[finding] measure-self-test-floor's --probe has zero coverage of the helper handshake spelling — its only carrier is an ENTRY_BY_HAND null

[claude]

Measured while landing #15324 (the probe's HELD-needs-a-printed-handshake fix) on c4d1354e3. Filed unassigned, as a finding only — no severity asserted, grading is triage's.

The reading

--probe in scripts/measure-self-test-floor.mjs has zero dynamic coverage of one of the three landed handshake spellings. Measured on c4d1354e3, whole population, 170 rows:

handshake spelling	carriers under scripts/	probe verdict coverage
sentinel — SELF_TEST_VERDICT	87	exercised (many HELD rows)
flag — selfTestReachedVerdict	78	exercised (many HELD rows)
helper — requireReachedVerdict	1	none

The helper form's only carrier in the tree is scripts/check-platform-checklist.mjs, and that file carries an ENTRY_BY_HAND row of null, so every probe run records:

{"verdict":"NOT MEASURED","why":"entry read by hand as not probeable -- see ENTRY_BY_HAND"}

The null is correct and deliberate — its dispatch calls four self-test functions and combines their statuses, so a one-function early return measures a sub-battery and would read a downstream crash as a handshake. Nothing here asks for that row to change.

Why it is worth recording

The consequence is about what a green sweep is evidence FOR. --probe is quoted as the completion picture for #13798/#13799, and for the helper spelling it says nothing at all — not "held", not "defeated", but no reading in either direction. So:

• no probe run in this tree has ever demonstrated that the helper form actually refuses when its self-test leaves early. It is believed to work by reading, exactly as the two forms with 87 and 78 carriers are not;
• if the helper form is ever adopted more widely (it exists precisely because ten inlined copies is ten drift sites — the ruling recorded in No canonical way to ask whether a self-test has a verdict handshake — the repair landed in three spellings and measure-self-test-floor reports none of them #14968), the population it covers grows while the probe's coverage of it stays at zero.

The instrument's inline controls do exercise a sentinel-shaped fixture (SOUND_GATE) in both directions, so the sentinel shape has a permanent control. The helper shape has neither a control fixture nor a live row.

Options

• A. Add a helper-shaped fixture to the instrument's inline controls. Cheapest, and it follows the file's own convention (controls run on every invocation, never behind a flag). It proves the probe can read that shape, without needing a probeable carrier in the tree. Cost: a third fixture string.
• B. Make check-platform-checklist.mjs probeable — e.g. teach ENTRY_BY_HAND to name a list of entries and combine their probes. Real coverage of the real file; considerably more instrument. Overlaps The self-test-floor census cannot measure check-workspace-manifest-cycles.mjs - two self-test-shaped functions and no ENTRY_BY_HAND row, so its handshake reads NOT MEASURED #14842's shape question.
• C. Record the gap in the instrument's header and leave the coverage at zero. Zero cost, and the NOT MEASURED row is already honest; but "which spellings has this sweep actually exercised" stays a question you answer by reading.

Recommendation: A. The claim a control can support here is "the probe reads this shape", and a fixture supports it without waiting on B; C leaves the reader to re-derive this table by hand, which is the recurring cost #14968 is about.

Related

#14968 (no canonical way to ask whether a self-test has a handshake; the three-spelling table), #14842 (two self-test-shaped functions and the ENTRY_BY_HAND row), #14963 (the probe anchors on the first selfTest match), #15324 (this probe's verdict).

Duplicate-checked before filing against 380 open issues read from the repo-scoped issues list plus a local grep; control terms returned the known family (#14842 / #14963 / #14968 / #15296 / #15339), confirming a live read. No existing card asks about per-spelling probe coverage.

---

Generated by Claude Code