The first-pass retrofit re-extracted via extractor.extract_text, which
re-runs Google Vision OCR on scanned pages. OCR is non-deterministic,
so the new text didn't match the chunk content stored in the DB
(produced by the original OCR run) — only ~7% of chunks were located.
New approach (no OCR cost):
1. Use the stored documents.extracted_text from the DB — the exact
text the chunks were produced from, so chunk lookups match.
2. Anchor page boundaries via PyMuPDF direct text reads (free, no
OCR). Pages with usable direct text are anchored by snippet match;
OCR-only pages are linearly interpolated between anchors.
3. Search each chunk in extracted_text using a whitespace-tolerant
helper — needed because the chunker joins paragraphs with single
'\\n' while extracted_text uses '\\n\\n' as page separators.
Verified on 8174-24 (5 docs, 307 chunks) + 8137-24 (9 docs, 512
chunks): 100% chunks tagged, 13s total, $0 cost.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The legacy chunker did not track which PDF page each chunk came from.
Stored chunks had page_number=NULL, which blocked the multimodal
hybrid retriever's text+image boost — it joins (chunk, image) on
(document_id, page_number) and the join could never fire.
This change:
- extractor.extract_text now returns (text, page_count, page_offsets);
page_offsets[i] is the start char offset of page (i+1) in the joined
text. None for non-PDFs.
- chunker.chunk_document accepts an optional page_offsets and tags
each chunk with the page that contains its first character (uses
the existing chunker logic; pages assigned post-hoc by content
search to keep the diff minimal).
- processor.process_document and precedent_library.ingest_precedent
forward page_offsets through the chunker. New uploads now carry
accurate page_number on every chunk.
- Other extract_text callers (tools/documents, tools/workflow,
web/app.py) updated to unpack the third element (ignored).
- scripts/backfill_chunk_pages.py: per-case retrofit. Re-extracts each
PDF (re-OCRs via Google Vision if needed, ~$0.0015/page), computes
page_offsets, and updates page_number on every chunk by content
search. Idempotent; --force re-runs on already-tagged docs.
Forward-only would leave the 419 image embeddings backfilled on
cases 8174-24 + 8137-24 unable to boost their corresponding text
chunks. The retrofit script closes that gap (cost ~$0.60).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Stage C: per-page image embeddings via voyage-multimodal-3 + hybrid
text+image search. Off by default; enable with MULTIMODAL_ENABLED=true.
- Schema V9: document_image_embeddings + precedent_image_embeddings
(vector(1024), page_number, image_thumbnail_path)
- extractor.render_pages_for_multimodal renders PDF pages at
MULTIMODAL_DPI (144) for embedding + JPEG thumbnails at
MULTIMODAL_THUMB_DPI (96) for UI preview, in one pass
- embeddings.embed_images calls voyage-multimodal-3 in 50-page batches
- services/hybrid_search.py orchestrator: rerank applied to text side
first (rerank-2 is text-only); image side cosine; weighted merge
with text_weight 0.65 (env-tunable); image-only pages surface as
match_type='image' so dense scanned content still appears
- processor.process_document and precedent_library.ingest_precedent
gated by flag — non-fatal on multimodal failure
- scripts/multimodal_backfill.py — idempotent per-case CLI to embed
existing documents without re-extracting text
Validated locally on a 5-page response brief: render 0.31s, embed 8.32s,
hybrid merge surfaces image rows correctly. Production rollout starts
with flag=false (no behavior change), then per-case A/B.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A — voyage-3 migration (executed):
- VOYAGE_MODEL=voyage-3 set in Coolify (legal-ai app) and ~/.env
- scripts/reembed_voyage.py: re-embeds document_chunks (6157),
case_law_embeddings (9), precedent_chunks (385), and halachot (400)
using the new model. paragraph_embeddings was empty. 6951 rows
re-embedded in 93s, ~75 rows/sec.
- Same 1024 dim → no schema change needed.
Why voyage-3 over voyage-law-2: benchmark on 3 Hebrew legal queries
with real passages from the corpus gave voyage-3 perfect ordering on
3/3 tests AND the largest separation (+0.483 vs voyage-law-2's
+0.238). voyage-4 family had bigger separation but missed top-1 on
the hardest test.
Phase B (voyage-context-3) and Phase C (voyage-multimodal-3.5 for
scanned + appraiser docs) are designed in docs/voyage-upgrades-plan.md
but deferred — to be picked up in a fresh conversation. The plan
includes:
- Phase B: contextualized embeddings refactor (~49% recall lift on
legal docs per Anthropic's research). Same dim, but ingestion
pipeline must pass full doc context per chunk.
- Phase C: page-level image embeddings via voyage-multimodal-3.5,
stored in a parallel *_image_embeddings table. Hybrid text+image
search. Targets appraiser report tables and scanned PDFs where
current OCR loses layout.
After this commit: MCP server needs a /mcp reconnect to pick up the
new VOYAGE_MODEL env, and the legal-ai container will pick it up on
its next redeploy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The claude_session bridge had two structural defects that made any
non-trivial document extraction unreliable:
1. subprocess.run() blocks the asyncio event loop in the MCP server
for the full duration of every LLM call (60-180s typical).
2. The 120-second timeout was below the cold-cache cost of any
document over ~12K Hebrew characters. Three back-to-back timeouts
on case 8174-24 dropped 43 appellant claims on the floor.
Phase 1 of the remediation plan — keeps claude_session as the engine
(no Anthropic API switch) and restructures around it:
claude_session.py
• query / query_json are now async — asyncio.create_subprocess_exec
instead of subprocess.run, so MCP server can serve other coroutines
while a call is in flight.
• DEFAULT_TIMEOUT 120 → 1800 (30 min). High enough that no realistic
document hits it; bounded so a runaway never zombifies forever.
• LONG_TIMEOUT 300 → 3600 for opus block writing on full case context.
• TimeoutError now actually kills the subprocess (asyncio.wait_for
cancellation alone leaves the child running).
claims_extractor.py
• _split_by_sections: chunks at numbered sections / Hebrew letter
headings / "פרק" markers / markdown ##, falls back to paragraph
breaks, then to hard splits. Targets 12K chars per chunk — small
enough that each chunk reliably finishes inside the timeout.
• _extract_chunk: per-chunk retry (1 attempt by default) with
structured logging on failure. Failed chunks no longer crash the
overall extraction; they're skipped with a partial-result warning.
• extract_claims_with_ai now runs chunks in parallel via
asyncio.gather bounded by a semaphore (CHUNK_CONCURRENCY=3).
For a 25K-char appeal: was sequential 150-300s, now ~70-90s.
Updated all 9 callers (claims, appraiser facts, block writer, qa
validator, brainstorm, learning loop, style analyzer × 3) to await
the now-async API.
The one-shot scripts/extract_claims_8174.py used to recover 43
appellant claims on case 8174-24 has been moved to .archive/ — phase 1
makes it obsolete. SCRIPTS.md updated.
Phase 2 (background-task wrapper around LLM-bound MCP tools, persistent
llm_tasks table, SSE progress) is the structural follow-up — separate PR.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fixes critical bug in 1033-25: user-uploaded עריכה-*.docx files were
orphaned on disk while exports kept rebuilding from stale DB blocks.
New architecture:
- User-uploaded DOCX becomes the source of truth (cases.active_draft_path)
- System edits via XML surgery with real Word <w:ins>/<w:del> revisions
- User can Accept/Reject each change from within Word
Components:
- docx_reviser.py: XML surgery for Track Changes (15 tests)
- docx_retrofit.py: retroactive bookmark injection with Hebrew marker
detection + heading heuristic (9 tests)
- docx_exporter.py: emits bookmarks around each of the 12 blocks
- 3 new MCP tools: apply_user_edit, list_bookmarks, revise_draft
- 4 new/updated endpoints: upload (auto-registers active draft),
/exports/revise, /exports/bookmarks, /exports/{filename}/retrofit,
/active-draft
- DB migration: cases.active_draft_path column
- UI: correct banner using real v-numbers, "מקור האמת" badge,
detailed upload toast with bookmarks_added/missing_blocks
- agents: legal-exporter (3 export modes), legal-ceo (stage G for
revision handling), legal-writer (revision mode)
Multi-tenancy:
- Works for both CMP (1xxx cases) and CMPA (8xxx/9xxx cases)
- New revise-draft skill added to both companies
- deploy-track-changes.sh syncs skills CMP ↔ CMPA
- retrofit_case.py: one-off retrofit of existing files
Tests: 34 passing (15 reviser + 9 retrofit + 4 exporter bookmarks + 6 e2e)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Active scripts (5): auto-sync-cases.sh, backup-db.sh, restore-db.sh,
notify.py, bidi_table.py
Archived (17): one-time migration/seeding scripts whose functionality
is now in MCP server or web API. Moved to scripts/.archive/
Deleted (5): zero-value scripts (duplicates, hardcoded single-case,
debug scripts)
Added scripts/SCRIPTS.md — registry of all scripts with purpose,
status, and what superseded them. CLAUDE.md updated with rule:
any script change requires SCRIPTS.md update.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- CLAUDE.md: clarify vault was deleted, knowledge is in docs/+training/
- Remove import-final-decisions.py (migration completed, all decisions in DB)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Assets live in ezer-mishpati/paperclip-config (cloned at ~/.paperclip).
Deploy via: ~/.paperclip/hebrew/apply-hebrew.sh
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- CEO agent now sends email via notify.py when awaiting human response
- CEO creates child issues (parentId) instead of flat disconnected issues
- Fix notify.py email address to chaim+paperclip@marcus-law.co.il
- Move Paperclip UI assets (RTL CSS + Hebrew JS) into repo under scripts/
- Add deploy.sh script to push assets to live Paperclip instance
- Fix comment box positioning: newest comment on top, input below it
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- auto-sync-cases.sh: fix broken directory scan (was looking for
status subdirs that don't exist), fix env var word-splitting bug,
add safe.directory handling and error logging
- cases.py: auto-create Gitea repo on case_create, fix
documents/original → documents/originals naming mismatch
- app.py: add GET /api/cases/{case_number}/git-status endpoint
- web-ui: add SyncIndicator component in case header showing
sync status (synced/pending/no remote) with last commit time
- pyproject.toml: add httpx dependency
- CLAUDE.md: update Paperclip wakeup API docs
- settings page: switch tag input from Select to free-text with datalist
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Visual dashboard at #/style-report with 4 sections:
- Hero: 24 decisions, char counts, subject donut, timeline
- Anatomy: average section-length breakdown (intro → ruling → conclusion)
- Signature Phrases Wall: pattern cards with real corpus frequencies, filter
chips by type, click → modal with examples
- Contribution: per-decision "new vs confirmed" patterns, growth curve SVG
Backend:
- /api/training/style-report endpoint computes all 4 sections in one call
- Headlines in Hebrew are computed server-side from real data
- Backfill script for style_patterns.frequency using _strip_nikud +
pattern-variant extraction (templates with [placeholders], / alternatives,
ellipsis all handled)
Real findings from the 24-decision corpus:
- דיון משפטי = 49% of avg decision (the focus)
- 23/24 use "לפנינו ערר" opening formula
- 21/24 use "ניתנה פה אחד" closing
- After 7 decisions we already learned 85% of her style patterns
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- New proofreader service strips Nevo editorial additions (front matter,
postamble, page headers, watermarks, inline codes) from DOCX/PDF/MD
- PDF pages use Google Vision OCR for clean Hebrew RTL extraction
- New training page at #/training with drag-and-drop upload, automatic
metadata extraction (decision number, date, categories), reviewable
preview, and style pattern report grouped by type
- API endpoints: /api/training/{analyze,upload,corpus,patterns,
analyze-style,analyze-style/status}
- Fix claude_session.query to pipe prompt via stdin, avoiding ARG_MAX
overflow when analyzing 900K+ char corpus
- CLI scripts for batch proofreading and corpus upload
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add delete_document_chunks for reprocessing, save extracted text to disk
- Expand case directory structure (original/extracted/proofread/backup)
- Update classifier patterns (תגובה, הודעת עמדה)
- Fix proofreader agent paths for new directory layout
- Update HEARTBEAT to notify on every task completion
- Improve bidi_table with LRE/PDF directional embedding
- Add Paperclip project verification and auto-close setup issue
- Add auto-sync-cases.sh for Gitea synchronization
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove cases/new|in-progress|completed subdivision (status managed in DB)
- Rename documents/original → documents/originals (consistent plural)
- Move exports from global data/exports/ into cases/{num}/exports/
- Add documents/research/ for case law and analysis files
- Update all agents, scripts, config, web API endpoints, and DB paths
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New: scripts/notify.py — sends via SMTP (notify@marcus-law.co.il → paperclip+chaim@marcus-law.co.il)
Updated: HEARTBEAT.md — agents must send email when waiting for human decision
Triggers: outcome choice, direction approval, QA failures, review ready.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
