You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Audit: the #1656 page-1 disposition for p1-fc6d6deb9fe7 records a refuted explanation that bbox reuse would copy forward #1858
assets/bugfixes/issue-1656/render-clusters-page-1.json disposes cluster p1-fc6d6deb9fe7 to #1660 with this note:
The cluster is the last glyph of a blocked cell against its column gridline: the
native export draws it whole and the output cuts it on the gridline (400% crop of
p1-fc6d6deb9fe7 reads rell against rel).
The explanation is refuted. Re-measured on main (1dc1ae29), the native Excel for
Mac export carries its own clip_path on the column gridline and paints the
overhanging glyph cut, exactly as we do — see the closing measurement on #1660.
What the crop actually caught is a whole-glyph count difference from the #1659
whole-point pacing drift, which is what every other cluster in that same report is
dispositioned to.
The cluster itself is gone: re-running that report's own census on main — compare_render.py tests/fixtures/xlsx/100-customers.xlsx output --page 1 --dpi 150,
the snapshot's 5% fuzz and 20pt² floor — yields 0 clusters against the 56 it
recorded. #1853 removed them.
Why this is worth a record rather than an edit
The report is a frozen per-PR evidence snapshot, the same status AGENTS.md gives before.jpg ("captures the pre-fix output"). Rewriting it would misrepresent what
that audit observed, and touching any assets/bugfixes/** cluster report makes scripts/check_visual_pr.py demand a full ## Visual audit with gt/before/after
evidence for a change that produces no pixels.
The risk is the documented re-audit shortcut: match new clusters to the committed
prior report by bounding box and hand-review only what moved. A cluster reappearing
near x 268.8, y 616.8, 6.24 x 9.12pt on that page would inherit a note whose
stated rule is wrong, and #1660 — the issue it points at — is closed.
Expected
A re-audit of 100-customers.xlsx page 1 re-derives that cluster's disposition from
the current measurement instead of reusing the note, or the note is refreshed as part
of the next audit that legitimately regenerates that report.
Actual
The note stands as committed, stating a rule the native export's own clip rectangle
refutes.
Completed the targeted re-audit in #2030. The current output PDF at assets/audits/issue-1858/current-output.pdf and its same-location crop were compared with the archived native Excel page; the historically identified glyph is intact, so the old cluster is absent from the current crop. The #1656 report remains frozen, and the audit rules now require deriving every disposition from current crops rather than inheriting notes by ID or bbox. The native PDF was unavailable, so this records no full-page cluster census or text-layer result.
Activity
developer0hye commentedon Oct 6, 2026
Completed the targeted re-audit in #2030. The current output PDF at assets/audits/issue-1858/current-output.pdf and its same-location crop were compared with the archived native Excel page; the historically identified glyph is intact, so the old cluster is absent from the current crop. The #1656 report remains frozen, and the audit rules now require deriving every disposition from current crops rather than inheriting notes by ID or bbox. The native PDF was unavailable, so this records no full-page cluster census or text-layer result.