Skip to content

Fix historical Reel Retina backfill and spotlight highlighting - #42

Open
Happy (huangyangyu) wants to merge 5 commits into
microsoft:mainfrom
FacePerceiver:fix/reel-history-layer
Open

Fix historical Reel Retina backfill and spotlight highlighting#42
Happy (huangyangyu) wants to merge 5 commits into
microsoft:mainfrom
FacePerceiver:fix/reel-history-layer

Conversation

@huangyangyu

@huangyangyu Happy (huangyangyu) commented Aug 18, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Preserve the canonical DPR1 poster PNG and historical poster HTML while generating quality-gated 2x/3x raster sources from the canonical PDF.
  • Install Retina assets and rewritten HTML transactionally, and enable the fallback only for one uniquely identified historical layer inside the marked pixel host.
  • Restore historical raster highlighting as a neutral 2px white spotlight with the surrounding poster faded; native HTML posters retain the existing DOM highlight path.
  • Scope screenshot/pixel highlight QA exclusively to the Reel-owned historical-raster fallback. Native poster HTML is checked only for DOM highlight activation and cleanup; its durable pixel contract remains owned by paper2poster's exact warm/cold decoded-RGBA generation gate.
  • Harden historical package QA for srcset/currentSrc, Retina rendering, proxy visibility and pixel effects, focus-core preservation, exact historical-proxy restoration, duplicate IDs, and failed-install recovery.

Retina sources are enabled only when PDF-derived output passes MAE <= 6, RMS <= 14, and maximum 32x32 tile RMS <= 30.

Validation

  • Full public Daily Paper catalog repair: 55/55 unique Reels passed HTTP and file:// QA with 0 errors and 0 warnings.
  • Public route reconciliation: 55/55 Reel pages and 55/55 nested posters returned HTTP 200; all 165 DPR1/2/3 raster URLs returned HTTP 200 with the expected PNG size.
  • Every catalog entry passed all 8 artifact gates and exposed the neutral spotlight plus high-DPR image-rendering: auto runtime, with a unique scoped historical layer and canonical 1x src plus 1x/2x/3x srcset.
  • Canonical DPR1 PNG and PDF hashes remained unchanged for all 55 papers.
  • Targeted production verification for 2608.10915 and 2608.12307: 0 errors, 0 warnings; public DPR1/2/3 and spotlight behavior verified.
  • Real 2608.06867 HTTP/file QA at DPR1/2/3: 0 errors.
  • Real 2608.10299 full HTTP/file QA: 0 errors, 0 warnings.
  • Native regression 2608.15089: the former mouseleave gate rejected 11 changed pixels at maximum channel delta 1; after scoping native QA to DOM lifecycle, full HTTP/file QA passes with 0 errors and 0 warnings.
  • Additional real native HTML posters 2608.08311 and 2608.09802: full HTTP/file QA passes with 0 errors and 0 warnings.
  • Historical-raster 2608.10915: final full HTTP/file QA still passes with 0 errors and 0 warnings after the native-scope correction.
  • Negative coverage passed for localized corruption, wrong/missing PDF, oversized images, partial generation, transaction rollback, duplicate IDs, and full-focus washout.
  • py_compile, both CLI help smoke tests, and git diff --check passed.
  • Fresh end-to-end ResearchStudio-Reel rebuild of native 2608.15089 from the complete paper/poster/blog/video bundle: regenerated 10 section clips and 12 slide clips, validated on-demand delivery plus HTTP and file:// browser gates with 0 errors and 0 warnings, and preserved the canonical poster.html and poster.png SHA-256 hashes.

No test files, screenshots, reports, generated QA artifacts, or production bundles are included in this PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants