- License breakdown bar: segmented colour bar + legend with counts and %
- Key metric blocks: entries, sources, writers, date range, transcript count
- Warn block shown if any entries have unclear rights
- compute_corpus_stats() helper in app.py; license short-names + colour map

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

5 entries stored only a PDF file, which browsers can't display inline
as an <img>. Used pdftoppm at 200 DPI to produce a JPEG thumbnail for
each, added as role=thumbnail prepended in the files list so the review
app picks it up immediately. Original PDF kept as role=original.

Affected entries:
- commons__auerbach_letter_shtenzel_1961__p0001
- commons__bendin_semichah_shtenzel_1933__p0001
- commons__weidenfeld_eruv_letter_1947__p0001
- commons__wosner_halachic_ruling_1981__p0001
- commons__wosner_support_letter_1990__p0001

Validation: 111 sources, 209 entries, 242 files verified.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Actions toggle (✎ Actions button in nav header):
- Hidden by default; one click reveals flag button + comment textarea on
  every entry card across all views (home, group pages, audit)
- State persists in localStorage; CSS-driven via body.show-actions class
- Audit submit button also gated behind the toggle

All Entries view (third tab on home page):
- Flat scrollable grid of all 209 entries, same card style as group pages
- Includes rights/transcript badges, lightbox zoom, action strips
- Browse save bar (sticky bottom) appears when Actions are on; saves to
  the same /api/audit/decide endpoint and merges with existing decisions

Group pages (writer/source):
- Flag/comment action strips added to each entry card (hidden by default)
- Floating browse save bar; loads + merges with existing audit decisions

New API endpoint GET /api/audit/decisions for client-side merge before save

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Removed entries (all flagged via audit UI 2026-05-25):
- commons__bodleian_geniza_ms_heb_d_41_4b__p0001  (Geniza fragment)
- commons__bodleian_geniza_ms_heb_e_39_78b__p0001  (Geniza fragment)
- commons__chief_rabbinate_letter_1921__p0001
- commons__chushiel_letter_geniza__p0001           (Geniza)
- commons__damascus_pentateuch_ms_heb_8_7088__p0001
- commons__geniza_education_ts_k5_13__p0001        (Geniza)
- commons__grodzinski_letter_about_kook__p0001
- commons__halper462_exilarch_genealogy__p0001
- loc__2024422570__p0003, p0004, p0005

9 now-orphaned source records marked rejected.
Corpus: 198 entries across 111 sources (228 files).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- scripts/ingest_loc.py: paginate LoC Hebraic Manuscripts JSON API,
  filter pre-1700/printed items, download up to 5 pages per item,
  write data/review/loc_pending.jsonl
- scripts/ingest_zucker.py: parse OPenn TEI manifests for the Zucker
  Ketubah Collection, write data/review/zucker_pending.jsonl
- scripts/merge_review.py: promote approved review decisions into
  entries.jsonl + sources.jsonl; auto-creates per-item source records
- scripts/review_app/requirements.txt: Flask dependency for review app
- scripts/review_app/templates/batch.html: batch review UI (invert
  accept pattern — all dim by default, click to accept)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- data/review/loc_pending.jsonl: 722 entries staged from LoC Hebraic
  Manuscripts collection; 166 items, up to 5 pages each
- data/review/loc_decisions.json: 722 decisions (27 approved, 695 rejected)
- data/review/zucker_pending.jsonl: 288 entries staged from OPenn
  Zucker Ketubah Collection
- data/review/zucker_decisions.json: 288 decisions (1 approved, 287 rejected)

These files serve as the audit trail for the two review sessions.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Adds only the scan directories for sources whose entries are in the
corpus index (entries.jsonl). Rejected/unreviewed scan directories
from the same download sessions remain untracked.

Sources included:
- 16 LoC Hebraic Manuscripts items (loc__2018757642 … loc__2023530858)
  accepted from the 166-item LoC review session (27 entries total)
- openn__zucker__ket_z_238 — single accepted Zucker ketubah
  (Hebrew-text panel, CC-BY-SA 4.0)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…eview

- app.py full rewrite: extract _enrich_entries() helper (eliminates 4
  copy-pasted enrichment blocks), add mtime-keyed module-level file
  cache, add load_audit_decisions(live_ids) to filter stale decisions
  at load time, fix path traversal in serve_scan() with resolve()+
  relative_to() check, fix save_decisions() to merge-not-clobber via
  existing.update(incoming), fix source_detail() 404 axis (check
  source_id not in sources, not len(entries)==0), fix review_batch()
  hardcoded source_id with primary_sid=max(set(...),key=count), fix
  walrus-operator double-call in group thumb helpers, remove dead
  imports (re, sys, datetime, timezone)

- templates: slim full-entry JSON blobs to ID-only arrays
  (ENTRIES→ENTRY_IDS, ALL_ENTRIES→ALL_ENTRY_IDS) — eliminates ~588 KB
  of tojson payload per page load; update save loops to iterate IDs

- group.html: remove dead fetch('/api/audit/status') try block in
  saveDecisions() that preceded the real merge fetch

- data/review/audit_decisions.json: clear 175 stale decisions
  (all referenced entries removed from corpus in audit passes)

- merge_review.py: prune stale IDs from audit_decisions.json after
  each batch merge so the file stays in sync with the live index

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>