Observed while landing #13799 batch 4 (PR #15198) — not fixed there, because it is a different defect class from that card's assertion floors and the batch is fenced to one file's self-test wiring.
The reading
scripts/partition-test-shards.mjs, the banner that opens the balancing pins:
// WRONG quantity satisfies every one of those assertions. These five pin the
// quantity and the outcome.
Six numbered pins follow it, not five:
| pin |
line |
what it pins |
| 1 |
523 |
the weight is TIME, not file count |
| 2 |
541 |
the committed dataset meets the acceptance bound at the CI shard count |
| 3 |
578 |
the bound is REACHABLE at this shard count |
| 4 |
596 |
the shard count in ci.yml agrees with SHARD_COUNT |
| 5 |
620 |
an unmeasured package is ESTIMATED, never silently free |
| 6 |
644 |
END-TO-END: a real run weighs a real package by duration |
Pin 6 was added after the banner was written and the count was not carried forward. Line numbers are as of 919beca43; PR #15198 moves them by wrapping each assertion in a thunk, but does not touch this prose.
Why it is worth a card rather than a silent fix
It is cosmetic today, but it is the same shape as the drift this file's own pin 4 exists to catch: a stated number that no longer matches the thing it counts, with nothing reading it. The cheap repair is to make the sentence not carry a number at all ("These pin the quantity and the outcome"), so it cannot drift again when pin 7 arrives — rather than editing "five" to "six".
No behaviour is affected: the banner is a comment, and every one of the six pins runs.
Generated by Claude Code
Observed while landing #13799 batch 4 (PR #15198) — not fixed there, because it is a different defect class from that card's assertion floors and the batch is fenced to one file's self-test wiring.
The reading
scripts/partition-test-shards.mjs, the banner that opens the balancing pins:Six numbered pins follow it, not five:
ci.ymlagrees withSHARD_COUNTPin 6 was added after the banner was written and the count was not carried forward. Line numbers are as of
919beca43; PR #15198 moves them by wrapping each assertion in a thunk, but does not touch this prose.Why it is worth a card rather than a silent fix
It is cosmetic today, but it is the same shape as the drift this file's own pin 4 exists to catch: a stated number that no longer matches the thing it counts, with nothing reading it. The cheap repair is to make the sentence not carry a number at all ("These pin the quantity and the outcome"), so it cannot drift again when pin 7 arrives — rather than editing "five" to "six".
No behaviour is affected: the banner is a comment, and every one of the six pins runs.
Generated by Claude Code