P2-02: Precedent DB is exact-match only — no semantic similarity search #12

New issue

Closed

opened 2026-06-16 13:57:01 +00:00 by Artur · 0 comments

Artur commented

2026-06-16 13:57:01 +00:00

Owner

Severity: P2 (Medium)
File: decider/precedent.py

Problem

The precedent database uses SHA256("normalized input") for hash-based lookup. This means:

"npm install express" and "npm install lodash" have different hashes → no match
"restart nginx" and "reload nginx configuration" → no match
Only EXACT normalized duplicates are found

The normalization (lowercase → remove punctuation → sorted words) helps slightly but still misses semantically similar situations.

Fix

Add a fallback similarity search when exact match fails:

Keep exact hash for fast lookups (O(1))
On miss: use embedding-based similarity (SentenceTransformers or local model) to find the nearest precedents
Return top-3 similar precedents with a similarity score
If similarity > 0.85 → apply precedent; if > 0.70 → present it as reference

Alternatively, implement a simpler keyword overlap approach using FTS5 in SQLite.

**Severity**: P2 (Medium) **File**: `decider/precedent.py` ## Problem The precedent database uses SHA256("normalized input") for hash-based lookup. This means: - "npm install express" and "npm install lodash" have different hashes → no match - "restart nginx" and "reload nginx configuration" → no match - Only EXACT normalized duplicates are found The normalization (`lowercase → remove punctuation → sorted words`) helps slightly but still misses semantically similar situations. ## Fix Add a fallback similarity search when exact match fails: 1. Keep exact hash for fast lookups (O(1)) 2. On miss: use embedding-based similarity (SentenceTransformers or local model) to find the nearest precedents 3. Return top-3 similar precedents with a similarity score 4. If similarity > 0.85 → apply precedent; if > 0.70 → present it as reference Alternatively, implement a simpler keyword overlap approach using FTS5 in SQLite.

Artur added the

enhancement

P2-Medium

labels

2026-06-16 13:57:01 +00:00

Artur referenced this issue from a commit

2026-06-16 14:18:13 +00:00

feat: P2-02 — semantic precedent search via FTS5 fallback

Artur closed this issue

2026-06-16 14:18:13 +00:00

Artur referenced this issue from a commit

2026-06-16 14:23:43 +00:00