lllin000_PaperForge/project/current/ocr-v2-active-queue.md
2026-07-14 19:04:20 +08:00

2.6 KiB

OCR-v2 Active Queue

Status: OCR-v2 is stable; OCR maintenance streaming is implemented, and the broader control-center UX is charted in Wayfinder: Make the PaperForge control center self-explanatory. Last updated: 2026-07-14

Current checkpoint

  • Retrieval recovery is merged to master; the real Literature-hub vault has a healthy 2560-dimensional vec0 index and working M / @ search paths.
  • OCR rebuild: streaming progress + maintenance UI redesign is implemented and reviewed.
  • Multi-key ocr rebuild and full ocr redo emit separate, flushed progress streams and accept a cross-platform cooperative stop request between papers.
  • The maintenance tab now exposes all papers plus the canonical _needs_derived_rebuild() recommendation set, selected batch actions, an above-table progress state, and full refresh on completion.
  • Source Corpus data remains authoritative and was not modified during verification. Only the deployed plugin bundle and disposable maintenance cache were refreshed.

Verification status

  • Focused Python OCR paths: 99 passed, 1 Windows SIGINT test skipped, 1 unrelated empty-result regression deselected.
  • Plugin: 93 passed; TypeScript check and production build passed.
  • Live Obsidian verification: PaperForge 1.5.15 loaded without captured errors; maintenance rendered 734 All rows and 700 Recommended rows from the canonical backend flag.
  • Live progress-state harness showed the floating progress bar, current key, Stop control, and disabled row actions.
  • The repository-wide Python suite remains blocked during collection by the pre-existing test_pr9a_resume_rebuild.py import of removed _assert_collections_healthy.

Frontier

Deferred

  • Vector rebuild UX (PRD Slice 1): deferred.
  • Memory/global maintenance cleanup (PRD Slice 3): deferred.
  • OCR ETA and real-time per-row mutation: out of scope for the completed OCR slice.