| { |
| "chart": "jevbench", |
| "metric": "JevBench v1.4.1 public accuracy (231 items)", |
| "rows": [ |
| { |
| "label": "Jev-Style 0.8B v3", |
| "key": "v3", |
| "accuracy": 0.6406926406926406, |
| "correct": 148, |
| "n": 231, |
| "colour": "#3456F0", |
| "source": "runs/macjev/received/ext_evals/main/jevbench/results.json :: public_accuracy" |
| }, |
| { |
| "label": "Qwen3.5-0.8B Decision Model (M. Ghafiri)", |
| "key": "mghafiri-qwen3.5-0.8b-decision-model", |
| "accuracy": 0.5930735930735931, |
| "correct": 137, |
| "n": 231, |
| "colour": "#C4C4CB", |
| "source": "data/external/jevbench_v1.4.1/results/v1.4.1/jevbench-v1.4.1-results.json :: systems[key=mghafiri-qwen3.5-0.8b-decision-model].public_accuracy", |
| "v3_lead_pts": 4.76 |
| }, |
| { |
| "label": "Laya (ModernBERT-large, 421M)", |
| "key": "laya", |
| "accuracy": 0.5844155844155844, |
| "correct": 135, |
| "n": 231, |
| "colour": "#D0D0D6", |
| "source": "data/external/jevbench_v1.4.1/results/v1.4.1/jevbench-v1.4.1-results.json :: systems[key=laya].public_accuracy", |
| "v3_lead_pts": 5.63 |
| }, |
| { |
| "label": "SimpleJev (Qwen3.5-0.8B)", |
| "key": "simplejev-qwen3.5-0.8b", |
| "accuracy": 0.5454545454545454, |
| "correct": 126, |
| "n": 231, |
| "colour": "#C4C4CB", |
| "source": "data/external/jevbench_v1.4.1/results/v1.4.1/jevbench-v1.4.1-results.json :: systems[key=simplejev-qwen3.5-0.8b].public_accuracy", |
| "v3_lead_pts": 9.52 |
| } |
| ], |
| "footnote": "JevBench v1.4.1, public items only (231). v3: self-run zero-shot with the vendored official harness (commit 24b9b5c); training-pool contamination scan: 0 hits; not an official leaderboard entry. Other rows: public accuracy as published in the board's v1.4.1 results file (github.com/fstandhartinger/jevbench). Shown: Laya plus every Qwen3.5-0.8B-based system on the board. Laya and M. Ghafiri lie inside v3's 95% CI (57.7\u201370.0%, Wilson): those two leads are point estimates, not significant at n = 231.", |
| "v3_run": { |
| "source_path": "runs/macjev/received/ext_evals/main/jevbench/results.json", |
| "field": "public_accuracy (148/231; independent check equal)", |
| "protocol_label": "v3 self-run with the vendored official harness on the 231 public items, zero-shot (contamination scan of the training pool: 0 hits, runs/macjev/external/jevbench/contamination.json); other rows as published in jevbench v1.4.1 results.json (full runs by the board); not an official leaderboard entry", |
| "ci95": [ |
| 0.577, |
| 0.6998 |
| ], |
| "ci_method": "Wilson 95%" |
| }, |
| "board": { |
| "source_path": "data/external/jevbench_v1.4.1/results/v1.4.1/jevbench-v1.4.1-results.json", |
| "protocol_label": "as published by the board (commit 24b9b5c1609a, tag v1.4.1)" |
| } |
| } |
