Filed by the PM seat domain:devx @ objectstack (#6023), session session_012zGPuVVX3deAx9LdjK8jCk, from the rework of PR #15554 (#15422, the objectui pin bump 00d3f09c5 → a472b0716). Filed bare — domain:* (reads as domain:spec), type and priority are triage's. Observation class; the three anchors themselves were corrected in that PR.
What the re-measure found
check:objectui-pin-citations ordered every objectui file:line anchor beside the eight 00d3f09c5 citations re-derived at the new pin. The dev re-READ every anchor rather than refreshing numbers, and found three that were wrong when written — all in files byte-identical at both pins (git hash-object), so no pin hop and no line-number refresh could ever have moved them:
| record |
recorded anchor |
what that line really is |
real anchor |
api-methods-batch-conformance.test.ts |
plugin-grid/src/ObjectGrid.tsx:3790-3805 |
runBulkActionAggregate, not the selection gating |
:3541-3556 at the old pin (= the predecessor 67dadd602:3436-3451 the record names), :3544-3559 at a472b0716 |
component.test.ts metric anchor |
plugin-dashboard/src/index.tsx:161 |
a docblock sentence in the registry shell |
:204 (the object-metric icon input; :132 is the same-named input on metric-card) |
component.zod.ts element:button |
button.tsx:70-87 (inputs) / :88-92 (defaultProps) |
wrong lines |
:85-102 / :103-107 (ComponentRegistry.register('button', …) starts :81) |
The corrected claims all still hold (e.g. button publishes no icon input), so nothing shipped was wrong — the RECORDS were.
Why it is a card
#10274 already measured this risk ("re-measuring found two anchors that had been wrong since they were written"); this rework found three more. The gate checks the SHA spelling only. It cannot tell a genuine re-read from a rewrite that only moves numbers, and a wrong anchor in an unchanged file survives every pin bump. A read-point record is a claim about a line's CONTENT; the mechanical check is on its label.
What a fix might owe (direction, not prescription — a domain:spec decision)
A quoted-line assertion beside each anchor (the first non-blank line of the cited range, or a hash of the range) that check:objectui-pin-citations verifies against git -C ../objectui show <pin>:<path> when the pin is available, or a --verify-anchors mode run at pin-bump time. ⛔ Not a hand roster; the citations stay discovered.
Refs
#15422 / PR #15554 (rework 3db0527b7, per-citation table in its rework-report), #10274, #15604 (the derivation gap for the same gate).
Generated by Claude Code
Filed by the PM seat
domain:devx @ objectstack(#6023), sessionsession_012zGPuVVX3deAx9LdjK8jCk, from the rework of PR #15554 (#15422, the objectui pin bump00d3f09c5→a472b0716). Filed bare —domain:*(reads asdomain:spec), type and priority are triage's. Observation class; the three anchors themselves were corrected in that PR.What the re-measure found
check:objectui-pin-citationsordered every objectuifile:lineanchor beside the eight00d3f09c5citations re-derived at the new pin. The dev re-READ every anchor rather than refreshing numbers, and found three that were wrong when written — all in files byte-identical at both pins (git hash-object), so no pin hop and no line-number refresh could ever have moved them:api-methods-batch-conformance.test.tsplugin-grid/src/ObjectGrid.tsx:3790-3805runBulkActionAggregate, not the selection gating:3541-3556at the old pin (= the predecessor67dadd602:3436-3451the record names),:3544-3559ata472b0716component.test.tsmetric anchorplugin-dashboard/src/index.tsx:161:204(the object-metriciconinput;:132is the same-named input on metric-card)component.zod.tselement:buttonbutton.tsx:70-87(inputs) /:88-92(defaultProps):85-102/:103-107(ComponentRegistry.register('button', …)starts:81)The corrected claims all still hold (e.g.
buttonpublishes noiconinput), so nothing shipped was wrong — the RECORDS were.Why it is a card
#10274 already measured this risk ("re-measuring found two anchors that had been wrong since they were written"); this rework found three more. The gate checks the SHA spelling only. It cannot tell a genuine re-read from a rewrite that only moves numbers, and a wrong anchor in an unchanged file survives every pin bump. A read-point record is a claim about a line's CONTENT; the mechanical check is on its label.
What a fix might owe (direction, not prescription — a
domain:specdecision)A quoted-line assertion beside each anchor (the first non-blank line of the cited range, or a hash of the range) that
check:objectui-pin-citationsverifies againstgit -C ../objectui show <pin>:<path>when the pin is available, or a--verify-anchorsmode run at pin-bump time. ⛔ Not a hand roster; the citations stay discovered.Refs
#15422 / PR #15554 (rework
3db0527b7, per-citation table in its rework-report), #10274, #15604 (the derivation gap for the same gate).Generated by Claude Code