Repository navigation
Audit: the LibreOffice reference for the #1220 deck seats page 8's value tick labels 7.68pt right of the native export #1860
Description
Summary
With #1663's horizontal calibration in place, page 8's value tick labels land
where the native PowerPoint export puts them, and the LibreOffice Impress
reference attached to #1220 is revealed to seat that label column 7.68pt to the
right of native while placing the plot's own edges in the same place. Every
layout-audit dx finding on 800..0 on this page therefore carries a
reference-side component that no converter change can remove.
This is the horizontal counterpart of #1676, which measures the same
reference's own vertical plot seat.
Measured
Fixture: page 8 of GENERAL SERVICES.pptx (#1220, input SHA-256
17924ec3b27646a2c1b2bb711b845360b713cc7052b3cdde9df752f9ceded3a9). Chart frame
689.187 x 221.058pt at (134.813, 270.0), value axis Arial 11.97pt, labels
0/200/400/600/800.
Ink extent of the 800 label, measured on 150 DPI pdftoppm renders (one pixel
is 0.48pt):
| native PowerPoint 16.113.1 | LibreOffice Impress 26.2.5.2 | office2pdf at #1663 | |
|---|---|---|---|
| label ink left | 142.000 | 149.680 | 142.000 |
| label ink right | 161.200 | 168.400 | 161.200 |
| plot left edge | 171.575 | 171.468 | 171.583 |
The two exporters agree on the plot's left edge to 0.107pt and disagree on where
the label column ends by 7.20pt, so the reference is not reserving a different
gutter — it is right-aligning the labels inside the one it reserves at its own
position.
The reference PDF is SHA-256
8c8f471d8baaf0abd79e752ef49951c240752d36e4512651f6cb9a7cfffb650f; the native
export is adec062605a45f1b6884252e9f4e199dee2da029e975845c258f4a7c38c820e1.
Effect on the audit
compare_layout.py --page 8 --audit --fine-shift 0.5 against the reference
reports 0 dx -7.34pt and 200..800 dx -7.30pt. Before #1663 those read
+0.98 to +1.01pt, because the uncalibrated gutter happened to put our labels
near the reference's rather than near PowerPoint's.
Expected
Recorded as a reference-side difference for this page, in the same way #1676
records the reference's own vertical plot seat, #1670 its right-legend layout
and #1664 its font substitutions.
Activity
fix(charts): measure a PowerPoint column plot's horizontal rectangle
docs(charts): record page 8's horizontal plot evidence
docs(pptx): publish the page-8 reference-exporter manifest
docs(pptx): publish the page-8 legend reference-exporter manifest
docs(evidence): record the composited axis labels against the page-8 GT
developer0hye commentedon Oct 6, 2026
Verified the page-8 horizontal tick-label reference difference against the fresh current-main output and the native PowerPoint export.
17924ec3b27646a2c1b2bb711b845360b713cc7052b3cdde9df752f9ceded3a98c8f471d8baaf0abd79e752ef49951c240752d36e4512651f6cb9a7cfffb650f6a18edd852804ae4cfd5fe10438b25fbbe4a140871b1506eacee2f7bc134bbc8e019d103c32a17268128401cb2d33d620ceccf33; PDF SHA-25647ddfd4ffaa608b27604cc4f987a923aef546ec4c4955bb530823bef57e99ff8; all three PDFs contain 15 pages at 960 x 540 pt.The fresh page-8 layout audit reports these reference-to-current-output shifts:
8006004002000At 150 DPI, the current-output tick-label crop aligns horizontally with the native PowerPoint image. For
800, the already-recorded ink starts are 142.000 pt in native PowerPoint and office2pdf versus 149.680 pt in the LibreOffice reference; the plot left edges remain within 0.12 pt across the three exports. This confirms that the supplied reference seats the value-label column to the right of the native/current output.The text-layer check also exposed a reference-side footer difference already tracked by #1664: every normalized word token from the LibreOffice extraction is present in the current output, while the output additionally extracts
Internal; the LibreOffice page image wraps that footer word onto a second line. The400token's extraction order differs around the legend, with no token loss. The native PDF rasterizes most page-8 chart/table/footer text, so it is not a text-layer oracle for those labels.The updated issue-local manifest records exact dispositions for all five tick-label shifts and the three tick-label render clusters. The fresh strict page-8 cluster report passes with 100/100 clusters dispositioned. The before and after evidence images are byte-identical because this PR records a verified reference difference without changing converter output.