Follow-up to the status SSoT (#382): analyst_verified + research_complete are
now first-class canonical statuses, so refresh the docs that still framed them
as legacy/decorative.
- HEARTBEAT.md §7: rewritten to point at the single source
(case_status_model.py / GET /api/status-model), list the 12 canonical statuses
by phase, and state the two intermediates are canonical (not legacy); genuinely
old markers (proofread/analysis_enriched/…) are display-only fallbacks.
- legal-qa.md / legal-ceo.md status maps: dropped the "legacy/optional" framing
on research_complete; point to the canonical model.
- scripts/backfill_case_status_trim.py: removed analyst_verified +
research_complete from the downgrade map (they are valid statuses again — must
NOT be downgraded) + updated the docstring.
Docs/prompts only — no behaviour change. py_compile clean.
NOTE: the host ~/legal-ai needs `git pull` for the Paperclip agents to pick up
the refreshed prompts (they run on the host tree, not the container).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Make the effort-calibration recommendation style-clean (style acquisition is
the goal — write like chair Dafna Tamir):
- recommend_effort ranks by anti_pattern_total (PRIMARY, the clean style-rule
violation count) → golden_ratio_deviation_pp (SECONDARY, structural-style,
None=worst) → distance (faint final tiebreak). change_percent is EXCLUDED from
the ranking (it mixes style with content-completeness per 07-learning §0.7 —
noise for STYLE) and kept in the report as context only.
- Add a confidence flag ("clear"/"weak"): a pick whose anti-pattern lead over the
runner-up is within noise (margin < epsilon = max(0.5, 0.20·spread)) is flagged
⚠️ weak, so an over-claimed pick (the old confident block-zayin "low") is never
shown as confident. Rendered as a confidence column in the report.
- Add --rerank <report.json>: re-apply the new key to a saved calibration JSON
offline (NO LLM, NO DB), print + write *-reranked.md. Re-ranks today's 8-final
run without regenerating.
- recommend_effort stays a single shared pure function reused by
scripts/calibrate_block_yod.py (G2); both --self-test suites pass.
- Spec note added (07-learning §0.7) + SCRIPTS.md updated.
Invariants: G8 (style-clean empirical ranking), G2 (single shared recommend_effort,
reused by the block-yod harness — no parallel metric path), INV-LRN5 (style-only
signal; change% reported-not-ranked).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Backend follow-up to INV-UI9 (X6 §"אודיט-נלווה"): make server-side
datetimes explicit so flipping the container to TZ=Asia/Jerusalem cannot
silently skew stored or displayed timestamps. Storage stays UTC; human
display is Israel time.
- audit.py: naive datetime.utcnow() → datetime.now(timezone.utc) for the
TIMESTAMPTZ audit_log.created_at. Under TZ=Asia/Jerusalem the naive UTC
wall-clock would be read as Israel-local and stored ~3h off. (§6: no
silent skew.)
- calibrate_effort.py: _ts() report stamp now Israel-time via
ZoneInfo("Asia/Jerusalem") with a filename-safe "-IL" suffix (script runs
on the UTC host); added a labeled "שעון ישראל" line to the report header
so the chair-facing timestamp is unambiguous. --self-test still ALL PASS.
- research_md.py: display fields made deterministic Israel-tz —
update_chair_position timestamp (datetime.now(IL_TZ)) and parse() mtime
(datetime.fromtimestamp(..., tz=IL_TZ)) — correct regardless of process TZ.
Left intentionally Israel-local-day date.today()/year-cutoff call sites
untouched (drafting/party_claims_summary/block_writer/court_fetch_supreme)
— a Jerusalem committee wants the Israel calendar day; they resolve
correctly once TZ flips.
Invariants: INV-UI9 (store-UTC / display-Israel, server side); §6
(timezone-aware datetimes, no silent ~3h skew).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A live calibration run crashed with no error handling and lost ~2h of work:
the per-cell loop in _run() reached block-yod (discussion), block_writer.write_block
raised ValueError("לא ניתן לכתוב בלוק דיון ללא כיוון מאושר…") because the discussion
block needs an approved direction calibration cases don't have, and since the report
was only written at the very end of main(), every completed generation was lost.
Three robustness fixes (no metric/scoring change — G2):
1. Drop block-yod from default CALIBRATABLE_BLOCKS — out of WS5 interim-draft scope
AND structurally non-calibratable standalone (needs brainstorm→approve_direction).
Kept in new VALID_BLOCKS so a user can still force it via --blocks block-yod.
2. Per-cell try/except in _run(): any single (case, block, effort) cell failure is
logged as a warning (case/block/effort/error) and skipped, never fatal —
defense-in-depth for a forced block-yod and for transient rate-limits mid-run.
3. Incremental persistence: _write_report flushes after each completed block (ts
threaded into _run), so a crash later in the grid never loses finished blocks.
JSON/MD shape unchanged — partial results omit not-yet-done blocks from by_block.
--self-test: ALL PASS (14/14). block-yod gone from --dry-run defaults.
Invariants — נוגע / מקיים:
- G8 (eval-harness robustness) — partial results persist; a single-cell failure is
non-fatal; the harness survives block-yod's missing-direction raise and rate-limits.
- G2 (מקור-מדידה יחיד) — no metric-path change; _score_cell / block_distance_to_final
/ recommend_effort / aggregate_cell untouched.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The voice-fingerprint (docs/daphna-voice-fingerprint.md) is the PRIMARY abstract
style channel the writer consumes (07-learning §0.2 channel A), but it was
hand-authored and never regenerated from the growing corpus — the "voice" the
writer adapts stayed frozen while finals accumulated. PR #345 unfroze the
structural half (corpus-measured ratios → writer); this is the prose half.
scripts/regenerate_voice_fingerprint.py reads the corpus (style_corpus stats +
a bounded block-level style_exemplars sample + measured section ratios) and asks
Opus (local) to PROPOSE refinements/additions to the fingerprint — STYLE / METHOD
/ VOICE / LEXICON only, grounded in corpus evidence, never case substance
(INV-LRN5), preserving curated insights rather than replacing them.
CHAIR-GATED, HARD (INV-LRN1): the prose profile is voice-knowledge, not a
low-risk style rule, so it does NOT auto-flow. The script writes a PROPOSAL to
data/curator-proposals/ (the existing curator-proposal surface) and NEVER
overwrites the live fingerprint; the chair reviews and hand-commits accepted parts
— the same manual gate the fingerprint already uses. Runs on the host (claude CLI).
Reuses claude_session + measure_corpus_ratios + the curator-proposals dir (G2 —
no parallel synthesis or gate). SCRIPTS.md + spec §0.2 updated.
Invariants: INV-LRN1 (hard chair gate for voice-knowledge prose), INV-LRN5
(style-only synthesis), G2 (reuses existing synthesis inputs + proposal surface).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The writer, learning panels, and legal_mcp tools run from the HOST tree
(Paperclip agents claude -p with cwd=~/legal-ai; pm2 services import from
mcp-server/src) — not the container. A merge + Coolify deploy only refreshes the
FastAPI container; host-side code stayed on the last manual `git pull`. On
2026-06-28 the host was 16 commits behind, so a day of learning-loop fixes sat
dormant until pulled by hand.
host_sync.sh (run every minute via cron, like the existing data/training pull)
fast-forwards the host tree to origin/main when it falls behind. Fast-forward
ONLY — never forces, so the perpetually-dirty .taskmaster/tasks/tasks.json is
never clobbered (clean ff leaves it; an upstream conflict logs + bails for manual
handling). Agent runs spawn fresh and pick up new code with no restart; only the
long-running pm2 services that import app code (chat_service, court_fetch_service,
reaper) are restarted, and only when their own files changed (precise, minimal
churn). Logs to data/logs/host_sync.log.
Cron entry (added on the host, flock-guarded):
* * * * * flock -n /tmp/host_sync.lock bash ~/legal-ai/scripts/host_sync.sh
SCRIPTS.md updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The style-exemplar corpus (channel B — the writer's block-level retrieval of
Dafna's real prose) was FROZEN at the one-time seed backfill: new finals were
enrolled into style_corpus but never broken into exemplars, so the richest style
channel never grew (8137/8126/8174 had 0 exemplars). The writer kept retrieving
only March–April seed paragraphs no matter how many finals were signed.
Extract the per-decision exemplar logic (section→paragraph→Voyage-embed→replace)
into a shared service `legal_mcp.services.style_exemplars.extract_and_store` —
the SINGLE implementation now used by BOTH the one-time backfill and the live
enrollment path (G2; no parallel extractor). `_enroll_final_in_library` calls it
on every final upload (source='internal_committee', the same source the writer's
search_style_exemplars reads). Voyage embeds over REST → container-safe;
best-effort, surfaced in the upload response, never fails the upload.
Effect: every signed final now grows the exemplar corpus, so the writer's
block-level style retrieval improves with each decision — the core "learn from
every decision" fix for channel B. Path A (style_distance_history) will track
whether the larger exemplar pool reduces style-distance over time.
Invariants: G2 (one extraction path shared by backfill + enroll), INV-LRN5
(style/structure prose only — substance routes elsewhere), INV-LRN4 (the
draft↔final loop now feeds the exemplar channel, not just the lesson channel).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The voice-learning panel produced 81 vetted style lessons — all stuck at
review_status='proposed' with 0 ever approved — so the writer (which reads only
'approved') received NONE of them. The system captured learning but never let it
flow. Chair decision (2026-06-28): a GRADUATED gate by content risk.
- STYLE lessons (categories style/structure/lexicon/tabular) the 2/2 panel kept →
created as review_status='approved' → flow to the writer immediately, reversibly
(chair vetoes in /training). The deepseek+gemini panel only emits style_method
and only on 2/2 consensus, so this is exactly the gate's criterion; substance is
already filtered out and skipped, and routes through the strict halacha gate.
- SUBSTANCE (halacha/precedent/fact) stays a HARD chair gate — unchanged.
style_lesson_panel.py: _review_status_for(category) sets the gate explicitly
(approved for _STYLE_CATEGORIES, else proposed); the apply loop passes it to
db.add_decision_lesson (which already accepts review_status — no DB change).
Spec: INV-LRN1 rewritten as the graduated gate (hard for substance; reversible
auto-flow for consensus style — still "under user control" via veto, per
NCSC/CEPEJ); §0.6 updated to match.
A one-time backfill of the 81 existing panel style lessons to 'approved' runs
separately (DB, post-merge).
Invariants: INV-LRN1 (amended — graduated), INV-LRN5 (style-only, substance never
auto-flows), G10 (human control preserved as reversible veto).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
renumber_cases.py step 6 only renamed the Paperclip project, leaving the
case↔issue linkage on the OLD case number. get_case_issues() keys on two
text surfaces the script never touched:
• plugin_state.legal-case-number (value_json) — the authoritative linkage
• issues.title — the '[ערר {cn}] …' tag the title-path lookup matches
Result: after a renumber the issues kept the old number, get_case_issues
returned [], and post-final actions (run-learning / run-halacha) silently
skipped with reason "no_issue" — surfaced in the UI as "לא הופעלה למידה".
Discovered on case 8137-11-24 (renamed from 8137-24).
Now the Paperclip step rewrites all three surfaces (projects.name +
plugin_state + issue titles); inspect_paperclip + the dry-run report show
the linkage-row and issue-title counts that will be rewritten.
The 11-case migration already ran; the matching DB rows were fixed
manually (incl. this commit's logic) for all stale cases. This change is
for any future renumber.
Invariants: G1 (normalize at source — the linkage, not a read-time
workaround). G12 note: this is a one-time host-side migration script that
already connects directly to the Paperclip DB by design (below the platform
port, like the existing projects/MinIO/Gitea surgery) — it does not add a
parallel runtime path through agent_platform_port.py.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Inline # comments in gitignore are not supported — they were silently
breaking three patterns (data/checkpoints/, data/adapter-migration-state.json,
.claude/agents/.generated/). Moved comments to their own lines and added
missing entries for runtime dirs (data/audit/, data/logs/, etc.) and
temp files (.interaction_tmp.json, .design-build/, .taskmaster bak files).
Also tracks previously untracked legitimate files: scripts, tests, docs,
skills references, .env.example, taskmaster templates.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Switches the scanned-PDF fallback from Google Cloud Vision to
Mistral OCR (mistral-ocr-latest) for better Hebrew accuracy and
robustness against broken embedded OCR layers (e.g. case 1044-03-26
which returned English garbage through Vision).
Routing strategy (document-level, not per-page):
- PyMuPDF extracts all pages; pages that pass _text_quality_ok()
use PyMuPDF output directly (free, ~50ms).
- If ANY page fails quality → Mistral OCR called once for the whole
PDF, returning per-page Markdown for all pages (consistent format,
no plain-text/Markdown mix within a document).
Markdown output preserved: Mistral returns ## headers and |tables|;
chunker updated to recognise ATX Markdown headers (##/###) as section
boundaries in _split_into_sections().
Config: GOOGLE_CLOUD_VISION_API_KEY → MISTRAL_API_KEY; allowlist
updated vision.googleapis.com → api.mistral.ai.
MISTRAL_API_KEY added to Coolify container env.
Invariants: G1 (single OCR fallback path, not parallel), G2 (no
duplicate extractor route), INV-AH (Mistral handles gershayim
natively; quote-fix only on PyMuPDF path).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The internal citation graph fed only RANKING (raw in-degree), and the per-citation
TREATMENT was never classified — so a precedent distinguished N times got the same
authority boost as one followed N times (INV-COR2 violation), and the signal never
reached the agents' reasoning. Wires the full path:
Phase 1 — scripts/classify_citation_treatments.py: classify each linked edge's
treatment (followed/distinguished/…) from its match_context via
corroboration.classify_treatment (Opus 4.8 @ xhigh, local), filling
precedent_internal_citations.treatment. Idempotent.
Phase 2 — db.refresh_verified_layer: count only NON-negative treatments toward
verified/cite_count (INV-COR2/COR4). Unclassified counts as neutral-positive so
the signal degrades gracefully before classification runs.
Phase 3 — db.citation_authority(ids): per-precedent {total, positive, negative,
unclassified, by_treatment}. Surfaced as `cited_by` in search_precedent_library
hits and precedent_library_get, and `treatment` per incoming citation.
Phase 4 — legal-researcher/analyst/writer prompts: weigh & ARGUE authority
("הלכה שאומצה ב-N החלטות ועדת-ערר"), flag distinguished/overruled, never invent
the count (INV-AH; writer is read-only of the analyst).
Auto-approval stays kill-switched off (chair gate preserved, INV-G10). No schema
change (treatment column already existed). Operational: run the classifier +
refresh_verified_layer over the 379 edges, then sync agents across companies.
Invariants: G2 (one classifier + one authority query, reused), INV-COR2/COR3/COR4
(negative never corroborates; point-specific; ≥N), INV-G10 (no auto-approval),
INV-AH (no invented numbers).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Phase C — scripts/cull_principles.py: re-adjudicates every existing 'original'
principle with the SAME panel regime (panel_keep_score → classify → apply_cap),
reversible (CSV backup + rejected canonical recoverable), usage-throttled.
panel_extraction.panel_keep_score + apply_cap (shared, G2). Dry-run on 3
decisions: 37→15 survive.
Phase D — services/principles.py: source-derived label הלכה (binding court) /
כלל פרשני (committee) / עיקרון (persuasive); umbrella עקרונות משפטיים. Wired into
canonical_halacha_get/list (principle_class+principle_label). UI string changes
deferred to the Claude Design gate. spec INV-LRN7; SCRIPTS.md; 7 new tests; 428 green.
Phase E needs no new code — synthesis already targets pending_synthesis, which the
cull leaves only on survivors (rejected canonicals → 'rejected').
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Grounded (INV-AH) multi-instance synthesis with drift guard + chair gate
(pending_review, G10). Single path used by backfill, MCP tool, nightly drain.
HELD from production run pending the principles-redesign (rename+cull, #152).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Scope halacha_panel_approve.py to a single case's pending halachot so the
3-judge deliberation (#133/FU-2) can be captured on demand for one case
(e.g. after a fresh extraction) instead of running the whole ~1.5k queue.
Capture-only path unchanged; no auto-approval. Applied before --limit.
Invariants: G1 (scope at source, not a parallel route) · G2 (reuses the
existing panel/capture path) · INV-G10 (chair gate untouched — dry-run
captures to halacha_panel_rounds only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- halacha_batch_reconcile.py --link now exits with error (V41 freezes
equivalent_halachot; equivalences live in canonical_id instead)
- link_equivalent_halachot emits DeprecationWarning (callers ≥ Python 3.2)
- SCRIPTS.md already marks --link as deprecated; no further doc change needed
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces canonical_halachot table: one row per unique legal principle,
replacing the equivalent_halachot bidirectional-link model (V28/G2 improvement).
Per-precedent halachot rows become instances that point to their canonical.
Schema (V41):
- canonical_halachot: canonical_statement, rule_type, practice_areas,
subject_tags, embedding (ivfflat), review_status (pending_synthesis→published),
first_established_in FK → case_law, instance_count.
- halachot: +canonical_id FK, +instance_type (original|citation|application),
+treatment; rule_statement + embedding become nullable for citation instances.
- halacha_citation_corroboration: +canonical_id FK so X11 aggregates at
principle level, not instance level. store_corroboration auto-populates it
via INSERT...SELECT.
New DB functions: create_canonical_halacha, nearest_canonical_halacha
(threshold search for Phase 3 lookup-before-insert), refresh_canonical_instance_count,
get_canonical_halacha (principle + instance list).
Backfill: scripts/backfill_canonical_halachot.py — dry-run by default,
--apply to execute. Uses union-find over equivalent_halachot pairs, picks
canonical representative (corroboration→confidence→earliest), creates canonicals,
sets canonical_id + instance_type on all instances.
Invariants: G2 (equivalent_halachot deprecated post-backfill), INV-G10
(canonical review_status gate), INV-DM7 (authority derived, not stored),
INV-AH (canonical_statement grounded in source statements, pending_synthesis
until chair approves).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
27/28 הלכות, 100% quote-verified, ביטחון חציון 0.86 מול 0.78 של Opus
אך 0 פריטים מתחת ל-0.7 (חשד ל-over-confidence). ריאלי כ-fallback.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
SCRIPTS.md: הסקריפטים החד-פעמיים (25, לפי עמודת Scheduled) הוצאו מסעיף
"## סקריפטים פעילים" אל סעיף חדש "## חד פעמי" בתחתית (לפני .archive),
מקובצים באותם תתי-כותרות ### לפי תת-נושא. כל 79 השורות (54 חוזרות + 25
חד-פעמיות) נשמרו מילה-במילה. הערת-הקונבנציה עודכנה.
/scripts: ה-parser מזהה את סעיף "## חד פעמי" כסקשן נפרד; הרינדור דו-רמתי —
כותרת-סעיף "פעילים" + קבוצות, ואז כותרת-סעיף "חד פעמי" + קבוצות-משנה
מוזחות (ms-5), מקופלות כברירת-מחדל. סטטוס-השורה נגזר מהסעיף.
מאושר דרך שער-העיצוב (מוקאפ 16-scripts עודכן; חיים אישר מראש).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ה-backfill של citation_formatted חשף קריסה ב-apply_to_record: כשפסק-דין
חיצוני מכיל docket שכבר שייך לרשומה כפולה אחרת, נרמול case_number → docket-נקי
נתקל ב-uq_case_law_external_number ומפיל את כל המיזוג (כולל הציטוט).
דוגמה: 'ע"א 3213/97' → '3213/97' שכבר קיים (כפילות נקר).
- db.case_number_collides(case_number, exclude_id) — בודק אם docket כבר שייך
לרשומה לא-internal אחרת (האינדקס החלקי).
- apply_to_record — מדלג על נרמול ה-case_number כשיש התנגשות (כפילות לדדופ
בהמשך, לא ענייננו כאן) וממשיך לכתוב את הציטוט. no-silent-swallow: מתעד warning.
- scripts/backfill_precedent_citations.py — try/except per-row + מונה שגיאות,
כך ששורה אחת לא מפילה את האצווה.
אומת: ריצה-מחדש מלאה ללא קריסה (0 שגיאות); ההתנגשות תועדה ודולגה כצפוי;
פסיקת בית-משפט: 224/228 מולאו, 4 נמנעו (חסר צדדים/תאריך — abstention, INV-AH).
test_fu2b_reconcile ✓.
Invariants: INV-AH (abstention) · G1 (נרמול-בכתיבה נשמר, רק לא קורס) ·
חוקה §6 (אין בליעה שקטה — דילוג מתועד).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
31 שורות case_law עם source_kind='cited_only' (ציטוט-בלבד, ללא full_text/chunks)
נושאות halacha_extraction_status='pending' רק כברירת-מחדל ומזהמות את מונה ה-pending
ובמתזמר/בדף-התפעול — אין להן מה לחלץ.
תיקון (G1 — תיקון-במקור, G2 — מסנן יחיד משותף):
- db.EXTRACTION_ELIGIBLE_PREDICATE — מקור-אמת יחיד ל"שורה ברת-חילוץ" (source_kind
<> 'cited_only' AND יש precedent_chunks). מוחל ב-list_pending_extraction_requests;
#139 יעשה בו שימוש-חוזר ל-reconcile (אותו כלל, לא כפול).
- מוני-snapshot מסננים cited_only: halacha_drain_supervisor.db_snapshot,
web/app.py meta+hal_ext (GROUP BY status).
- reconcile_metadata_status.py מורחב לכסות גם את תור-ההלכות: cited_only→'skipped'
(אותו terminal-state כמו צד-המטא, תור-תאום, G2). בוצע על ה-DB החי: 31 הועברו
ל-'skipped' (metadata כבר היה מיושב — אידמפוטנטי). התפלגות-אחרי: halacha
pending=9 (עבודה אמיתית), skipped=31, completed=309.
בדיקות: test_extraction_queue_eligibility (predicate + list_pending מחיל אותו,
שני ה-kinds). כל 345 בדיקות mcp עוברות. guards נקיים.
Invariants: G1 (terminal-state אמיתי במקור), G2 (predicate יחיד, ללא תור מקביל),
INV-DM1 (stub לא-searchable אינו מועמד-חילוץ), G12 (leak-guard נקי).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
בהעלאה דרך "פסיקה-חסרה" (ענף ועדת-ערר), כשטופס case_number ריק המסלול נפל-לאחור
לציטוט המלא (committee_case_number = case_number.strip() or citation), כך שמחרוזת-
תצוגה עם שמות-צדדים הושתלה בשדה-המזהה — הפרת INV-ID2/INV-ID1 (X1). נצפה על
precedent 1bf0bae0 (ערר 85074-04-25 רפאל לוי/חולון): case_number=85074/0425,
case_name=ציטוט שלם.
תיקון (G1 — נרמול-במקור, G2 — שימוש-חוזר בפרסר הקנוני):
- court_citation.case_number_from_citation(citation) — מחזיר את אסימון-המספר
המנורמל בלבד (classify; '' כשאין מספר). חולץ נכון 85074-04-25 גם מתוך
"ערר (ת\"א 85074-04-25) ...". reuse של הפרסר היחיד, בלי regex מקביל.
- web/app.py (ענף ועדת-ערר): fallback דרך case_number_from_citation; אם אין
מספר — HTTPException 400 "נא להזין מספר-תיק ידנית" במקום השתלת ציטוט-מלא.
- db._canonical_case_number: מוקשח לחלץ את אסימון-המספר (זורק זנב שמות-צדדים),
כך ששדה-המזהה לעולם לא נשמר מזוהם — גם בקריאה ישירה (committee + active cases).
מספר נקי חוזר ללא שינוי; חודש לא מומצא (X1 §1).
- תיקון-נתון: scripts/fix_137_committee_case_number.py (בוצע) — 1bf0bae0:
case_number→85074-04-25, case_name→צדדים, token ב-citation_formatted.
אומת היחיד עם canon(num)≠num ב-internal_committee. אידמפוטנטי.
מחוץ-לתחום (תועד כ-follow-up): מסלול external (precedent_library) משתמש בציטוט-
מלא כמזהה-מורשת — זהו פריט-המיגרציה X1 §5 (138 רשומות external/cited_only),
לא הבאג הזה. prefill ב-UI של /missing-precedents — דורש שער Claude Design.
בדיקות: test_court_citation (case_number_from_citation: party-strip/forms/empty),
test_canonical_case_number (harden). כל 339 בדיקות mcp עוברות. guards נקיים.
Invariants: G1 (נרמול-במקור), INV-ID1/ID2 (מזהה מנורמל, אין ציטוט-מלא כמזהה),
G2 (פרסר יחיד), G12 (leak-guard נקי).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
PR #251 made the OAuth usage endpoint the PRIMARY rate-limit signal and the log
429 only a fallback for when the endpoint is unreachable. Observed 2026-06-15: the
endpoint reported the window <100% (available) while the claude CLI kept 429-ing
("session limit"). The supervisor then read 'rate_limited=false', classified the
drain 'hung', and restart-churned it — RE-EXTRACTING already-completed precedents
under the rate limit and DEGRADING them (e.g. 4624/21 lost halachot 3→1, only
4/18 chunks). delta_done went negative (completed cases reverting).
Fix: a FRESH CLI 429 is ground truth — the call is literally failing.
• ENTER cooldown on EITHER signal (endpoint-exhausted OR fresh 429), so a 429
overrides an endpoint that wrongly reports the window available.
• VETO the early resume while a fresh 429 remains (the endpoint can lie
"available" mid-storm → without the veto we'd bounce straight back to churn).
• DEFAULT_COOLDOWN_MIN=30 when a fresh 429 has no parseable reset time.
While limited the drain STOPS (no 429-hammering, no degrading completed cases) and
re-ignites only once quota is back AND no fresh 429 remains.
Tested: 8 unit-tests over the decision matrix (endpoint×429×stored-cooldown),
incl. the exact tonight case and the veto. py_compile clean.
Immediate mitigation already applied out-of-band: drain stopped + disabled
(drain_controls.disabled) to halt the degradation until this deploys.
Invariants: G1 (fix at source — trust the failing call, not a lagging endpoint),
G2 (same cooldown path, no parallel control). Builds on PR #251.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
