Skip to content

fix: Stage 1 symbol ranking by fuzzy score — MRR@3 0.39→0.58 #25

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
silongtan merged 1 commit into from
Apr 17, 2026
Merged

fix: Stage 1 symbol ranking by fuzzy score — MRR@3 0.39→0.58 #25

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 11 additions & 5 deletions adapters/code_locator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,11 +212,14 @@ def _ground_single(
# collapses to BM25-only.
# 2. Rank symbols within each qualifying file (relevant ones
# come first).
matched_ids: set[int] = set()
matched_scores: dict[int, float] = {}
if tokens:
try:
validated = self._validate_with_threshold(tokens, fuzzy_threshold)
matched_ids = {v["symbol_id"] for v in validated if v.get("symbol_id")}
for v in validated:
sid = v.get("symbol_id")
if sid:
matched_scores[sid] = max(matched_scores.get(sid, 0), v["match_score"])
except Exception as exc:
logger.debug("[ground] fuzzy validate failed for '%s': %s", description[:60], exc)

Expand All @@ -225,10 +228,12 @@ def _ground_single(
try:
rows = db.lookup_by_name(name)
for row in rows:
matched_ids.add(row["id"])
matched_scores[row["id"]] = max(matched_scores.get(row["id"], 0), 100.0)
except Exception as exc:
logger.debug("[ground] symbol name lookup failed for '%s': %s", name, exc)

matched_ids = set(matched_scores)

# Stage 1: multi-file fused retrieval (FC-2 fix, v0.4.6).
# Previously this stage took only the top-1 BM25 hit via next(),
# which collapsed multi-file features to a single anchor — the
Expand Down Expand Up @@ -273,7 +278,7 @@ def _ground_single(
relevant_ids = matched_ids & file_symbol_ids
ranked = sorted(
file_symbols,
key=lambda r: (r["id"] not in relevant_ids, r["start_line"]),
key=lambda r: (r["id"] not in relevant_ids, -matched_scores.get(r["id"], 0)),
)
for row in ranked[:per_file_budget]:
code_regions.append({
Expand All @@ -294,8 +299,9 @@ def _ground_single(
# Unchanged from v0.4.5 semantics.
if not code_regions and matched_ids:
try:
score_ranked_ids = sorted(matched_scores, key=matched_scores.get, reverse=True)
code_regions = self._regions_from_symbol_ids(
sorted(matched_ids)[:max_symbols], db, description,
score_ranked_ids[:max_symbols], db, description,
)
except Exception as exc:
logger.warning("[ground] fuzzy fallback failed: %s", exc)
Expand Down