fix: reports 500, NaN scores, hypothesis dedup blocking

Reports 500 (Internal Server Error):
- app.py: Replace NaN/Infinity->null before json.loads() in both
  /reports and /api/runs routes. Old summary.json files with invalid
  NaN are now handled transparently.
- reports/writer.py: Added _safe_val() helper using math.isfinite().
  All summary.json values now use safe rounding (NaN->0, Inf->0).

No new hypotheses (optimizer stops after 1 iter):
- optimizer_loop.py: Changed dedup from DB-wide history to
  session-scoped (self.session_tested_deltas). Previous runs from
  old sessions no longer block new hypothesis proposals.
- Each tested hypothesis is tracked within the session only.

Result: optimizer now runs multiple iterations, scores are valid
numbers, reports page loads without 500.

app.py

@@ -94,13 +94,23 @@ def history():
 @app.route("/reports/")
 def reports_index():
     """Reports browser page — fixes the 404 on the Reports button."""
+    import json, re
     runs = []
     for run_dir in sorted(REPORTS_DIR.iterdir(), reverse=True) if REPORTS_DIR.exists() else []:
+        if not run_dir.is_dir():
+            continue
         summary = run_dir / "summary.json"
         if summary.exists():
-            import json
             try:
-                runs.append(json.loads(summary.read_text()))
+                txt = summary.read_text(encoding="utf-8")
+                # Fix legacy NaN values (invalid JSON) written before the fix
+                txt = re.sub(r'\bNaN\b', 'null', txt)
+                txt = re.sub(r'\bInfinity\b', 'null', txt)
+                data = json.loads(txt)
+                # Replace None scores with 0 for display
+                data["score"]       = data.get("score") or 0
+                data["score_delta"] = data.get("score_delta") or 0
+                runs.append(data)
             except Exception:
                 pass
     return render_template("reports_index.html", runs=runs[:100])
@@ -114,14 +124,22 @@ def reports_file(filename):
 
 @app.route("/api/runs")
 def runs_list():
+    import json, re
     runs = []
     if REPORTS_DIR.exists():
         for run_dir in sorted(REPORTS_DIR.iterdir(), reverse=True):
+            if not run_dir.is_dir():
+                continue
             summary = run_dir / "summary.json"
             if summary.exists():
-                import json
                 try:
-                    runs.append(json.loads(summary.read_text()))
+                    txt = summary.read_text(encoding="utf-8")
+                    txt = re.sub(r'\bNaN\b', 'null', txt)
+                    txt = re.sub(r'\bInfinity\b', 'null', txt)
+                    data = json.loads(txt)
+                    data["score"]       = data.get("score") or 0
+                    data["score_delta"] = data.get("score_delta") or 0
+                    runs.append(data)
                 except Exception:
                     pass
     return jsonify(runs[:50])

optimizer_loop.py

@@ -66,6 +66,7 @@ class OptimizerLoop:
         self.current_run_id: Optional[str] = None
         self.score_history: list[dict]     = []
         self.run_start_ts: Optional[float] = None
+        self.session_tested_deltas: list[dict] = []  # dedup within this session only
 
         with open(config_path) as f:
             self.cfg = yaml.safe_load(f)
@@ -177,12 +178,11 @@ class OptimizerLoop:
                 self._emit("log", {"level": "warn", "msg": "No actionable findings. Stopping."})
                 break
 
-            # Mutation proposals
-            recent_deltas = store.get_recent_param_deltas(cfg["mutation"]["dedup_lookback_runs"])
+            # Mutation proposals — only dedup within this session
             hypotheses = mutator.propose(
                 findings=findings,
                 current_params=current_params,
-                recent_deltas=recent_deltas,
+                recent_deltas=self.session_tested_deltas,
                 max_proposals=cfg["mutation"]["max_hypotheses_per_cycle"],
             )
 
@@ -249,6 +249,7 @@ class OptimizerLoop:
                              hypothesis=hyp, baseline_score=current_metrics.composite_score)
 
                 store.update_hypothesis_status(hyp.hypothesis_id, "tested", run_id)
+                self.session_tested_deltas.append(hyp.param_delta)  # track for session dedup
 
                 if iteration_best is None or test_metrics.composite_score > iteration_best.composite_score:
                     iteration_best        = test_metrics