fix: add gate file rotation to prevent unbounded growth (#628)

The quality gate stores SHA-256 hashes and eval results in gate files.
Without rotation, these files accumulate indefinitely.

Changes:
- Added _rotate_gate_files() function
- Deletes files older than 7 days (GATE_FILE_MAX_AGE_DAYS)
- Caps directory at 50 historical files (GATE_FILE_MAX_COUNT)
- Always preserves eval_gate_latest.json
- Called automatically after each evaluate_candidate()

Closes #628

bin/soul_eval_gate.py

@@ -25,9 +25,11 @@ Usage:
     result = evaluate_candidate(scores_path, baseline_path, candidate_id)
 """
 
+import glob
 import json
+import os
 import sys
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
 from typing import Optional
 
@@ -63,6 +65,10 @@ MAX_METRIC_REGRESSION = -0.15
 # Default paths
 DEFAULT_GATE_DIR = Path.home() / ".timmy" / "training-data" / "eval-gates"
 
+# Gate file rotation settings (fixes #628: hash dedup growth)
+GATE_FILE_MAX_AGE_DAYS = 7       # Delete gate files older than this
+GATE_FILE_MAX_COUNT = 50         # Keep at most this many gate files (excluding latest)
+
 
 def evaluate_candidate(
     scores_path: str | Path,
@@ -239,6 +245,9 @@ def evaluate_candidate(
     latest_file = gate_dir / "eval_gate_latest.json"
     latest_file.write_text(json.dumps(result, indent=2))
 
+    # Rotate old gate files to prevent unbounded growth (#628)
+    _rotate_gate_files(gate_dir)
+
     return result
 
 
@@ -287,6 +296,58 @@ def _find_category_score(
     return None
 
 
+def _rotate_gate_files(gate_dir: Path) -> int:
+    """Rotate and clean up old eval gate files.
+
+    Prevents unbounded growth of the gate file directory by:
+    1. Deleting files older than GATE_FILE_MAX_AGE_DAYS
+    2. Keeping at most GATE_FILE_MAX_COUNT historical files
+    3. Always preserving eval_gate_latest.json
+
+    Returns the number of files deleted.
+    """
+    if not gate_dir.exists():
+        return 0
+
+    deleted = 0
+    now = datetime.now(timezone.utc)
+    cutoff = now - timedelta(days=GATE_FILE_MAX_AGE_DAYS)
+
+    # Find all eval_gate_*.json files, excluding latest
+    pattern = str(gate_dir / "eval_gate_*.json")
+    all_files = glob.glob(pattern)
+    gate_files = [f for f in all_files if not f.endswith("eval_gate_latest.json")]
+
+    # Sort by modification time (oldest first)
+    gate_files.sort(key=lambda f: os.path.getmtime(f))
+
+    for filepath in gate_files:
+        try:
+            mtime = datetime.fromtimestamp(os.path.getmtime(filepath), tz=timezone.utc)
+
+            # Delete if older than max age
+            if mtime < cutoff:
+                os.remove(filepath)
+                deleted += 1
+                continue
+
+        except OSError:
+            pass
+
+    # Enforce max count (delete oldest first)
+    remaining = [f for f in gate_files if os.path.exists(f)]
+    if len(remaining) > GATE_FILE_MAX_COUNT:
+        excess = remaining[:len(remaining) - GATE_FILE_MAX_COUNT]
+        for filepath in excess:
+            try:
+                os.remove(filepath)
+                deleted += 1
+            except OSError:
+                pass
+
+    return deleted
+
+
 # ── CLI ──────────────────────────────────────────────────────────────
 
 def main():