Chaim
d5ccf03e4c
Includes: - docs/: architecture, block-schema, migration-plan, product-specification - scripts/: bidi_table, decompose-decisions, extract-claims, seed-knowledge, etc. - skill-legal-decision/: SKILL.md + references + block-schema - skill-legal-assistant/: SKILL.md - skill-legal-docx/: SKILL.md + references - .claude/commands/: bidi-table skill - .taskmaster/: task config + PRDs - .gitignore: exclude legacy/, kiryat-yearim/, node_modules/, memory/ Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
119 lines
4.3 KiB
Python
119 lines
4.3 KiB
Python
#!/usr/bin/env python3
|
|
"""Link claims to discussion paragraphs using semantic similarity.
|
|
|
|
For each claim, finds the most similar paragraph in block-yod of the same decision.
|
|
Updates claims.addressed_in_paragraph with the paragraph number.
|
|
"""
|
|
|
|
import asyncio
|
|
import sys
|
|
from pathlib import Path
|
|
from uuid import UUID
|
|
|
|
sys.path.insert(0, str(Path(__file__).parent.parent / "mcp-server" / "src"))
|
|
|
|
from legal_mcp.services.db import get_pool, init_schema, close_pool
|
|
from legal_mcp.services.embeddings import embed_texts
|
|
|
|
|
|
async def main():
|
|
await init_schema()
|
|
pool = await get_pool()
|
|
|
|
async with pool.acquire() as conn:
|
|
# Get all cases with both claims and discussion blocks
|
|
cases = await conn.fetch(
|
|
"""SELECT DISTINCT c.id as case_id, c.case_number
|
|
FROM cases c
|
|
JOIN claims cl ON cl.case_id = c.id
|
|
JOIN decisions d ON d.case_id = c.id
|
|
JOIN decision_blocks db ON db.decision_id = d.id AND db.block_id = 'block-yod' AND db.word_count > 0
|
|
ORDER BY c.case_number"""
|
|
)
|
|
|
|
total_linked = 0
|
|
|
|
for case in cases:
|
|
case_id = case["case_id"]
|
|
case_number = case["case_number"]
|
|
|
|
async with pool.acquire() as conn:
|
|
# Get claims for this case
|
|
claims = await conn.fetch(
|
|
"SELECT id, claim_text, party_role, claim_index FROM claims WHERE case_id = $1 ORDER BY claim_index",
|
|
case_id,
|
|
)
|
|
|
|
# Get discussion paragraphs (split block-yod into paragraphs)
|
|
yod_content = await conn.fetchval(
|
|
"""SELECT db.content FROM decision_blocks db
|
|
JOIN decisions d ON d.id = db.decision_id
|
|
WHERE d.case_id = $1 AND db.block_id = 'block-yod'""",
|
|
case_id,
|
|
)
|
|
|
|
if not yod_content or not claims:
|
|
continue
|
|
|
|
# Split discussion into paragraphs
|
|
disc_paragraphs = [p.strip() for p in yod_content.split("\n") if p.strip() and len(p.strip()) > 30]
|
|
|
|
if not disc_paragraphs:
|
|
continue
|
|
|
|
print(f"\n{case_number}: {len(claims)} טענות ← {len(disc_paragraphs)} פסקאות דיון")
|
|
|
|
# Embed all claims and discussion paragraphs
|
|
claim_texts = [c["claim_text"][:500] for c in claims]
|
|
all_texts = claim_texts + disc_paragraphs
|
|
|
|
embeddings = await embed_texts(all_texts, input_type="document")
|
|
|
|
claim_embeddings = embeddings[:len(claims)]
|
|
disc_embeddings = embeddings[len(claims):]
|
|
|
|
# For each claim, find the best matching discussion paragraph
|
|
linked = 0
|
|
async with pool.acquire() as conn:
|
|
for i, claim in enumerate(claims):
|
|
claim_emb = claim_embeddings[i]
|
|
|
|
# Cosine similarity
|
|
best_score = -1
|
|
best_para_idx = -1
|
|
for j, disc_emb in enumerate(disc_embeddings):
|
|
dot = sum(a * b for a, b in zip(claim_emb, disc_emb))
|
|
norm_a = sum(a * a for a in claim_emb) ** 0.5
|
|
norm_b = sum(b * b for b in disc_emb) ** 0.5
|
|
score = dot / (norm_a * norm_b) if norm_a > 0 and norm_b > 0 else 0
|
|
|
|
if score > best_score:
|
|
best_score = score
|
|
best_para_idx = j
|
|
|
|
if best_para_idx >= 0 and best_score > 0.3:
|
|
# paragraph_number is 1-indexed
|
|
para_num = best_para_idx + 1
|
|
await conn.execute(
|
|
"UPDATE claims SET addressed_in_paragraph = $1 WHERE id = $2",
|
|
para_num, claim["id"],
|
|
)
|
|
linked += 1
|
|
|
|
total_linked += linked
|
|
print(f" קושרו: {linked}/{len(claims)} טענות (ציון מינימלי: 0.3)")
|
|
|
|
# Summary
|
|
async with pool.acquire() as conn:
|
|
total_claims = await conn.fetchval("SELECT count(*) FROM claims")
|
|
linked_claims = await conn.fetchval("SELECT count(*) FROM claims WHERE addressed_in_paragraph IS NOT NULL")
|
|
|
|
await close_pool()
|
|
|
|
print(f"\n{'='*50}")
|
|
print(f"סיכום: {linked_claims}/{total_claims} טענות קושרו לפסקאות דיון ({linked_claims/total_claims*100:.0f}%)")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
asyncio.run(main())
|