הפרוסה האחרונה של GAP-48 (INV-TOOL1). 18 כלי drafting הומרו ל-{status,data,message}
דרך tools/envelope.py — כולל מסלול הפקת-ההחלטה הקריטי.
עיקרון לכלים עם כשל משמעותי (export_docx/revise_draft/apply_user_edit): err()
ברמת-המעטפת — כך שהסוכן והמשתמש רואים את הכשל; failed_gates רוכב ב-data.
שאר הכלים: ok(data=payload) להצלחה, err להיעדר-תיק/קלט-שגוי/חריגה.
6 צרכני-app.py חוּוטו (get_decision_template, apply_user_edit ×2, revise_draft,
list_bookmarks, export_docx) עם envelope_unwrap + בדיקת status=="error"→4xx,
לשמירת חוזה-ה-API (X6) ללא-שינוי. test_export_qa_gate עודכן לחוזה החדש.
בדיקות: 182/182 עוברים (כולל שערי-QA של הייצוא).
GAP-48 סגור: כל ~12 משפחות-הכלים אחידות. נותר ב-FU-14: GAP-49/50 (שובר), GAP-54.
Invariants: משלים INV-TOOL1 + G2. מתועד ב-X9 (נסגר) + gap-audit פרוסה 7.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
סוגר את לולאת-הלמידה (INV-LRN4): כל החלטה נסגרת מול הסופי, וכל סופי
מנותח מול הטיוטה. מזין את הטבלאות ש-T15 כבר קורא מהן.
T5 — פנקס-התאמה:
- SCHEMA_V26: טבלת draft_final_pairs (snapshot draft + final + diff + analysis + status).
- db: create/update/list_draft_final_pairs.
- mark-final (app.py): תופס snapshot של הטיוטה (decision_blocks) ברגע החתימה,
לפני שאפשר לדרוס אותו, ופותח שורת-פנקס (status=final_received).
T4 — דיסטילציה אוטומטית:
- learning_loop.process_final_version: משתמש ב-snapshot (לא בבלוקים שאולי השתנו),
מסווג style_method↔substance, שומר הצעה ב-pair (status=analyzed).
**הוסר ה-auto-upsert של style_patterns** — ביטל את ה-bug שדרס את שער-היו"ר
וזיהם סגנון במהות (INV-LRN1 + INV-LRN5).
- LESSONS_PROMPT: הפרדת style_method↔substance מפורשת + לקח מופשט בלבד.
- curator wake + hermes-curator.md: מריץ ingest_final_version ראשון; מציע רק
style_method שלא תועד; substance→מסלול precedent.
INV-LRN1 (שער-יו"ר, אין auto-commit) · INV-LRN4 (ניגוד-אמת) · INV-LRN5 (טוהר).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
עונה ל"להתחשב במה שכבר למדנו": הכותב התעלם מעריכות היו"ר ב-/methodology
(נשמרו ב-appeal_type_rules אך block_writer קרא רק קבועי lessons.py) ומ-
decision_lessons של /training. עכשיו הכל מגיע לכתיבה.
- db.get_methodology_overrides(category) — overrides של היו"ר (יחסי-זהב,
כללי-דיון, צ׳קליסטים) מ-appeal_type_rules (כמו merge של ה-API).
- db.get_recent_decision_lessons(limit, practice_area) — לקחי /training.
- _build_style_context(practice_area): מוסיף סעיף "⭐ למידה מצטברת — גובר
על ברירת-מחדל" עם שניהם, אחרי voice-fingerprint (T0). שני ה-callers מעבירים
practice_area. עובד יחד עם הלולאה (T4/T5) שתזין לאותן טבלאות.
תיקון-מספור (חלק מ-T9, דחוף כי T0 הזריק את הטעות): voice-fingerprint §3.1
תוקן — ההחלטה ממוספרת תמיד (מספור-אוטומטי ב-Word); "ללא מספור" היה
ארטיפקט-חילוץ. האנטי-דפוס האמיתי: רשימת-מיני בתוך פסקה + מספרים ידניים.
INV-LRN4 (הזרמת למידה) · INV-LRN5 (טוהר). G11.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
שתי בעיות UX בדף /precedents:
1. חילוץ מטא-דאטה לא נתן שום אינדיקציה שהוא רץ. בניגוד לחילוץ טקסט/הלכות
(extraction_status / halacha_extraction_status) למטא-דאטה היתה רק חותמת-זמן
metadata_extraction_requested_at — אין מצב "processing", לכן StatusPill לא
הציג כלום. נוספה עמודת metadata_extraction_status ('pending'|'processing'|
'completed'|'failed') במתכונת העמודות הקיימות, וה-worker
(process_pending_extractions + reextract_metadata) מעדכן אותה: processing
בתחילת פריט, completed בסיום (מנקה גם את החותמת), pending בכשל (לריטריי).
ה-UI מציג תג "מחלץ מטא-דאטה" + באנר מונה-אצווה עם אחוז התקדמות (high-water-mark
של עומק-התור) שמתעדכן אוטומטית דרך ה-polling הקיים (5ש').
2. שתי טבלאות מוערמות (בתי משפט / ועדות ערר) חייבו גלילה ארוכה. הוחלפו במתג-
מקטעים — טבלה אחת בכל פעם, עם שמירה על העמודות הייעודיות לכל סוג.
Invariants: G2 (מרחיב מנגנון-סטטוס קיים, לא מסלול מקביל), INV-TOOL4/GAP-45
(המשך חשיפת תור-החילוץ הסמוי). אין נגיעה בתוכן משפטי (G11).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ה-provenance (document_id, page_number, score) כבר נשלף ב-search_similar אך
נזרק בבניית פלט draft_section. כעת מוחזר לכל קטע ב-case_documents/precedents,
כך שהכותב יכול לעקוב אחורה אל מסמך-המקור והעמוד ולצטטם, ולא לסמוך על תוכן
חסר-מקור. תוספתי בלבד — אין צרכן שמפרסר את מפתחות-הפלט, תואם-לאחור.
נותר ב-GAP-47: העברת הנחיות-יו"ר מ-analysis-and-research.md ל-DB
(get_chair_directions) — שינוי-מסלול גדול יותר, לפרוסה נפרדת.
Invariants: מקיים INV-TOOL4 (מקור-אמת נגיש) + G9 (provenance). לא נוגע ב-G2/G1.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
INV-TOOL3 (idempotency על מפתח דטרמיניסטי). כל שלושת הכלים מחזירים את הרשומה
הקיימת במקום ליצור כפילות:
- case_create — מפתח case_number (כבר UNIQUE ב-schema): מחזיר את התיק הקיים
במקום unique-violation.
- precedent_attach — מפתח (case_id, section_id, citation, quote): צירוף חוזר
של אותו ציטוט לאותו סעיף מחזיר את הקיים.
- document_upload — מפתח (case_id, SHA-256 של בייטי הקובץ): העלאה חוזרת של אותו
קובץ מחזירה את המסמך הקיים ו**מדלגת על copy+OCR+embed** (החלק היקר). נוספה
עמודת documents.content_hash (תוספתי, DEFAULT '') + get_document_by_hash.
נבחרה בדיקת-מפתח ברמת-אפליקציה (SELECT-לפני-INSERT) ולא UNIQUE-constraint —
כדי לא לשבור startup אם קיימים נתונים-כפולים legacy. אין מיגרציה הרסנית.
עודכנו docs/spec/X9 (INV-TOOL3 ✅) ו-gap-audit (GAP-52 ✅, פרוסה 2).
py_compile עבר על 4 קבצי הקוד. אימות runtime (restart MCP server) נדחה עד
שהחילוץ הפעיל יסתיים.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
תוספתי בלבד, אפס שבירת-תאימות. שני invariants מחוזה-כלי-ה-MCP (X9):
GAP-44 (INV-TOOL4, סימטריית extract/get): נוסף get_appraiser_facts — ה-get
המקביל ל-extract_appraiser_facts. קורא list_appraiser_facts + detect_appraiser_conflicts
מה-DB ללא חילוץ-LLM יקר ולא-דטרמיניסטי. מחזיר count=0 (לא שגיאה) אם טרם חולץ.
GAP-53 (INV-TOOL5, limit-caps / OWASP API4:2023): נוסף _clamp_limit (תקרה 200,
non-positive→max) על ~13 כלי list/search ב-server.py (case_list, search_*,
precedent_library_list, halachot_pending, missing_precedent_list, list_*_citations…).
list_chair_feedback קיבל param limit חדש (server→workflow→db עם LIMIT) — היה ללא תקרה כלל.
לא הוסף get_appraiser_facts ל-frontmatter של סוכנים (INV-AG3 "לא עודף" — ההוראות
עוד לא מפנות אליו; חיווט = follow-up). נותר ב-FU-14: GAP-45/48/49/50/51/52.
עודכנו docs/spec/X9 (INV-TOOL4/5) ו-gap-audit (סטטוס פרוסה 1).
אומת: py_compile על 4 קבצי הקוד. אימות runtime (restart MCP server) נדחה עד
שהחילוץ הפעיל של היו"ר יסתיים.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
סוגר את לולאת פידבק-יו"ר→ידע-סוכנים. עד כה resolve רק עדכן את ה-DB; עכשיו
לחיצה ב-/feedback מעירה את ה-CEO שמקפל את הלקח לקובץ לפי הקטגוריה.
- paperclip_client.py: wake_ceo_for_feedback_fold() — יוצר issue ב-Paperclip
עם הלקח + rubric ניתוב (style→SKILL.md, wrong_structure→block-schema,
אחר→lessons.md), מעיר CEO. משכפל את דפוס wake_for_precedent_extraction
- db.py: get_chair_feedback(id) — שליפת הערה בודדת עם case_number/appeal_type
- app.py: resolve endpoint מקבל fold (ברירת מחדל true); BackgroundTask
fire-and-forget; guard — רק עם lesson_extracted. מחזיר fold_queued
- legal-ceo.md: dispatch ל-feedback_fold_ + סעיף "קיפול הערת יו"ר" עם rubric
- frontend: useResolveFeedback מקבל fold; /feedback שולח fold=true עם toast;
drafts-panel שולח fold=false (bookkeeping per-case, בלי קיפול כפול)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- בקאנד: GET לפני ה-async task — אם citation כבר קיים כ-external_upload מחזיר 409
- DB: get_external_case_law_by_citation — lookup לפי citation + source_kind
- פרונט: banner אדום עם פרטי הרשומה הקיימת ושני כפתורות:
• "הפעל חילוץ מחדש" — request-halachot ל-ID הקיים וסגירת הטופס
• "מחק את הרשומה" — DELETE עם confirm, ניקוי conflict לאחר מכן
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
strip_nevo_preamble's _DECISION_START only matched ועדת-ערר openings (בפנינו /
הערר שבנדון / ...), so Nevo COURT judgments — exactly the ones carrying a
מיני-רציו — slipped through unstripped. The editorial mini-ratio then leaked into
the chunked body, risking that the halacha extractor reads Nevo's answer key
(contamination) and polluting the corpus. Proven on בג"ץ 1764/05: its full_text
still contained the מיני-רציו (unstripped).
Fix:
- Extend _DECISION_START with court-ruling openings: פסק-דין/פסק דין header and
the authoring-judge line (השופט/ת, כב' השופט, הנשיא, המשנה לנשיא). re.search
picks the earliest line-start match → the real opinion start, not the prose
ratio above it.
- Widen the Nevo-marker detection window 400→1500 chars so a long court/parties
header doesn't push חקיקה שאוזכרה:/מיני-רציו: out of range.
Verified on the real 1764/05 full_text: strips 2702 chars, body now starts at
'השופט ס' ג'ובראן:', מיני-רציו gone. Regression: ועדת-ערר openings still strip;
non-Nevo text untouched; markers-past-400 now detected. Suite 182 passed (6 new).
This is the anti-contamination prerequisite for the Nevo-ratio gold-set (#86.3/#81.7).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After a precedent finishes extracting, a claude_session pass folds facets of the
SAME legal question (below #82's dedup cosine — the שפר 14-vs-4 / 403-17→89
granularity gap) into one canonical; the rest are marked 'rejected' (reversible:
out of the active corpus AND the review queue, but recoverable). FOLD-ONLY —
never merges distinct legal questions, never invents.
- Engine: claude_session-as-judge (local CLI, zero cost), 'high' effort — folding
needs careful judgment. One pass per precedent, runs in _extract_impl once all
chunks are done (the prompt dedups within a chunk; this catches across chunks).
- Pure, unit-tested helpers in halacha_quality: CONSOLIDATE_SYSTEM,
build_consolidation_prompt, parse_fold_groups (fails SAFE → [] on any malformed
shape; drops <2-member groups; coerces/dedups indices).
- halacha_extractor._consolidate_precedent picks the canonical per group
(approved>pending, higher confidence, quote_verified, longer) and rejects the
rest via the existing update_halachot_batch (#84). Never rejects a canonical.
Fails OPEN on any error (no CLI / parse fail → 0 folds, data untouched).
- config: HALACHA_CONSOLIDATE_ENABLED/MODEL/EFFORT.
Verified: suite 176 passed (10 new); integration vs dev DB — a 2-facet group
folds to 1 canonical + 1 rejected (tagged), distinct rules untouched, claude
error → 0 folds (fail-open).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
#81.3 — a post-extraction validator that flags halachot whose rule_statement is
NOT entailed by its supporting_quote (the model over-reaching beyond its source).
- Engine: claude_session-as-judge (local CLI, zero API cost) per chaim's standing
preference — one batched judge call per chunk, NOT a hosted NLI model.
- Pure, unit-tested helpers in halacha_quality: NLI_SYSTEM, build_nli_prompt,
parse_nli_verdicts (fails OPEN — any shape/label ambiguity → 'entailed').
- halacha_extractor._nli_check wraps the call; fails OPEN on any error (e.g. no
CLI in the container) so a flaky judge never blocks a genuine halacha.
- Non-entailed (neutral/contradiction) → quality_flag 'nli_unsupported' which
blocks auto-approve (routes to pending_review) via the existing store gate.
- config: HALACHA_NLI_ENABLED/MODEL/EFFORT (effort 'low' — entailment is simple).
Verified: suite 166 passed (10 new); LIVE smoke test against the real claude CLI
returned ['entailed','neutral'] for a supported vs unsupported rule.
Also commits TaskMaster #86 (Nevo preamble/ratio: anti-contamination strip fix +
gold-set benchmark) capturing today's strip_nevo_preamble findings.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Legacy Hebrew .doc precedents (e.g. nevo.co.il CP1255 OLE2) can now be
uploaded directly through the precedent-library, missing-precedent, and
training upload paths — the frontend already advertised .doc but the
backend gate rejected it before reaching the extractor.
- web/app.py: add .doc to ALLOWED_EXTENSIONS (covers all paths that share
the set: precedent library, missing-precedent, training).
- Dockerfile: install libreoffice-writer-nogui (no X11/Java) so the
extractor's existing _extract_doc LibreOffice conversion works in the
Coolify container (was missing → would fail at runtime).
- extractor.py: isolate the LibreOffice user profile per call to avoid a
profile-lock failure on concurrent .doc conversions.
Verified in python:3.12-slim (prod base): .doc→.docx→text yields text
byte-identical to a native Word .docx save (103 paragraphs, 24,341 chars).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Make the chair's pending-halacha review faster and less exhausting.
Backend:
- New 'deferred' review_status (snooze): stays out of the active library AND
out of the default pending queue, without the finality of 'rejected'.
update_halacha stamps reviewer+reviewed_at on defer; HALACHA_REVIEW_STATUSES
is the single source of valid statuses (PATCH validation now uses it).
- db.update_halachot_batch(ids, status, reviewer) — one atomic UPDATE for a
whole group; invalid status / empty ids are a no-op.
- POST /api/halachot/batch (HalachaBatchReviewRequest) wraps it.
- update_halacha now RETURNs quality_flags too (parity with list_halachot).
Frontend (halacha-review-panel):
- Quality-flag badges (#81: non_decision / truncated_quote / thin_restatement /
quote_unverified) so the chair sees WHY an item was held back.
- Defer action — button + keyboard 'D' — to snooze without rejecting (fixes the
'leave in pending forever' anti-pattern; reject stays the junk verb).
- Per-precedent batch bar: 'אשר הכל' / 'דחה הכל' via useBatchReviewHalachot
(one request, one refetch) with confirm guards.
- Halacha/HalachaPatch types gain quality_flags + 'deferred'.
Verified: mcp-server suite 156 passed; web build green; end-to-end integration
against dev DB (batch approve/reject, defer sets status+timestamp, pending
excludes approved+deferred, deferred queryable, invalid status no-op).
Note: api:types regen deferred until deploy (the batch hook is hand-typed, not
dependent on generated types).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
#83 pipeline robustness — the index-numbering correctness guarantee:
- Add CREATE UNIQUE INDEX idx_halachot_unique_index ON halachot(case_law_id,
halacha_index). The extractor assigns the index as MAX+1 under an in-process
store-lock + a cross-process pg advisory lock, so collisions shouldn't occur
in normal operation — but per the research (FireHydrant/OneUptime) the
constraint is the actual correctness guarantee while the lock is the
optimization. A racing/double run now fails LOUDLY (UniqueViolation, chunk
left un-checkpointed → clean resume) instead of silently appending the
duplicates that were the 2026-05/06 over-extraction root cause.
Data prep (run against the live DB before the constraint, backed up to
data/audit/halacha-reindex-backup-*.sql): the 6 precedents that still carried
colliding halacha_index values (9 groups, distinct principles that shared a
number — NOT content dups) were renumbered to unique sequential indices.
Verified: advisory lock holds cross-process and the DB path is direct asyncpg
(no transaction-pooler), so the session lock is safe (83.1); force=True does
delete+checkpoint-clear in one transaction (83.5); constraint rejects a
duplicate-index insert (integration-checked). Full suite 156 passed.
Also commits the TaskMaster tracking for the whole halacha-quality initiative
(#81-#84 + research-backed subtasks, statuses).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bake the 2026-06-03 strict-cleanup rubric into the extraction pipeline so the
corpus stays clean at the source instead of accumulating duplicates, obiter
dicta, truncated quotes and thin restatements that clog the review queue.
#81 — quality gate:
- New pure module halacha_quality.py with unit-tested validators:
non-decision/obiter (Wambaugh markers), truncated-quote (mid-word cut),
thin-restatement (rule≈quote), quote-unverified.
- Validators run in halacha_extractor._process; a non-decision is re-typed
obiter; flags persist in new halachot.quality_flags column.
- Auto-approve now requires confidence>=threshold AND no quality flags;
flagged items route to pending_review regardless of confidence.
- Both extraction prompts hardened: reject undecided dicta, exclude
case-specific applications, require abstraction, forbid over-splitting.
#82 — dedup-on-insert (store_halachot_for_chunk):
- Within the same precedent, skip a halacha whose normalized supporting_quote
already exists, or whose rule-embedding has cosine>=HALACHA_DEDUP_COSINE
(0.93) against an already-stored one. Makes re-runs idempotent.
Migration: halachot.quality_flags TEXT[] (additive, idempotent ALTER).
Tests: 19 new unit tests; full suite 156 passed. Validated end-to-end against
dev DB (dedup skips dups, flag blocks auto-approve, re-run inserts 0).
Calibration: flags fire on only ~10% of current survivors (low false-positive).
Spec: docs/halacha-strict-rubric.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A section that opens with a short header line ('דיון', 'טענות המשיבים')
followed by a paragraph larger than chunk_size flushed the header alone as a
tiny chunk. #55 added a query-time >=50 filter to hide these; this removes
them at the source.
_split_section: (1) don't flush a buffer still below MIN_CHUNK_CHARS — let it
absorb the next paragraph even if that overflows chunk_size, so a short header
rides with its following content; (2) fold a trailing tiny chunk back into its
predecessor.
Verified: re-chunked the 4 corpus docs that still had a tiny chunk
(ע"א 5138/04, בר"מ 2340/02, בג"ץ 6525/15, 403-17) — corpus-wide chunks<50
went 4 -> 0; all 4 stay embedded/searchable and rank top in a relevant search
(נווה שלום #1 for the s.19(ג)(1) exemption query). No regression.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
xhigh is the quality sweet-spot for a single precedent but very slow at scale
(64-chunk case ≈ 20 min). Bulk queue-drains (process_pending over many
precedents) now use a lighter effort to cut wall-clock; interactive single
re-extraction keeps xhigh quality.
- config.HALACHA_BULK_EXTRACT_EFFORT (env, default 'high'; set 'medium' for max
speed, 'xhigh' to match single).
- extract()/_extract_impl()/_extract_chunk() take an `effort` override threaded
to claude_session.query_json; None falls back to HALACHA_EXTRACT_EFFORT (xhigh).
- process_pending_extractions(kind='halacha') passes the bulk effort; single
reextract_halachot keeps xhigh.
Verified end-to-end (mocked LLM): _extract_chunk(effort='medium') → query_json
effort='medium'; effort=None → 'xhigh' fallback. Closes the open item in #72.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Halacha extraction held ALL chunk results in memory and stored once at the very
end — a crash/interrupt mid-run (e.g. the 2026-05-31 freeze) lost everything and
re-paid the full LLM cost on retry.
Now each chunk's halachot are stored AND the chunk is checkpointed
(precedent_chunks.halacha_extracted_at) the moment it finishes:
- V25 schema: precedent_chunks.halacha_extracted_at (per-chunk checkpoint).
- db.store_halachot_for_chunk: atomic per-chunk insert (halacha_index continues
from MAX, caller serializes via an in-process store-lock) + checkpoint mark.
- db.reset_halacha_extraction (force) / mark_all_chunks_extracted (legacy backfill).
- _extract_impl rewritten: resume by default (skip checkpointed chunks; failed
chunks stay pending and are retried; status stays 'processing' until all done);
force=True wipes + redoes all. reextract_halachot passes force=True; the queue
drain (process_pending) resumes by default.
- Legacy guard: a pre-V25 precedent (halachot exist, no checkpoints) is
backfilled and treated as complete — never re-extracted (would duplicate).
Verified on 9002-24 (55 halachot, legacy): resume → legacy-backfill, NO
duplication (stays 55), all chunks checkpointed. Index continuation: store at
55,56 after max 54, no collision. Tracks #72.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-05-31: opus-4-8 @ xhigh extraction + overlapping driver processes (agent
fallback retries each spawn an independent `python -c` driver; process_pending is
serial WITHIN a process but the box ran 4-5 drivers in parallel) → 12-16 concurrent
xhigh `claude -p` procs → load 69 → hard reboot.
Fix: halacha_extractor.extract() now takes a Postgres advisory lock
(pg_try_advisory_lock, key 'HALA') before any work. If another extraction (any
process/agent/driver — all share the legal-ai DB) holds it, the call returns
status='busy' and the precedent stays pending for the next drain. Guarantees ONE
extraction at a time ACROSS PROCESSES — an in-process Semaphore cannot (drivers
are separate OS processes). Core logic moved to _extract_impl (unchanged) under
the lock. CHUNK_CONCURRENCY now env-tunable (HALACHA_CHUNK_CONCURRENCY, default 3).
Verified: while a lock is held, extract() returns 'busy' with no LLM call; lock
releases cleanly and the next extraction proceeds. Tracks #72.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Covers GAP-11 (INV-RET4/G8) and GAP-14 (INV-QA1/G10). Retrieval quality was
never measured (only telemetry observation) and the halacha review backlog was
invisible (the 10/19 gap was found by accident).
Unit B — backlog visibility (pure code, container):
- metrics.halacha_backlog(conn) → {pending_review, approved, rejected, published,
total, oldest_pending_at}; surfaced in metrics.get_dashboard() (get_metrics MCP
tool) and /api/system/diagnostics. Live count revealed 178 pending / 1552 total,
oldest from 2026-05-03 — previously invisible.
Unit A — retrieval eval harness (host-side scripts):
- scripts/eval_gold_bootstrap.py — seeds data/eval/gold-set.jsonl. Two sources:
citations (cited==relevant via search_relevance_feedback — empty until decisions
cite precedents) and known_item (query=case_name → relevant=self; a real
citation-free signal, the methodology #52 checked by hand). Idempotent; preserves
source='chair' rows.
- scripts/eval_retrieval.py — runs the production retrieval path (search_library /
search_internal) over the gold-set; computes precision@k, recall@k, MRR, nDCG@k
(k=5,10); aggregates overall + per-corpus + per-practice_area; writes a report and
a delta vs committed baseline.json (which records the retrieval_config it reflects).
--self-test unit-checks the metric math offline.
Gold-set strategy = hybrid (chair decision): bootstrap + chair review. The citation
source is empty today (0 cited precedents in decisions), so the seed is known-item
(77 queries: 54 internal_decisions + 23 precedent_library). The gold-set is
PROVISIONAL until Dafna reviews it (the domain chair-gate).
Baseline (production config: multimodal+rerank on): R@10=0.987, MRR=0.837,
nDCG@10=0.872. Finding: MULTIMODAL_ENABLED=true slightly lowers known-item recall
(image-page results displace exact name matches) — relevant to #15. precedent_library
weaker than internal (R@10 0.957 vs 1.0) — one external precedent unfindable by name.
"CI gate" realized as discipline (re-runnable harness + committed baseline + run
before/after any retrieval-layer change) — retrieval needs prod DB + Voyage, no CI
runner has that access.
Spec: docs/superpowers/specs/2026-05-31-fu5-eval-harness-design.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wakeup-INSERT rule is universal (never allowlisted — hard invariant). Raw-HTTP
rule exempts the sanctioned helpers + standalone operator/admin scripts (a
distinct category per fitness-function scope differentiation + DRY: tooling
needn't reuse the FastAPI wrapper). Repo scanned clean under these rules.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dry-run surfaced 2 rows with בל"מ prefix but proceeding_type=ערר. Since the
migration strips the prefix, a wrong proceeding_type would silently lose the
בל"מ signal — must be chair-adjudicated, not auto-applied. Chair table now
flags 4 rows: 2 DUP_CHECK (8047-23) + 2 PROC_MISMATCH.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
