Merge pull request 'feat(style-acq T1-T3): קורפוס-דוגמאות של דפנה לכותב (style_exemplars)' (#80) from worktree-style-acquisition-mvp into main

mcp-server/src/legal_mcp/services/block_writer.py

@@ -725,19 +725,43 @@ async def _build_precedents_context(
     style_parts: list[str] = []
     caselaw_parts: list[str] = []
     case_law_ids: list[str] = []
+    # block → golden-ratio section, for targeted exemplar retrieval (T2)
+    _BLOCK_SECTION = {
+        "block-vav": "background", "block-zayin": "claims",
+        "block-yod": "discussion", "block-yod-alef": "summary",
+    }
     try:
         case = await db.get_case(case_id)
         case_number = case.get("case_number", "") if case else ""
         subject = case.get("subject", "") if case else ""
+        practice_area = case.get("practice_area", "") if case else ""
+        decision = await db.get_decision_by_case(case_id)
+        outcome = (decision or {}).get("outcome", "")
         query = f"דיון משפטי בנושא {subject}" if subject else "דיון משפטי ועדת ערר"
         query_emb = await embeddings.embed_query(query)
+        section = _BLOCK_SECTION.get(block_id)
 
-        # Stream 1: paragraph_embeddings — Dafna's own prose (STYLE exemplars, not content)
+        # Stream 1a (PRIMARY): Dafna's own block-level prose from her corpus
+        # (style_exemplars) — matched by section + outcome + practice_area (T2/T3).
+        if section:
+            exemplars = await db.search_style_exemplars(
+                query_embedding=query_emb, section=section,
+                outcome=outcome or None, practice_area=practice_area or None, limit=6,
+            )
+            exemplars = [e for e in exemplars if e.get("decision_number", "") != case_number]
+            for e in exemplars[:4]:
+                style_parts.append(
+                    f"[דוגמת-סגנון (מבנה/קול בלבד — התאם, אל תעתיק תוכן) — "
+                    f"{e.get('decision_number', '?')}, {section}, "
+                    f"outcome={e.get('outcome') or '—'}]\n{e['paragraph_text'][:1100]}"
+                )
+
+        # Stream 1b: paragraphs from pipeline cases (legacy path; may be empty)
         para_results = await db.search_similar_paragraphs(
             query_embedding=query_emb, limit=10, block_type="block-yod",
         )
         para_results = [r for r in para_results if r.get("case_number", "") != case_number]
-        for r in para_results[:4]:
+        for r in para_results[:2]:
             style_parts.append(
                 f"[דוגמת-סגנון — החלטת {r.get('case_number', '?')} "
                 f"{r.get('case_title', '')}, בלוק {r.get('block_type', '')}]\n{r['content'][:500]}"

mcp-server/src/legal_mcp/services/db.py

@@ -1204,6 +1204,28 @@ CREATE INDEX IF NOT EXISTS idx_draft_final_pairs_case ON draft_final_pairs(case_
 CREATE INDEX IF NOT EXISTS idx_draft_final_pairs_status ON draft_final_pairs(status);
 """
 
+SCHEMA_V27_SQL = """
+-- style_exemplars (T1-T3): block-level paragraphs from Dafna's OWN decisions
+-- (style_corpus + internal_committee finals), embedded for retrieval as
+-- style exemplars at write-time. Purpose-built so we DON'T fabricate synthetic
+-- cases just to reuse decision_paragraphs. INV-LRN5: style material only — the
+-- writer is told to adapt structure/voice, copy only boilerplate, never substance.
+CREATE TABLE IF NOT EXISTS style_exemplars (
+    id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
+    decision_number TEXT DEFAULT '',
+    source TEXT DEFAULT '',                 -- style_corpus | internal_committee
+    practice_area TEXT DEFAULT '',
+    outcome TEXT DEFAULT '',                -- rejection | partial_acceptance | full_acceptance | ''
+    section TEXT DEFAULT 'other',           -- background | claims | discussion | summary | other
+    paragraph_text TEXT NOT NULL,
+    word_count INTEGER DEFAULT 0,
+    embedding vector(1024),
+    created_at TIMESTAMPTZ DEFAULT now()
+);
+CREATE INDEX IF NOT EXISTS idx_style_exemplars_section ON style_exemplars(section);
+CREATE INDEX IF NOT EXISTS idx_style_exemplars_decision ON style_exemplars(decision_number, source);
+"""
+
 
 async def _run_schema_migrations(pool: asyncpg.Pool) -> None:
     async with pool.acquire() as conn:
@@ -1234,7 +1256,8 @@ async def _run_schema_migrations(pool: asyncpg.Pool) -> None:
         await conn.execute(SCHEMA_V24_SQL)
         await conn.execute(SCHEMA_V25_SQL)
         await conn.execute(SCHEMA_V26_SQL)
-    logger.info("Database schema initialized (v1-v26)")
+        await conn.execute(SCHEMA_V27_SQL)
+    logger.info("Database schema initialized (v1-v27)")
 
 
 async def init_schema() -> None:
@@ -2329,6 +2352,85 @@ async def list_draft_final_pairs(status: str | None = None, limit: int = 200) ->
     return [dict(r) for r in rows]
 
 
+async def insert_style_exemplar(
+    decision_number: str, source: str, practice_area: str, outcome: str,
+    section: str, paragraph_text: str, word_count: int, embedding: list[float],
+) -> None:
+    """Insert one block-level style exemplar (T1 backfill)."""
+    pool = await get_pool()
+    async with pool.acquire() as conn:
+        await conn.execute(
+            """INSERT INTO style_exemplars
+                   (decision_number, source, practice_area, outcome, section,
+                    paragraph_text, word_count, embedding)
+               VALUES ($1, $2, $3, $4, $5, $6, $7, $8)""",
+            decision_number, source, practice_area, outcome, section,
+            paragraph_text, word_count, str(embedding),
+        )
+
+
+async def delete_style_exemplars(decision_number: str, source: str) -> int:
+    """Idempotent backfill: clear a decision's exemplars before re-inserting."""
+    pool = await get_pool()
+    async with pool.acquire() as conn:
+        res = await conn.execute(
+            "DELETE FROM style_exemplars WHERE decision_number = $1 AND source = $2",
+            decision_number, source,
+        )
+    try:
+        return int(res.split()[-1])
+    except (ValueError, IndexError):
+        return 0
+
+
+async def search_style_exemplars(
+    query_embedding: list[float],
+    section: str | None = None,
+    outcome: str | None = None,
+    practice_area: str | None = None,
+    limit: int = 6,
+) -> list[dict]:
+    """Retrieve Dafna's own block-level paragraphs as STYLE exemplars (T2).
+    Filters by section (block) + optionally outcome/practice_area for the closest
+    match to the block being written. Soft filters: outcome/practice_area narrow but
+    never zero-out — section is the hard filter."""
+    pool = await get_pool()
+    conditions, params, idx = [], [query_embedding, limit], 3
+    if section:
+        conditions.append(f"section = ${idx}"); params.append(section); idx += 1
+    if outcome:
+        conditions.append(f"(outcome = ${idx} OR outcome = '')"); params.append(outcome); idx += 1
+    if practice_area:
+        conditions.append(f"(practice_area = ${idx} OR practice_area = '')"); params.append(practice_area); idx += 1
+    where = f"WHERE {' AND '.join(conditions)}" if conditions else ""
+    sql = f"""
+        SELECT decision_number, source, section, outcome, practice_area,
+               paragraph_text, word_count,
+               1 - (embedding <=> $1) AS score
+        FROM style_exemplars
+        {where}
+        ORDER BY embedding <=> $1
+        LIMIT $2
+    """
+    async with pool.acquire() as conn:
+        rows = await conn.fetch(sql, *params)
+    return [dict(r) for r in rows]
+
+
+async def count_style_exemplars() -> dict:
+    """Coverage check for the backfill."""
+    pool = await get_pool()
+    async with pool.acquire() as conn:
+        total = await conn.fetchval("SELECT count(*) FROM style_exemplars")
+        by_section = await conn.fetch(
+            "SELECT section, count(*) AS n FROM style_exemplars GROUP BY section ORDER BY n DESC"
+        )
+        decisions = await conn.fetchval(
+            "SELECT count(DISTINCT decision_number) FROM style_exemplars"
+        )
+    return {"total": total, "decisions": decisions, "by_section": [dict(r) for r in by_section]}
+
+
 async def upsert_style_pattern(
     pattern_type: str,
     pattern_text: str,

scripts/SCRIPTS.md

@@ -45,6 +45,7 @@
 | `backfill_multimodal_precedents.py` | python | Backfill voyage-multimodal-3 page embeddings על רשומות `case_law` (external_upload + internal_committee) שחסרות `precedent_image_embeddings`. בונה אינדקס קבצים מ-`data/precedent-library/` ו-`data/internal-decisions/`, מנסה התאמה לפי tokens של מספרי תיק (כולל parts-match לפורמטים שונים של Nevo doc-id). מדלג על רשומות בלי קובץ-מקור או עם MD בלבד (PyMuPDF לא מרנדר MD). תומך `--dry-run` (default) / `--apply` / `--only external_upload\|internal_committee` / `--limit N`. רץ בקונטיינר (יש `/data` + Voyage env). **הופעל 2026-05-26**: 70 חסרים → 26 backfilled (503 pages, ~$0.21 voyage tokens), 44 אין-קובץ-מקור. ניתן להריץ שוב אחרי שיועלו עוד PDF/DOCX לספרייה | ידני |
 | `monitor_halacha_quality.py` | python | מנטר איכות חילוץ הלכות. בודק drift של `avg(confidence)` בין baseline היסטורי לחלון אחרון. מחזיר JSON מטריקות + alert ב-stderr אם drift > threshold (ברירת מחדל 5%). 2 סדרות: trusted (approved+published) ו-all_extracted. תומך `--window N` / `--threshold X` / `--min-sample N` / `--silent` / `--exit-on-alert`. רץ ב-container או מקומית עם `mcp-server/.venv` (אין תלות ב-LLM, רק SQL). **תזמון מומלץ**: `0 8 * * 1` (יום ראשון 08:00, שבועי) | `0 8 * * 1` (לתזמן) |
 | `audit_training_corpus.py` | python | audit של `style_corpus` — לכל החלטה: שדות מטא-דאטה מאוכלסים (`summary`/`outcome`/`key_principles`/`appeal_subtype`/`subject_categories`), קישור ל-`documents` (FK + chunks + embeddings). מפיק `data/audit/corpus-YYYY-MM-DD.json` + summary בקונסול. דרוש `POSTGRES_URL` או POSTGRES_*. אין תלויות חיצוניות מלבד asyncpg. **רץ מהמכונה המקומית** (לא קונטיינר) — חיבור ישיר ל-Postgres :5433 | ידני / קדם-עבודה לפני enrichment של מטא-דאטה |
+| `backfill_style_exemplars.py` | python | **T1 (style-acquisition)** — מאכלס `style_exemplars` מקורפוס דפנה (`style_corpus` + `internal_committee` chair=דפנה): מפצל לסעיפים (`chunker._split_into_sections`) → פסקאות (25-450 מילים) → embed (Voyage) → שמירה עם `section`/`outcome`/`practice_area`. מאפשר לכותב לאחזר פסקאות-בלוק אמיתיות של דפנה (T2/T3). מקור-סגנון בלבד (INV-LRN5). אידמפוטנטי (מנקה per-decision). `--dry-run` (default) / `--apply`. דורש POSTGRES_URL + Voyage. **רץ מקומית** (venv). | ידני (`python scripts/backfill_style_exemplars.py --apply`) |
 
 ## תיקיית `.archive/` — סקריפטים שהושלמו
 

scripts/backfill_style_exemplars.py

@@ -0,0 +1,131 @@
+#!/usr/bin/env python3
+"""T1 — אכלוס style_exemplars מקורפוס דפנה (style_corpus + internal_committee).
+
+מפצל כל החלטה של דפנה לסעיפים (chunker._split_into_sections), ומכל סעיף לפסקאות,
+מטמיע (Voyage) ושומר ב-style_exemplars עם section/outcome/practice_area — כדי
+שהכותב יוכל לאחזר פסקאות-בלוק אמיתיות של דפנה בזמן כתיבה (T2/T3).
+
+מקור-סגנון בלבד (INV-LRN5) — לא מהות. אידמפוטנטי: מנקה לכל decision לפני הכנסה.
+
+שימוש:
+    python3 scripts/backfill_style_exemplars.py            # dry-run (סופר בלבד)
+    python3 scripts/backfill_style_exemplars.py --apply    # מטמיע ושומר
+
+דורש POSTGRES_URL + מפתח Voyage בסביבה (כמו שאר ה-MCP).
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import logging
+
+from legal_mcp.services import db, embeddings
+from legal_mcp.services.chunker import _split_into_sections
+
+logging.basicConfig(level=logging.INFO, format="%(message)s")
+log = logging.getLogger("backfill_exemplars")
+
+# chunker section_type → style_exemplars.section
+_SECTION_MAP = {
+    "facts": "background",
+    "appellant_claims": "claims",
+    "respondent_claims": "claims",
+    "legal_analysis": "discussion",
+    "conclusion": "summary",
+    "ruling": "summary",
+    "intro": "other",
+    "other": "other",
+}
+
+MIN_WORDS = 25      # skip tiny fragments
+MAX_WORDS = 450     # skip over-long blobs (likely un-split)
+MAX_PER_SECTION = 15
+
+
+def _paragraphs(section_text: str) -> list[str]:
+    """Split a section into paragraph units (blank-line separated; fall back to lines)."""
+    raw = [p.strip() for p in section_text.split("\n\n")]
+    if len(raw) <= 1:
+        raw = [p.strip() for p in section_text.split("\n")]
+    out = []
+    for p in raw:
+        wc = len(p.split())
+        if MIN_WORDS <= wc <= MAX_WORDS:
+            out.append(p)
+    return out[:MAX_PER_SECTION]
+
+
+async def _gather_sources() -> list[dict]:
+    """All of Dafna's decisions: style_corpus + internal_committee (chair דפנה)."""
+    pool = await db.get_pool()
+    rows: list[dict] = []
+    async with pool.acquire() as conn:
+        sc = await conn.fetch(
+            "SELECT decision_number, full_text, outcome, practice_area "
+            "FROM style_corpus WHERE full_text <> ''"
+        )
+        for r in sc:
+            rows.append({
+                "decision_number": r["decision_number"] or "",
+                "source": "style_corpus",
+                "full_text": r["full_text"],
+                "outcome": r["outcome"] or "",
+                "practice_area": r["practice_area"] or "",
+            })
+        ic = await conn.fetch(
+            "SELECT case_number, full_text, practice_area FROM case_law "
+            "WHERE source_kind = 'internal_committee' AND coalesce(chair_name,'') LIKE '%דפנה%' "
+            "AND coalesce(full_text,'') <> ''"
+        )
+        for r in ic:
+            rows.append({
+                "decision_number": r["case_number"] or "",
+                "source": "internal_committee",
+                "full_text": r["full_text"],
+                "outcome": "",
+                "practice_area": r["practice_area"] or "",
+            })
+    return rows
+
+
+async def main(apply: bool) -> None:
+    sources = await _gather_sources()
+    log.info("מקורות: %d החלטות של דפנה (style_corpus + internal_committee)", len(sources))
+
+    total_paras = 0
+    for src in sources:
+        units: list[tuple[str, str]] = []  # (section, paragraph)
+        for section_type, section_text in _split_into_sections(src["full_text"]):
+            section = _SECTION_MAP.get(section_type, "other")
+            for para in _paragraphs(section_text):
+                units.append((section, para))
+        if not units:
+            continue
+        total_paras += len(units)
+        log.info("  %-14s %-16s → %d פסקאות", src["source"], src["decision_number"], len(units))
+        if not apply:
+            continue
+
+        await db.delete_style_exemplars(src["decision_number"], src["source"])
+        texts = [u[1] for u in units]
+        vecs = await embeddings.embed_texts(texts, input_type="document")
+        for (section, para), vec in zip(units, vecs):
+            await db.insert_style_exemplar(
+                decision_number=src["decision_number"], source=src["source"],
+                practice_area=src["practice_area"], outcome=src["outcome"],
+                section=section, paragraph_text=para, word_count=len(para.split()),
+                embedding=vec,
+            )
+
+    if apply:
+        cov = await db.count_style_exemplars()
+        log.info("הושלם. style_exemplars: %s", cov)
+    else:
+        log.info("dry-run: %d פסקאות יוטמעו. הרץ --apply לביצוע.", total_paras)
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--apply", action="store_true", help="embed + insert (default: dry-run)")
+    args = ap.parse_args()
+    asyncio.run(main(args.apply))