Filed unassigned by the domain:devx @ objectui PM seat as an out-of-scope observation returned by #6778's implementation (PR #7045, R39). Recording only — deliberately not fixed there, and the implementer was explicitly instructed not to ride it along.
Attribution, as prose because the footer block is stripped from issue bodies written through this path (see #6970): generated by Claude Code, PM session 86aefd79-6ef3-5a63-98f9-701da03c59ab.
The structural claim
scripts/check-eager-closure-budget.mjs is built on the principle that its reasoning is checked rather than argued. Every ceiling is weighed against the report the run just read, and evaluateHeadroomSensitivity exists precisely so that a claim about the gauge cannot go stale in silence.
The doc comments are the one part with no such guard. They are prose, nothing reads them, and nothing can tell when they stop describing the file they sit in.
The instance that proves it is not theoretical
#6778 measured, and PR #7045 repaired, a single paragraph on PER_CHUNK_BASELINE in which four separate statements had gone false:
- it named
4c1623c0c as what BASELINE carries — three re-baselines out of date;
- its "707,291 BELOW" arithmetic was computed against that retired reading;
- its conclusion was the reverse of the gate's live verdict — the paragraph told its reader the aggregate ceiling was ~787 KB wider than the regression it must catch (i.e. decorative) while the same script, in the same run, printed
headroom 42.2 KB = 0.47x the 89.0 KB regression (i.e. in range and working);
~787 KB was a units slip — 4,086,000 − 3,298,620 = 787,380 bytes, i.e. ~769 KB.
Point 3 is the sharp one: an author sizing a re-baseline off that paragraph would conclude the aggregate half is decorative at exactly the moment the tool is telling them it is working.
⚠️ Two further facts about how long this survived:
The cheap backstop that already has its hooks in place
VERDICT_CEILING_CONSTANTS already deliberately excludes BASELINE and PER_CHUNK_BASELINE — the exclusion is a decision the file has already made and documented. A unit test could turn "the prose names a retired commit" into a red test, along one or both of:
- (a) positive pin — each baseline's own commit string must appear in the doc block that cites it;
- (b) negative pin — that block must carry no other 9-hex-digit token.
⛔ The design question this card exists to settle — (b) is not safe as stated
(b) would fire on legitimate text. PR #7045 deliberately kept two past-tense citations of 4c1623c0c elsewhere in the file — one in the "Why this number has moved" narrative, one in the MAX_EAGER_CLOSURE_GZIP_BYTES re-baseline list — because as history they are correct and load-bearing. The re-baseline list is, by construction, a list of superseded hashes.
⇒ A rule that cannot tell "this is what the constant carries" (must be current) from "this is what it used to carry" (must not change) will either red the honest history or be scoped so narrowly it stops catching anything. That distinction is the card, not the regex.
Whoever takes this decides:
⛔ Fences for whoever takes it
Related
Filed unassigned by the
domain:devx @ objectuiPM seat as an out-of-scope observation returned by #6778's implementation (PR #7045, R39). Recording only — deliberately not fixed there, and the implementer was explicitly instructed not to ride it along.Attribution, as prose because the footer block is stripped from issue bodies written through this path (see #6970): generated by Claude Code, PM session
86aefd79-6ef3-5a63-98f9-701da03c59ab.The structural claim
scripts/check-eager-closure-budget.mjsis built on the principle that its reasoning is checked rather than argued. Every ceiling is weighed against the report the run just read, andevaluateHeadroomSensitivityexists precisely so that a claim about the gauge cannot go stale in silence.The doc comments are the one part with no such guard. They are prose, nothing reads them, and nothing can tell when they stop describing the file they sit in.
The instance that proves it is not theoretical
#6778 measured, and PR #7045 repaired, a single paragraph on
PER_CHUNK_BASELINEin which four separate statements had gone false:4c1623c0cas whatBASELINEcarries — three re-baselines out of date;headroom 42.2 KB = 0.47x the 89.0 KB regression(i.e. in range and working);~787 KBwas a units slip — 4,086,000 − 3,298,620 = 787,380 bytes, i.e. ~769 KB.Point 3 is the sharp one: an author sizing a re-baseline off that paragraph would conclude the aggregate half is decorative at exactly the moment the tool is telling them it is working.
PER_CHUNK_BASELINEprovenance comment argues from a reading two re-baselines old, and its conclusion contradicts the gate's own live sensitivity verdict #6778 namedbd2a7ec50/3_254_004; correct when filed on 2026-08-29, wrong by dispatch on 08-31 because finding(console): metadata-admin's 172,651 B eager closure is held by a REGISTRY CONTRACT, not by chunking — registerAppComponent takes a component value #6776 re-baselined again. ⇒ Even a card about stale citations could not stay current by hand.The cheap backstop that already has its hooks in place
VERDICT_CEILING_CONSTANTSalready deliberately excludesBASELINEandPER_CHUNK_BASELINE— the exclusion is a decision the file has already made and documented. A unit test could turn "the prose names a retired commit" into a red test, along one or both of:⛔ The design question this card exists to settle — (b) is not safe as stated
(b) would fire on legitimate text. PR #7045 deliberately kept two past-tense citations of
4c1623c0celsewhere in the file — one in the "Why this number has moved" narrative, one in theMAX_EAGER_CLOSURE_GZIP_BYTESre-baseline list — because as history they are correct and load-bearing. The re-baseline list is, by construction, a list of superseded hashes.⇒ A rule that cannot tell "this is what the constant carries" (must be current) from "this is what it used to carry" (must not change) will either red the honest history or be scoped so narrowly it stops catching anything. That distinction is the card, not the regex.
Whoever takes this decides:
PER_CHUNK_BASELINEprovenance comment argues from a reading two re-baselines old, and its conclusion contradicts the gate's own live sensitivity verdict #6778 defect (the false hash was in the adjacent block, so probably yes —⛔ Fences for whoever takes it
main. thePER_CHUNK_BASELINEprovenance comment argues from a reading two re-baselines old, and its conclusion contradicts the gate's own live sensitivity verdict #6778 demonstrated three times over that citations in this area rot within days.Related
PER_CHUNK_BASELINEprovenance comment argues from a reading two re-baselines old, and its conclusion contradicts the gate's own live sensitivity verdict #6778 / PR docs(scripts): re-derive the PER_CHUNK_BASELINE provenance against a live gate run #7045 — the instance; the repair, and the source of this observation.PER_CHUNK_BASELINEprovenance comment argues from a reading two re-baselines old, and its conclusion contradicts the gate's own live sensitivity verdict #6778, open on "the headroom moved and nobody could establish which side". This card is the mechanism by which that becomes unknowable: the constant's own documentation is the first place anyone looks.scripts/vite-declared-lazy-views.ts. ⭐ Two files, same week, same defect — which is the argument that this wants a mechanism rather than another hand fix.