diff --git a/METRICS.md b/METRICS.md
index 26d2b9b..5e85b33 100644
--- a/METRICS.md
+++ b/METRICS.md
@@ -188,3 +188,42 @@ the signal does not catch confidently wrong rankings.
 - Sandbox repo; each case's diffs are small and fit the 3000-char truncation.
 - The baseline is already 90% top-1, leaving little room for runbook context to
   help; retrieval only delivers the right runbook in about half the cases.
+
+## Model comparison and re-run (2026-10-07)
+
+Same eval set v2, no runbooks, 24 cases × 3 trials per model. Raw results:
+`eval/results/baseline_v2_sonnet5_rerun_2026-10-07.json` and
+`eval/results/baseline_v2_haiku45_2026-10-07.json`; reproduce with
+`python -m eval.report <sonnet file> <haiku file>`.
+
+| Metric | Claude Sonnet 5 (re-run) | Claude Haiku 4.5 |
+|---|---|---|
+| Top-1 | **91.7% (66/72)** | 80.6% (58/72) |
+| Top-3 | 100% | 100% |
+| MRR | 0.958 | 0.903 |
+| Null-runbook cases top-1 | 20/21 | 18/21 |
+| Ranking latency, median | 6.5s | 4.5s |
+| Tokens per call, mean (in / out) | 3066 / 665 | 2698 / 536 |
+| Cost per ranking call, list price | ~$0.013 | ~$0.005 |
+
+- **Stability:** the Sonnet 5 re-run (91.7%) matches the Phase 9 baseline
+  (90.3%) within one trial, so the 90% figure is not a lucky run.
+- **Haiku 4.5 is ~2.4x cheaper and ~2s faster per ranking but 11 points less
+  accurate** (paired: 2 case wins, 5 losses, 17 ties). Its losses include
+  `config_reformat_hidden_string` and `worker_queue_renamed` (0/3 each), the
+  cases where the decoy is most surface-plausible. At ~$0.013 per incident
+  the accuracy is worth more than the saving, so **Sonnet 5 stays the
+  default**. `SENTINEL_MODEL` makes the choice configurable.
+- Haiku 4.5 runs without extended thinking by default, while Sonnet 5 thinks
+  adaptively; this compares the models as shipped by default, not at matched
+  reasoning effort.
+- Costs use list prices ($2/$10 per MTok for Sonnet 5, $1/$5 for Haiku 4.5)
+  times the metered tokens, for the ranking call only (the postmortem is a
+  separate call).
+
+**Close-call rule, out of sample.** The 0.1 rule was chosen on the Phase 9
+trials. On the fresh Sonnet 5 re-run it flags **4/6 wrong picks with 7/66
+false alarms** (in-sample it was 9/13 and 9/131), so it holds. On Haiku 4.5 it
+flags 7/14 wrong but 22/58 right picks, and Haiku's raw confidence is identical
+for right and wrong picks (0.93 / 0.93): the rule is calibrated for Sonnet 5
+and should be re-checked if the model changes.
diff --git a/core/services/llm.py b/core/services/llm.py
index 456e4fb..24a3605 100644
--- a/core/services/llm.py
+++ b/core/services/llm.py
@@ -5,7 +5,7 @@
 from dotenv import load_dotenv
 
 load_dotenv()
-MODEL = "claude-sonnet-5"
+MODEL = os.getenv("SENTINEL_MODEL", "claude-sonnet-5")
 
 
 @cache
diff --git a/eval/report.py b/eval/report.py
index 425d2cc..37021ee 100644
--- a/eval/report.py
+++ b/eval/report.py
@@ -40,9 +40,24 @@ def summarize(records: list[dict]) -> dict:
         "n_right": len(right),
         "n_wrong": len(wrong),
         "latency_median_s": statistics.median(r["rank_latency_s"] for r in records) if records else None,
+        # runs before token metering have no usage; report None rather than a fake zero
+        "tokens_in_mean": _mean_usage(records, "input_tokens"),
+        "tokens_out_mean": _mean_usage(records, "output_tokens"),
     }
 
 
+def _mean_usage(records: list[dict], key: str):
+    vals = [r["usage"][key] for r in records if r.get("usage")]
+    return round(statistics.mean(vals)) if vals else None
+
+
+def _label(a: dict, b: dict) -> tuple[str, str]:
+    """Name the two runs by whatever differs: condition for an A/B, model for a model comparison."""
+    if a["condition"] != b["condition"]:
+        return a["condition"], b["condition"]
+    return a["model"], b["model"]
+
+
 def retrieval(records: list[dict]) -> dict:
     """Retrieval is deterministic per case, so score one record per case."""
     per_case = {}
@@ -114,6 +129,8 @@ def print_summary(name: str, res: dict):
         f"wrong {_fmt_conf(s['conf_wrong'])} (n={s['n_wrong']})"
     )
     print(f"  rank latency median {s['latency_median_s']}s (n={s['n']})")
+    if s["tokens_in_mean"] is not None:
+        print(f"  tokens/call  in {s['tokens_in_mean']}, out {s['tokens_out_mean']} (mean)")
     if res.get("aborted"):
         print(f"  WARNING run ABORTED, not a valid measurement: {res['aborted']}")
     truncated = sorted({r["case_id"] for r in res["records"] if r["truncated_diffs"]})
@@ -137,20 +154,21 @@ def print_retrieval(res: dict):
 
 def print_comparison(a: dict, b: dict):
     rows = paired(a["records"], b["records"])
+    la, lb = _label(a, b)
     wins = sum(hb / tb > ha / ta for _, ha, ta, hb, tb in rows)
     losses = sum(hb / tb < ha / ta for _, ha, ta, hb, tb in rows)
-    print(f"\n== Paired per-case top-1 hits ({a['condition']} vs {b['condition']})")
-    print(f"  {'case':<34} {a['condition']:>9} {b['condition']:>9}")
+    print(f"\n== Paired per-case top-1 hits ({la} vs {lb})")
+    print(f"  {'case':<34} {la:>15} {lb:>15}")
     for cid, ha, ta, hb, tb in rows:
         mark = "+" if hb / tb > ha / ta else "-" if hb / tb < ha / ta else " "
-        print(f"  {cid:<34} {ha:>5}/{ta:<3} {hb:>5}/{tb:<3} {mark}")
+        print(f"  {cid:<34} {ha:>11}/{ta:<3} {hb:>11}/{tb:<3} {mark}")
     ties = len(rows) - wins - losses
-    print(f"  {b['condition']} wins {wins}, losses {losses}, ties {ties} (n={len(rows)} cases)")
+    print(f"  {lb} wins {wins}, losses {losses}, ties {ties} (n={len(rows)} cases)")
 
     print("\n== Null-runbook cases only (does irrelevant context hurt?)")
-    for res in (a, b):
+    for label, res in ((la, a), (lb, b)):
         s = summarize([r for r in res["records"] if not r["expected_runbook_id"]])
-        print(f"  {res['condition']:<9} top-1 {_pct(s['top1'], s['n'])}, MRR {s['mrr']:.3f}")
+        print(f"  {label:<15} top-1 {_pct(s['top1'], s['n'])}, MRR {s['mrr']:.3f}")
 
 
 def main(argv=None):
diff --git a/eval/results/README.md b/eval/results/README.md
index 22d69cc..e2c2dac 100644
--- a/eval/results/README.md
+++ b/eval/results/README.md
@@ -6,4 +6,6 @@ Each file records `sentinel_git_sha`, the commit the run used. Phase 9 was rebas
 those SHAs (`7b4dc90`, `2e1eb8d`, `530b73f`) live on the preserved branch `feat/phase9-rag-eval`, not on
 `main`. Do not delete that branch.
 
+The 2026-10-07 model-comparison runs record `dfd2797`, preserved on branch `feat/model-comparison`; keep it too.
+
 Runs under `aborted/` are kept for the record and are never scored as measurements.
diff --git a/eval/results/baseline_v2_haiku45_2026-10-07.json b/eval/results/baseline_v2_haiku45_2026-10-07.json
new file mode 100644
index 0000000..35872f9
--- /dev/null
+++ b/eval/results/baseline_v2_haiku45_2026-10-07.json
@@ -0,0 +1,3736 @@
+{
+  "condition": "baseline",
+  "dry_run": false,
+  "eval_set_version": "v2",
+  "runbook_corpus_version": "v1",
+  "runbooks_ingested": 10,
+  "runbook_floor": 0.35,
+  "model": "claude-haiku-4-5",
+  "sentinel_git_sha": "dfd27973fd1c0e2bb514413e4e258782cc7b48fd",
+  "started_at": "2026-10-07T21:11:11.972865+00:00",
+  "trials": 3,
+  "n_cases": 24,
+  "records": [
+    {
+      "trial": 0,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.78,
+      "usage": {
+        "input_tokens": 2729,
+        "output_tokens": 556
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.12,
+      "usage": {
+        "input_tokens": 2582,
+        "output_tokens": 507
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.36,
+      "usage": {
+        "input_tokens": 2657,
+        "output_tokens": 533
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.75,
+      "usage": {
+        "input_tokens": 2254,
+        "output_tokens": 448
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.11,
+      "usage": {
+        "input_tokens": 2852,
+        "output_tokens": 564
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.44,
+      "usage": {
+        "input_tokens": 2834,
+        "output_tokens": 551
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.09,
+      "usage": {
+        "input_tokens": 2788,
+        "output_tokens": 509
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.72
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.59,
+      "usage": {
+        "input_tokens": 3050,
+        "output_tokens": 535
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.95,
+      "usage": {
+        "input_tokens": 2596,
+        "output_tokens": 513
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 3.72,
+      "usage": {
+        "input_tokens": 2496,
+        "output_tokens": 436
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.7
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.2,
+      "usage": {
+        "input_tokens": 2934,
+        "output_tokens": 512
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.64,
+      "usage": {
+        "input_tokens": 2867,
+        "output_tokens": 548
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.94,
+      "usage": {
+        "input_tokens": 2676,
+        "output_tokens": 572
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.99
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.17,
+      "usage": {
+        "input_tokens": 2610,
+        "output_tokens": 535
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.7
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.41,
+      "usage": {
+        "input_tokens": 2816,
+        "output_tokens": 532
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.19,
+      "usage": {
+        "input_tokens": 2726,
+        "output_tokens": 523
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.06,
+      "usage": {
+        "input_tokens": 2600,
+        "output_tokens": 529
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.5,
+      "usage": {
+        "input_tokens": 3001,
+        "output_tokens": 556
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.23,
+      "usage": {
+        "input_tokens": 2944,
+        "output_tokens": 524
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.98,
+      "usage": {
+        "input_tokens": 2363,
+        "output_tokens": 599
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.88,
+      "usage": {
+        "input_tokens": 2532,
+        "output_tokens": 568
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.03,
+      "usage": {
+        "input_tokens": 2702,
+        "output_tokens": 526
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.48,
+      "usage": {
+        "input_tokens": 2431,
+        "output_tokens": 559
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.45,
+      "usage": {
+        "input_tokens": 2724,
+        "output_tokens": 555
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.33,
+      "usage": {
+        "input_tokens": 2729,
+        "output_tokens": 553
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.22,
+      "usage": {
+        "input_tokens": 2582,
+        "output_tokens": 516
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.44,
+      "usage": {
+        "input_tokens": 2657,
+        "output_tokens": 548
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.7,
+      "usage": {
+        "input_tokens": 2254,
+        "output_tokens": 453
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.68
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.54
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.86,
+      "usage": {
+        "input_tokens": 2852,
+        "output_tokens": 568
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.88,
+      "usage": {
+        "input_tokens": 2834,
+        "output_tokens": 551
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.7
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.03,
+      "usage": {
+        "input_tokens": 2788,
+        "output_tokens": 515
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.78
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 3.91,
+      "usage": {
+        "input_tokens": 3050,
+        "output_tokens": 481
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.99
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.55,
+      "usage": {
+        "input_tokens": 2596,
+        "output_tokens": 534
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.44,
+      "usage": {
+        "input_tokens": 2496,
+        "output_tokens": 429
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.72
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.91,
+      "usage": {
+        "input_tokens": 2934,
+        "output_tokens": 569
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.5,
+      "usage": {
+        "input_tokens": 2867,
+        "output_tokens": 526
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.59,
+      "usage": {
+        "input_tokens": 2676,
+        "output_tokens": 562
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.98,
+      "usage": {
+        "input_tokens": 2610,
+        "output_tokens": 512
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.69,
+      "usage": {
+        "input_tokens": 2816,
+        "output_tokens": 532
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.2,
+      "usage": {
+        "input_tokens": 2726,
+        "output_tokens": 561
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.11,
+      "usage": {
+        "input_tokens": 2600,
+        "output_tokens": 535
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.59,
+      "usage": {
+        "input_tokens": 3001,
+        "output_tokens": 546
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.77,
+      "usage": {
+        "input_tokens": 2944,
+        "output_tokens": 519
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.2,
+      "usage": {
+        "input_tokens": 2363,
+        "output_tokens": 599
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.64,
+      "usage": {
+        "input_tokens": 2532,
+        "output_tokens": 549
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.45,
+      "usage": {
+        "input_tokens": 2702,
+        "output_tokens": 544
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.3,
+      "usage": {
+        "input_tokens": 2431,
+        "output_tokens": 551
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.59,
+      "usage": {
+        "input_tokens": 2724,
+        "output_tokens": 535
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.53,
+      "usage": {
+        "input_tokens": 2729,
+        "output_tokens": 581
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.39,
+      "usage": {
+        "input_tokens": 2582,
+        "output_tokens": 522
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.38,
+      "usage": {
+        "input_tokens": 2657,
+        "output_tokens": 528
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.66,
+      "usage": {
+        "input_tokens": 2254,
+        "output_tokens": 453
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.03,
+      "usage": {
+        "input_tokens": 2852,
+        "output_tokens": 565
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 5.36,
+      "usage": {
+        "input_tokens": 2834,
+        "output_tokens": 615
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.12,
+      "usage": {
+        "input_tokens": 2788,
+        "output_tokens": 513
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.53,
+      "usage": {
+        "input_tokens": 3050,
+        "output_tokens": 542
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.86,
+      "usage": {
+        "input_tokens": 2596,
+        "output_tokens": 522
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 3.81,
+      "usage": {
+        "input_tokens": 2496,
+        "output_tokens": 439
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.44,
+      "usage": {
+        "input_tokens": 2934,
+        "output_tokens": 549
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.58,
+      "usage": {
+        "input_tokens": 2867,
+        "output_tokens": 538
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.19,
+      "usage": {
+        "input_tokens": 2676,
+        "output_tokens": 579
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.99
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 3.81,
+      "usage": {
+        "input_tokens": 2610,
+        "output_tokens": 498
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.59,
+      "usage": {
+        "input_tokens": 2816,
+        "output_tokens": 553
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.65
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.66,
+      "usage": {
+        "input_tokens": 2726,
+        "output_tokens": 550
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.45,
+      "usage": {
+        "input_tokens": 2600,
+        "output_tokens": 558
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.64,
+      "usage": {
+        "input_tokens": 3001,
+        "output_tokens": 548
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.28,
+      "usage": {
+        "input_tokens": 2944,
+        "output_tokens": 540
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.05
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.41,
+      "usage": {
+        "input_tokens": 2363,
+        "output_tokens": 566
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.45,
+      "usage": {
+        "input_tokens": 2532,
+        "output_tokens": 561
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.7
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.58,
+      "usage": {
+        "input_tokens": 2702,
+        "output_tokens": 563
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.72,
+      "usage": {
+        "input_tokens": 2431,
+        "output_tokens": 558
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.72
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.68
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.12
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 4.7,
+      "usage": {
+        "input_tokens": 2724,
+        "output_tokens": 550
+      }
+    }
+  ],
+  "finished_at": "2026-10-07T21:21:23.757228+00:00"
+}
\ No newline at end of file
diff --git a/eval/results/baseline_v2_sonnet5_rerun_2026-10-07.json b/eval/results/baseline_v2_sonnet5_rerun_2026-10-07.json
new file mode 100644
index 0000000..e9d41ac
--- /dev/null
+++ b/eval/results/baseline_v2_sonnet5_rerun_2026-10-07.json
@@ -0,0 +1,3736 @@
+{
+  "condition": "baseline",
+  "dry_run": false,
+  "eval_set_version": "v2",
+  "runbook_corpus_version": "v1",
+  "runbooks_ingested": 10,
+  "runbook_floor": 0.35,
+  "model": "claude-sonnet-5",
+  "sentinel_git_sha": "dfd27973fd1c0e2bb514413e4e258782cc7b48fd",
+  "started_at": "2026-10-07T21:21:32.348350+00:00",
+  "trials": 3,
+  "n_cases": 24,
+  "records": [
+    {
+      "trial": 0,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 8.81,
+      "usage": {
+        "input_tokens": 3108,
+        "output_tokens": 701
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.22,
+      "usage": {
+        "input_tokens": 2900,
+        "output_tokens": 659
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.17,
+      "usage": {
+        "input_tokens": 3034,
+        "output_tokens": 707
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.08,
+      "usage": {
+        "input_tokens": 2555,
+        "output_tokens": 643
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.59,
+      "usage": {
+        "input_tokens": 3255,
+        "output_tokens": 626
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.53,
+      "usage": {
+        "input_tokens": 3170,
+        "output_tokens": 685
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.42,
+      "usage": {
+        "input_tokens": 3161,
+        "output_tokens": 679
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.56,
+      "usage": {
+        "input_tokens": 3504,
+        "output_tokens": 687
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.8,
+      "usage": {
+        "input_tokens": 2902,
+        "output_tokens": 737
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 4.88,
+      "usage": {
+        "input_tokens": 2773,
+        "output_tokens": 534
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.41,
+      "usage": {
+        "input_tokens": 3344,
+        "output_tokens": 759
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.73,
+      "usage": {
+        "input_tokens": 3259,
+        "output_tokens": 615
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 8.55,
+      "usage": {
+        "input_tokens": 3060,
+        "output_tokens": 644
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.42,
+      "usage": {
+        "input_tokens": 2971,
+        "output_tokens": 667
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.5
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.56,
+      "usage": {
+        "input_tokens": 3260,
+        "output_tokens": 790
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.14,
+      "usage": {
+        "input_tokens": 3174,
+        "output_tokens": 656
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.86,
+      "usage": {
+        "input_tokens": 2930,
+        "output_tokens": 641
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.06,
+      "usage": {
+        "input_tokens": 3442,
+        "output_tokens": 555
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.36,
+      "usage": {
+        "input_tokens": 3376,
+        "output_tokens": 691
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.86,
+      "usage": {
+        "input_tokens": 2647,
+        "output_tokens": 724
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.53,
+      "usage": {
+        "input_tokens": 2831,
+        "output_tokens": 718
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.42,
+      "usage": {
+        "input_tokens": 3051,
+        "output_tokens": 615
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.64,
+      "usage": {
+        "input_tokens": 2774,
+        "output_tokens": 665
+      }
+    },
+    {
+      "trial": 0,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.96
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.91,
+      "usage": {
+        "input_tokens": 3095,
+        "output_tokens": 614
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.2,
+      "usage": {
+        "input_tokens": 3108,
+        "output_tokens": 691
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.27,
+      "usage": {
+        "input_tokens": 2900,
+        "output_tokens": 580
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.55,
+      "usage": {
+        "input_tokens": 3034,
+        "output_tokens": 729
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.52,
+      "usage": {
+        "input_tokens": 2555,
+        "output_tokens": 559
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.73,
+      "usage": {
+        "input_tokens": 3255,
+        "output_tokens": 707
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.75,
+      "usage": {
+        "input_tokens": 3170,
+        "output_tokens": 732
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.84,
+      "usage": {
+        "input_tokens": 3161,
+        "output_tokens": 749
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 6.72,
+      "usage": {
+        "input_tokens": 3504,
+        "output_tokens": 656
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.81,
+      "usage": {
+        "input_tokens": 2902,
+        "output_tokens": 635
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 5.02,
+      "usage": {
+        "input_tokens": 2773,
+        "output_tokens": 549
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.5,
+      "usage": {
+        "input_tokens": 3344,
+        "output_tokens": 598
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.28,
+      "usage": {
+        "input_tokens": 3259,
+        "output_tokens": 624
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.52,
+      "usage": {
+        "input_tokens": 3060,
+        "output_tokens": 703
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.98
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.64,
+      "usage": {
+        "input_tokens": 2971,
+        "output_tokens": 614
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.17,
+      "usage": {
+        "input_tokens": 3260,
+        "output_tokens": 765
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.67,
+      "usage": {
+        "input_tokens": 3174,
+        "output_tokens": 684
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.73,
+      "usage": {
+        "input_tokens": 2930,
+        "output_tokens": 613
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.25,
+      "usage": {
+        "input_tokens": 3442,
+        "output_tokens": 599
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.8
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.75
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.72,
+      "usage": {
+        "input_tokens": 3376,
+        "output_tokens": 590
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.59,
+      "usage": {
+        "input_tokens": 2647,
+        "output_tokens": 686
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.06,
+      "usage": {
+        "input_tokens": 2831,
+        "output_tokens": 733
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.44,
+      "usage": {
+        "input_tokens": 3051,
+        "output_tokens": 714
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.39,
+      "usage": {
+        "input_tokens": 2774,
+        "output_tokens": 729
+      }
+    },
+    {
+      "trial": 1,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.69,
+      "usage": {
+        "input_tokens": 3095,
+        "output_tokens": 594
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "auth_issuer_mesh_hosts",
+      "category": "auth_rejection",
+      "culprit_hash": "5e68f8a",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.245
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.236
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5e68f8a",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "45a24d5",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "df0ec4c",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "298be8d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d20025a",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.62,
+      "usage": {
+        "input_tokens": 3108,
+        "output_tokens": 677
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cache_contains_recursion",
+      "category": "infinite_recursion",
+      "culprit_hash": "4584056",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.333
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.31
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.233
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "4584056",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "6484609",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "7e63ea3",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "6cc141a",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "a98de00",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.5,
+      "usage": {
+        "input_tokens": 2900,
+        "output_tokens": 538
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cart_price_field_v2",
+      "category": "null_or_missing_field",
+      "culprit_hash": "d0ab213",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.26
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.207
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.179
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "d0ab213",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "73e1933",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "4fbde4c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2bbbf3f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "f618f59",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.62,
+      "usage": {
+        "input_tokens": 3034,
+        "output_tokens": 729
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "cart_sku_lowercased",
+      "category": "identifier_format",
+      "culprit_hash": "c2ec824",
+      "expected_runbook_id": null,
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.293
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.247
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.229
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c2ec824",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "64dde29",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "d0cb0b3",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91f8d36",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.06,
+      "usage": {
+        "input_tokens": 2555,
+        "output_tokens": 621
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "checkout_deadline_extra_call",
+      "category": "latency_timeout",
+      "culprit_hash": "efebc07",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.487
+        },
+        {
+          "id": "db_failures",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.353
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "efebc07",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "a035477",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "2b765f4",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "be85cc9",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91b8b07",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.67,
+      "usage": {
+        "input_tokens": 3255,
+        "output_tokens": 689
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "config_reformat_hidden_string",
+      "category": "config_type_error",
+      "culprit_hash": "c3629ef",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.246
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.19
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.184
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "c3629ef",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "3af442c",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "976c988",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "a6a3c41",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "89436d3",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.41,
+      "usage": {
+        "input_tokens": 3170,
+        "output_tokens": 700
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "db_fetch_one_leak",
+      "category": "db_pool_exhaustion",
+      "culprit_hash": "af9b861",
+      "expected_runbook_id": "db_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.631
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.35
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "af9b861",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "f46cd8c",
+          "confidence_score": 0.5
+        },
+        {
+          "commit_hash": "1618815",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "6d50d20",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "1bf1024",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.77,
+      "usage": {
+        "input_tokens": 3161,
+        "output_tokens": 649
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "export_tempfile_never_closed",
+      "category": "file_handle_leak",
+      "culprit_hash": "6eea0fc",
+      "expected_runbook_id": "file_handle_exhaustion",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.625
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.432
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.291
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "5cedb21",
+          "confidence_score": 0.55
+        },
+        {
+          "commit_hash": "6eea0fc",
+          "confidence_score": 0.5
+        },
+        {
+          "commit_hash": "b245524",
+          "confidence_score": 0.3
+        },
+        {
+          "commit_hash": "0be651d",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "7453733",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 7.12,
+      "usage": {
+        "input_tokens": 3504,
+        "output_tokens": 695
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "guest_email_normalise",
+      "category": "null_or_missing_field",
+      "culprit_hash": "7dd25fe",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.241
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.208
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.161
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "7dd25fe",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "9519064",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "1c0135b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "30a7434",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "cde392a",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.62,
+      "usage": {
+        "input_tokens": 2902,
+        "output_tokens": 647
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "httpx_028_drops_proxies",
+      "category": "dependency_version",
+      "culprit_hash": "b35322f",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 4,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.294
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.285
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.263
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "741f699",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "b35322f",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "b45189f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "e3601f6",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 5.19,
+      "usage": {
+        "input_tokens": 2773,
+        "output_tokens": 555
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "inventory_cache_key_per_request",
+      "category": "rate_limit_429",
+      "culprit_hash": "526f8dd",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.382
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.381
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.309
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "526f8dd",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "bf4cc5f",
+          "confidence_score": 0.45
+        },
+        {
+          "commit_hash": "93320f2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "2e1b06b",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "d9caa65",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.16,
+      "usage": {
+        "input_tokens": 3344,
+        "output_tokens": 747
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "inventory_v3_payload",
+      "category": "null_or_missing_field",
+      "culprit_hash": "1c29d94",
+      "expected_runbook_id": "null_field_handling",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.33
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.281
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.28
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "1c29d94",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "c6f0797",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "16d2191",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "5823d1f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "fb9b710",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.52,
+      "usage": {
+        "input_tokens": 3259,
+        "output_tokens": 575
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "jwt_leeway_milliseconds",
+      "category": "auth_rejection",
+      "culprit_hash": "2eeb222",
+      "expected_runbook_id": "auth_failures",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.488
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.198
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.159
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "2eeb222",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "2e11931",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "e8548c5",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "a028c62",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "716e22d",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.8,
+      "usage": {
+        "input_tokens": 3060,
+        "output_tokens": 709
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "money_returns_decimal",
+      "category": "json_serialization",
+      "culprit_hash": "b14897b",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.298
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.151
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.13
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "b14897b",
+          "confidence_score": 0.97
+        },
+        {
+          "commit_hash": "dedc41a",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "626c22d",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "cd4a49f",
+          "confidence_score": 0.01
+        },
+        {
+          "commit_hash": "92cb3da",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.31,
+      "usage": {
+        "input_tokens": 2971,
+        "output_tokens": 721
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "order_read_sync_audit",
+      "category": "latency_timeout",
+      "culprit_hash": "8aed86c",
+      "expected_runbook_id": "request_timeouts",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.539
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.369
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.35
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "8aed86c",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "d3d6f87",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "94578cc",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "5abe2e4",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "b7ea08d",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 7.0,
+      "usage": {
+        "input_tokens": 3260,
+        "output_tokens": 794
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "order_sql_missing_param",
+      "category": "sql_param_mismatch",
+      "culprit_hash": "9ef79b2",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.322
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.23
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.196
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "9ef79b2",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "661f871",
+          "confidence_score": 0.2
+        },
+        {
+          "commit_hash": "93ebe2f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "66d5f2f",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "411ddfa",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.53,
+      "usage": {
+        "input_tokens": 3174,
+        "output_tokens": 681
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "page_size_zero_division",
+      "category": "pagination_boundary",
+      "culprit_hash": "6265f24",
+      "expected_runbook_id": "pagination_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.146
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.129
+        },
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.127
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "6265f24",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "ed3861b",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "0d241ff",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "ef48acf",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "debe268",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.61,
+      "usage": {
+        "input_tokens": 2930,
+        "output_tokens": 649
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "payments_body_without_content_type",
+      "category": "http_content_type",
+      "culprit_hash": "33a9e2a",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.274
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.164
+        },
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.143
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "33a9e2a",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "e60edf2",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "91f3040",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "5430f37",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "64ee5a8",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.28,
+      "usage": {
+        "input_tokens": 3442,
+        "output_tokens": 559
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "payments_retry_loop",
+      "category": "rate_limit_429",
+      "culprit_hash": "f07e927",
+      "expected_runbook_id": "rate_limiting",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.357
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.352
+        },
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.323
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f07e927",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "07a85f8",
+          "confidence_score": 0.35
+        },
+        {
+          "commit_hash": "c1fc55f",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "429c4dd",
+          "confidence_score": 0.0
+        },
+        {
+          "commit_hash": "51bc043",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.06,
+      "usage": {
+        "input_tokens": 3376,
+        "output_tokens": 601
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "redis_legacy_lrem_args",
+      "category": "dependency_version",
+      "culprit_hash": "0751831",
+      "expected_runbook_id": "dependency_regressions",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "null_field_handling",
+          "similarity_score": 0.324
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.317
+        },
+        {
+          "id": "auth_failures",
+          "similarity_score": 0.269
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "0751831",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "3059910",
+          "confidence_score": 0.4
+        },
+        {
+          "commit_hash": "fd92e84",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "7b9dd2c",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "957a42e",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.5,
+      "usage": {
+        "input_tokens": 2647,
+        "output_tokens": 660
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "settings_env_quotes",
+      "category": "config_type_error",
+      "culprit_hash": "15aba07",
+      "expected_runbook_id": "config_parse_errors",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "db_failures",
+          "similarity_score": 0.36
+        },
+        {
+          "id": "config_parse_errors",
+          "similarity_score": 0.278
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.241
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "15aba07",
+          "confidence_score": 0.9
+        },
+        {
+          "commit_hash": "cae9a6c",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "423c51f",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "3e36aa4",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "55cfa0d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.88,
+      "usage": {
+        "input_tokens": 2831,
+        "output_tokens": 753
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "since_filter_naive_cutoff",
+      "category": "timezone_mismatch",
+      "culprit_hash": "266c09c",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "pagination_errors",
+          "similarity_score": 0.279
+        },
+        {
+          "id": "dependency_regressions",
+          "similarity_score": 0.224
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.215
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f3ad8b7",
+          "confidence_score": 0.85
+        },
+        {
+          "commit_hash": "266c09c",
+          "confidence_score": 0.6
+        },
+        {
+          "commit_hash": "7f71cc9",
+          "confidence_score": 0.1
+        },
+        {
+          "commit_hash": "ef6b4a1",
+          "confidence_score": 0.02
+        },
+        {
+          "commit_hash": "5948e6d",
+          "confidence_score": 0.02
+        }
+      ],
+      "culprit_rank": 2,
+      "error": null,
+      "rank_latency_s": 7.41,
+      "usage": {
+        "input_tokens": 3051,
+        "output_tokens": 730
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "worker_queue_renamed",
+      "category": "queue_stall",
+      "culprit_hash": "f8d8258",
+      "expected_runbook_id": null,
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "request_timeouts",
+          "similarity_score": 0.411
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.409
+        },
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.383
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "f8d8258",
+          "confidence_score": 0.92
+        },
+        {
+          "commit_hash": "0f497ec",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "110eec6",
+          "confidence_score": 0.05
+        },
+        {
+          "commit_hash": "aacbcd7",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "3434d6b",
+          "confidence_score": 0.01
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 6.36,
+      "usage": {
+        "input_tokens": 2774,
+        "output_tokens": 733
+      }
+    },
+    {
+      "trial": 2,
+      "case_id": "worker_retains_job_payloads",
+      "category": "memory_growth",
+      "culprit_hash": "3037583",
+      "expected_runbook_id": "memory_leaks",
+      "commits_in_window": 5,
+      "truncated_diffs": 0,
+      "retrieved_runbooks": [
+        {
+          "id": "file_handle_exhaustion",
+          "similarity_score": 0.48
+        },
+        {
+          "id": "memory_leaks",
+          "similarity_score": 0.478
+        },
+        {
+          "id": "rate_limiting",
+          "similarity_score": 0.424
+        }
+      ],
+      "ranking": [
+        {
+          "commit_hash": "3037583",
+          "confidence_score": 0.95
+        },
+        {
+          "commit_hash": "947ac52",
+          "confidence_score": 0.25
+        },
+        {
+          "commit_hash": "088e964",
+          "confidence_score": 0.15
+        },
+        {
+          "commit_hash": "15f34e0",
+          "confidence_score": 0.03
+        },
+        {
+          "commit_hash": "63c2633",
+          "confidence_score": 0.0
+        }
+      ],
+      "culprit_rank": 1,
+      "error": null,
+      "rank_latency_s": 5.33,
+      "usage": {
+        "input_tokens": 3095,
+        "output_tokens": 593
+      }
+    }
+  ],
+  "finished_at": "2026-10-07T21:33:51.939699+00:00"
+}
\ No newline at end of file
diff --git a/eval/run_eval.py b/eval/run_eval.py
index aa06c73..74161af 100644
--- a/eval/run_eval.py
+++ b/eval/run_eval.py
@@ -47,14 +47,24 @@ def run_trial(case: dict, condition: str) -> dict:
             "culprit_rank": None,
             "error": None,
         }
+        usage = {}
+        real_chat = llm_analyzer.chat
+
+        def metered_chat(*args, **kwargs):  # records token usage without touching production code
+            resp = real_chat(*args, **kwargs)
+            if getattr(resp, "usage", None):
+                usage.update(input_tokens=resp.usage.input_tokens, output_tokens=resp.usage.output_tokens)
+            return resp
+
         t0 = time.monotonic()
         try:
             # the harness never builds a prompt itself; it calls the production function
-            ranked = (
-                llm_analyzer.rank_suspect_commits(diffs, alert, runbooks=runbooks)
-                if rag
-                else llm_analyzer.rank_suspect_commits(diffs, alert)
-            )
+            with patch.object(llm_analyzer, "chat", metered_chat):
+                ranked = (
+                    llm_analyzer.rank_suspect_commits(diffs, alert, runbooks=runbooks)
+                    if rag
+                    else llm_analyzer.rank_suspect_commits(diffs, alert)
+                )
             record["ranking"] = [
                 {"commit_hash": r["commit_hash"][:7], "confidence_score": r["confidence_score"]}
                 for r in ranked
@@ -65,6 +75,7 @@ def run_trial(case: dict, condition: str) -> dict:
         except llm_analyzer.LLMUnavailable as e:
             record["error"] = f"LLMUnavailable: {e}"  # counted as a miss by report.py, never dropped
         record["rank_latency_s"] = round(time.monotonic() - t0, 2)
+        record["usage"] = usage or None
         return record