Clarify metrics and add inference costs

README.md

@@ -12,7 +12,11 @@ The pairs come from open-source projects in Rust, JavaScript, TypeScript, Python
 
 ## Reading the results
 
-AP (average precision) measures ranking across score cutoffs. For the other metrics, a score of at least 0.5 counts as an alert. Precision is the share of alerts that are correct; recall is the share of issue-bearing methods found; F1 is their harmonic mean. TP is a correct alert, FP a false alert, and FN a missed issue. Pair order is the share of eligible before/after pairs where the before method scores higher than the fixed after method; ties do not count as correct. Each table is ranked by F1 at 0.5. Bold marks the best value in each column (lowest for FP and FN).
+- **AP (average precision):** How well the model ranks issue-bearing methods across all score cutoffs.
+- **0.5 cutoff:** A score of at least 0.5 counts as an alert for precision, recall, F1, TP, FP, and FN.
+- **Precision, recall, and F1:** Precision is the share of alerts that are correct. Recall is the share of actual issues found. F1 combines both into one score.
+- **TP, FP, and FN:** Correct alerts, false alerts, and missed issues, respectively.
+- **Pair order:** The share of eligible before/after pairs where the issue-bearing before method scores higher than the fixed after method. Ties do not count as correct.
 
 ## Bugs
 
@@ -20,21 +24,25 @@ AP (average precision) measures ranking across score cutoffs. For the other metr
 
 [Download public bug pairs](https://huggingface.co/datasets/perchscan/bug-pairs-public)
 
-| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **47.6%** | **49.9%** | **22.0%** | **30.6%** | **410** | 412 | **1451** | 42.5% |
-| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 44.0% | 44.9% | 13.9% | 21.2% | 258 | 317 | 1603 | **46.9%** |
-| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 44.1% | 47.5% | 7.6% | 13.1% | 141 | 156 | 1720 | 32.1% |
-| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 43.7% | 44.8% | 2.3% | 4.4% | 43 | **53** | 1818 | 22.4% |
+| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order | Inference cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **47.6%** | **49.9%** | **22.0%** | **30.6%** | **410** | 412 | **1451** | 42.5% | $0.537 |
+| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 44.0% | 44.9% | 13.9% | 21.2% | 258 | 317 | 1603 | **46.9%** | ~$0.057 |
+| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 44.1% | 47.5% | 7.6% | 13.1% | 141 | 156 | 1720 | 32.1% | — |
+| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 43.7% | 44.8% | 2.3% | 4.4% | 43 | **53** | 1818 | 22.4% | — |
+
+_Sorted by F1 at 0.5. Bold marks the best quality metric in each column; fewer FP and FN are better._
 
 ### Private — 498 pairs
 
-| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **48.6%** | 50.7% | **24.1%** | **32.6%** | **102** | 99 | **322** | **43.4%** |
-| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 45.2% | 44.6% | 17.5% | 25.1% | 74 | 92 | 350 | **43.4%** |
-| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 45.8% | 46.4% | 9.2% | 15.4% | 39 | 45 | 385 | 32.7% |
-| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 45.2% | **60.0%** | 3.5% | 6.7% | 15 | **10** | 409 | 22.7% |
+| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order | Inference cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **48.6%** | 50.7% | **24.1%** | **32.6%** | **102** | 99 | **322** | **43.4%** | $0.119 |
+| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 45.2% | 44.6% | 17.5% | 25.1% | 74 | 92 | 350 | **43.4%** | ~$0.013 |
+| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 45.8% | 46.4% | 9.2% | 15.4% | 39 | 45 | 385 | 32.7% | — |
+| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 45.2% | **60.0%** | 3.5% | 6.7% | 15 | **10** | 409 | 22.7% | — |
+
+_Sorted by F1 at 0.5. Bold marks the best quality metric in each column; fewer FP and FN are better._
 
 ## Security
 
@@ -42,29 +50,36 @@ AP (average precision) measures ranking across score cutoffs. For the other metr
 
 [Download public security pairs](https://huggingface.co/datasets/perchscan/security-pairs-public)
 
-| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 44.0% | 43.6% | **64.9%** | **52.1%** | **741** | 959 | **401** | **41.5%** |
-| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 44.8% | 46.4% | 12.4% | 19.6% | 142 | 164 | 1000 | 30.6% |
-| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **49.2%** | **54.4%** | 10.2% | 17.2% | 117 | 98 | 1025 | 40.5% |
-| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 43.3% | 36.0% | 3.2% | 5.8% | 36 | **64** | 1106 | 22.9% |
+| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order | Inference cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 44.0% | 43.6% | **64.9%** | **52.1%** | **741** | 959 | **401** | **41.5%** | ~$0.033 |
+| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 44.8% | 46.4% | 12.4% | 19.6% | 142 | 164 | 1000 | 30.6% | — |
+| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **49.2%** | **54.4%** | 10.2% | 17.2% | 117 | 98 | 1025 | 40.5% | $0.312 |
+| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 43.3% | 36.0% | 3.2% | 5.8% | 36 | **64** | 1106 | 22.9% | — |
+
+_Sorted by F1 at 0.5. Bold marks the best quality metric in each column; fewer FP and FN are better._
 
 ### Private — 439 pairs
 
-| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 47.1% | 46.9% | **67.8%** | **55.4%** | **270** | 306 | **128** | **52.8%** |
-| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 47.4% | 45.0% | 4.5% | 8.2% | 18 | 22 | 380 | 37.9% |
-| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **48.9%** | **52.6%** | 2.5% | 4.8% | 10 | **9** | 388 | 36.2% |
-| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 47.3% | 43.8% | 1.8% | 3.4% | 7 | **9** | 391 | 29.4% |
+| Model | AP | Precision at 0.5 | Recall at 0.5 | F1 at 0.5 | TP | FP | FN | Pair order | Inference cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| [CLM-v0.1-8B](https://huggingface.co/Contrastive-LM/CLM-v0.1-8B)† | 47.1% | 46.9% | **67.8%** | **55.4%** | **270** | 306 | **128** | **52.8%** | ~$0.013 |
+| [JevK5](https://huggingface.co/alibiserikbay/JevK5) | 47.4% | 45.0% | 4.5% | 8.2% | 18 | 22 | 380 | 37.9% | — |
+| [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) | **48.9%** | **52.6%** | 2.5% | 4.8% | 10 | **9** | 388 | 36.2% | $0.112 |
+| [jev-lite](https://huggingface.co/vagmi/jev-lite) | 47.3% | 43.8% | 1.8% | 3.4% | 7 | **9** | 391 | 29.4% | — |
+
+_Sorted by F1 at 0.5. Bold marks the best quality metric in each column; fewer FP and FN are better._
 
 † CLM-v0.1-8B uses its choice head with a 6,144-token encoder. All models are scored on the same method sides and labels within each track.
 
+Inference cost is for scoring the method sides in that table. Jev uses API-reported billable input tokens at [TypeSafe's published $0.042 per million input tokens](https://typesafe.ai/blog/introducing-system-one-models-and-jev); output tokens are free. CLM is an estimated share of 11 minutes 49 seconds of active Runpod GPU time at $0.59/hour, allocated by input tokens; model startup and idle time are excluded. JevK5 and jev-lite ran on a local Mac Studio, so no metered GPU cost was recorded for them. Costs are not ranked.
+
 ## Files
 
 - [`benchmark.py`](https://huggingface.co/datasets/perchscan/benchmark-results/blob/main/benchmark.py): run a model on the public pairs and calculate the metrics in these tables.
 - [`adapters/`](https://huggingface.co/datasets/perchscan/benchmark-results/tree/main/adapters): example runners for the unmodified open-source models in the tables.
 - `results.jsonl`: one aggregate report per domain, track, model, and protocol.
+- `benchmark-costs.json`: token usage, GPU timing, rates, and per-track cost calculations.
 - `split-summary.json`: split sizes and whole-file SHA-256 values. It contains no private rows or repository list.
 - `predictions/`: public-pair probabilities for auditing the public scores. No private predictions are included.
 

benchmark-costs.json

@@ -0,0 +1,79 @@
+{
+  "benchmark_sha256": {
+    "bug:private": "3df4ade65fedd1128d30bc17c9c9869c8ab9bd526e666156d0e16685e78f8a73",
+    "bug:public": "32bc68bc85d5535ad28e103adbfa0c09c73e09d0f53a53f7ca35ec219a96e6e2",
+    "security:private": "a4b81728e9e520cf1c6f342974bec369d72f5b7d8aabc1c9cf2aceba966f8051",
+    "security:public": "2407059caa3486371749884138b997d89043eec646e548ee8f44ba3ab08514b8"
+  },
+  "models": {
+    "clm-base": {
+      "basis": "Observed active benchmark GPU window times $0.59/hour; allocated to tracks by embedded input tokens; excludes model startup and idle time",
+      "completed_at_utc": "2026-09-26T04:03:07+00:00",
+      "elapsed_seconds": 709,
+      "gpu": "RTX PRO 6000 Blackwell MIG",
+      "rate_usd_per_hour": 0.59,
+      "started_at_utc": "2026-09-26T03:51:18+00:00",
+      "total_usd": 0.116197,
+      "tracks": {
+        "bug:private": {
+          "embedded_input_tokens": 1939404,
+          "method_sides": 996,
+          "usd": 0.012813
+        },
+        "bug:public": {
+          "embedded_input_tokens": 8686196,
+          "method_sides": 4440,
+          "usd": 0.057386
+        },
+        "security:private": {
+          "embedded_input_tokens": 1903206,
+          "method_sides": 878,
+          "usd": 0.012574
+        },
+        "security:public": {
+          "embedded_input_tokens": 5059264,
+          "method_sides": 2622,
+          "usd": 0.033424
+        }
+      }
+    },
+    "gemma-jev-lite-base": {
+      "basis": "Local Mac Studio run; no metered GPU charge recorded",
+      "tracks": {}
+    },
+    "jev-1.13.0": {
+      "basis": "API-reported billable input tokens times published $0.042 per million input tokens; output tokens free",
+      "rate_usd_per_million_input_tokens": 0.042,
+      "tracks": {
+        "bug:private": {
+          "input_tokens": 2841996,
+          "method_sides": 996,
+          "output_tokens": 20916,
+          "usd": 0.119364
+        },
+        "bug:public": {
+          "input_tokens": 12780147,
+          "method_sides": 4440,
+          "output_tokens": 93240,
+          "usd": 0.536766
+        },
+        "security:private": {
+          "input_tokens": 2672270,
+          "method_sides": 878,
+          "output_tokens": 18438,
+          "usd": 0.112235
+        },
+        "security:public": {
+          "input_tokens": 7427799,
+          "method_sides": 2622,
+          "output_tokens": 55062,
+          "usd": 0.311968
+        }
+      }
+    },
+    "jevk5-base": {
+      "basis": "Local Mac Studio run; no metered GPU charge recorded",
+      "tracks": {}
+    }
+  }
+}