ledger is maintainer-only.

The production-path half (ACCEPT Q2 → A)

check-filter-alias-parity.mjs, check-meta-type-normalized.mjs and
check-test-completeness.mjs also run their self-test on the production path and
discarded the result there — the same silence one seam over, and one the census never
drives because it only runs --self-test. Each now compares the sentinel on that path
too. Ablated by hand with the census's own injector, run with no --self-test on
the command line: baseline exit 0 / 0 / 3, mutated exit 1 / 1 / 1, each printing the
named refusal.

check-workspace-manifest-cycles.mjs is fixed but still NOT MEASURED

Its dispatch calls runSelfTest(), which wraps a selfTest() that returns a failure
list. The census picks an entry only when a file defines exactly one self-test-shaped
function or has an ENTRY_BY_HAND row; this file has two and no row, so it reads
NOT MEASURED (ambiguous entry) both before and after — the handshake is real, the
instrument just cannot see it.

Ablated by hand instead, with the census's own probeEarlyReturn:

• entry runSelfTest (what the dispatch actually calls): baseline exit 0, mutated
  exit 1, head ✗ check-workspace-manifest-cycles self-test: runSelfTest() returned without reaching its verdict, — HELD.
• entry selfTest (what the census would pick if the ambiguity were resolved the
  wrong way): mutated exit 1 with a TypeError stack — an accident, not a handshake.

⚠️ The one-line 'scripts/check-workspace-manifest-cycles.mjs': 'runSelfTest' row that
would make this measurable lives in scripts/measure-self-test-floor.mjs, which is
outside this batch's declared file surface. It is not edited here; it is filed as
#14842 instead, with the two readings that say which entry the row must name.

Local verification

origin/main merged once more (a real merge, no rebase) before opening this PR. The gate
union was re-derived AFTER the final commit and run on that head, 61850abbf:

• node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands,
  DG_EXIT 0, 24 paths vs merge base f3ae441fa, 47 families, every one run from
  --commands. Plus the dispatch-named check:declared-population-live, the
  any-edit check:nul-bytes, and scripts/pm/check-governed-queue-guard.mjs --self-test
  (that file is edited here and governed-surface-guard.yml runs it, but the derivation
  did not list it).
• 51 RAN-PASS, 1 NOT MEASURED. The one is
  node scripts/check-test-completeness.mjs — exit 3, PREREQUISITE NOT MET (it grades a
  saved turbo run test log and none exists locally). Never read as a pass.
  node scripts/check-engine-split-ratio.mjs first came back exit 2 on the shallow
  clone; after git fetch --shallow-since=2026-05-29 origin it is exit 0 with a real
  ratio, so it counts as RAN-PASS.
• pnpm check:pm-dispatch-gates ran to completion: dispatch-gates self-test: 1241 cases pass.
• pnpm build exit 0. pnpm lint over the WHOLE repo, not narrowed: exit 0, 5779 files,
  0 errors, 0 warnings (counts read from --format json).
• Every one of the 24 files runs its own --self-test to the SAME exit code and
  byte-identical output as before the change (captured before the codemod, again after,
  then diffed). One file differs on one nondeterministic field:
  release-rehearsal-clone.mjs prints an elapsed-milliseconds figure (115 ms vs
  102 ms).
• pnpm check:nul-bytes exit 0, plus a direct control-byte sweep of the touched files:
  no match.

Every exit code above was captured by redirecting first and reading $? after — never
through a pipe.

skip-changeset: scripts/** only, nothing published from any package.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LAwHpn4uVuf4N1geBcD5i3

---

Generated by Claude Code

[claude]

PM status (domain:devx execution seat, session session_01LAwHpn4uVuf4N1geBcD5i3): reviewed and ACCEPTED on #13798 (batch-2 checklist there). Held as draft on one red check that is not this PR's: Validate Package Dependencies on 61850abbf — run 33719242879, job 100534794466 — fails only at step 12 "Audit dependencies for known vulnerabilities (OSV-Scanner)"; step 11 (this PR's own check-osv-exemptions.mjs --self-test + gate) passed. The workflow ran here because the PR edits scripts/check-osv-exemptions.mjs (path filter), and the scanner reads the base's lockfile, which still carries the four fast-uri advisories (#14732). The fix is PR #14783 (fast-uri 3.1.5 → 3.1.7), already in the merge queue. Plan: once #14783 lands, merge origin/main into this branch via update-branch (real merge commit), let the check re-run, then flip ready + arm. No re-run is requested for the red itself — it is deterministic on the current base.

---

Generated by Claude Code

[claude]

PM action (domain:devx seat, session session_01LAwHpn4uVuf4N1geBcD5i3, closing shift): update-branch issued 10:3xZ — PR #14783 (the fast-uri lift that clears Validate Package Dependencies on the base) MERGED 08:38Z as b69da43d5. Next sitter: on GREEN flip ready + arm; any red on a shard this diff does not touch ⇒ read the anchor #14822 first.

---

Generated by Claude Code