Oct 23 26 days ago 16:30 Add the released perch 0.4.8 on Jev to both benchmarks 16:19 Add the Greptile and ReviewBench benchmark data The data behind lakeday-org/perch's .benchmarks: the Greptile benchmark's labels, manifest, saved runs and scores, and ReviewBench's frozen requests, answers, labels and judgments with the rerun's answers and scores. Each folder's README lists its files. Oct 13 18:38 Rerun SubQ with full method context (#43 ) - Rerun SubQ with full method context (516fa6fa1fb672be8216e1a37f7beb13d24bd1f1) 16:03 Add SubQ Snap benchmark results (#42 ) - Add SubQ Snap benchmark results (b31ccab3783d5a73f1ddc4ffbc8ebfd934648f22) Oct 12 19:25 Remove failed requests from benchmark tables (#41 ) - Remove failed requests from benchmark tables (66ecefd4829314346775ade0e9bd65ebfd7f8085)
- Distinguish benchmark models with colors and shapes (1f98f74d7c2eaa2594596dfce5a3cfe6c34f19e5) 17:49 Plot benchmark performance against inference cost (#40 ) - Plot benchmark performance against inference cost (cd67583812688815b1c0b80a9f8ddfc2127885a8) 17:09 Add Mercury and Microsoft decision benchmarks (#39 ) - Add Mercury Decide benchmark results (305c895395ab08325224e076cdc9e50c557dfc92)
- Add Microsoft Decision 1 benchmark results (df01fc6205e02e20269955dbf2593c9a71a69f4a) Oct 8 19:32 Describe the benchmark as a test of SystemOne decision models (#38 ) - Describe the benchmark as a test of SystemOne decision models (2890e5fd2d20dbe0d3be68d600f61a0994b01631) 18:37 Add CC BY 4.0 license and source credits (#37 ) - Add CC BY 4.0 license and source credits (17348c4bcf59b77e983db0b14678867472180fdf) 17:49 Show recall against false positive rate (#30 ) - Align benchmark curves with table alert rules (dcaa711548ae2643f957ad86595059223db7e935)
- Show recall against false positive rate (58161df591b6c02ac1a0cd0630119f22dd5dd35c) 17:36 Update GPT-6 Luna benchmark results (#36 ) - Add Decisions choice benchmark (b7be346ed78ca544ac47b7e356864d6ca80f1594)
- Update GPT-6 Luna benchmark results (6dfdfdd3b71bc14191087358610d0e3dcde9abdd)
- Show model separation in benchmark graphs (9381ba02de0cdb4fa7bb551173a10aa992dab116) Oct 7 02:03 Add OpenAI Decisions benchmark results (#35 ) - Add OpenAI Decisions benchmark results (00c1490ff3cdcf050999766fc8e966907fc675a5)
- Standardize benchmark graphs (221ee22a7ea069655fad6a8563d912f30f7785f9) Oct 5 03:00 Update completed CodeQL benchmark results (#34 ) - Scan all eligible CodeQL benchmark pairs (c8b91c2c8453b21bd55bada2ef0b3d805a51e392)
- Update completed CodeQL benchmark results (e5351c3f7e5bc0df66de5ae4f62efb16a46ff931) Oct 2 21:44 Add GLiNER-2.5-Decide benchmark results (#33 ) - feat: benchmark GLiNER-2.5-Decide (f1540c49c26f00793c43247b5e6ae6c254be9222)
- fix: size Decide benchmark requests with Perch (a5891f1d9b0f02592878a56c584e08657a2a2f44) Oct 1 21:25 Show CodeQL in the bug and security benchmark tables (#32 ) - Include CodeQL in bug and security result tables (d0cb15591a65d1bedc230bb828f9079e353e1a2f) 20:34 Compare CodeQL and Jev on matched public methods (#31 ) - Add matched CodeQL benchmark comparison (ae8330811ada99719df1fba7d37dd98d62bf3f25) Sep 30 19:12 Rank results by recall minus false alerts and rescale charts (#29 ) - Rank benchmark results by recall minus false alerts (ed841cee298d885d1b5b11fbcef27e3ab50ab484)
- Show false positives as percentages of clean methods (f0583ba7a1450e531429eb66d7485002dd97ac6c)
- Highlight benchmark column winners (57cf3a020c0be4413961b62aaf0bb477a14a0614) 17:22 Compare benchmark recall against false alerts (#28 ) - Show benchmark recall versus false-alert tradeoffs (ebf2345c8f2a7d9158feb6f10a210eef12cbc765)
- Clarify plotted alert settings (bc1a9aebad2ff730956381fdf3e8e912ee26c16d) 16:14 Add TypeLLM benchmark results (#27 ) - Add TypeLLM benchmark results (ac01d0697c8a55229f690b5986ea0b173cb6470a) 00:47 Remove benchmark chart from dataset card (#26 ) - Remove benchmark chart from dataset card (42611b0684b0728b51e1ca897f1ed117110d4f3f) 00:43 Update Jev and Liquid bug benchmark results (#25 ) - Update Jev and Liquid bug benchmark results (c636ba3c324d4d2d3980ee364b88f1e477f8bbd5)
- Keep benchmark card neutral (e5675abe06fb7479cb45cdefb1185dc908b8dda1) Sep 29 20:28 Add Liquid d1 benchmark results (#24 ) - Add Liquid d1 benchmark results (401d5c862fbde2cfebfcc653158e781981cc7ba2)
- Add public Lint Bench graphic (22442de02597f7851b26224ee2ccc17105cf78d9) Sep 28 19:37 Add Span-01 results and benchmark runner (#23 ) - Add Span-01 results and benchmark runner (59ceae0f2059cb0570244987e51d015b09d8cbe6)
- Record inference sources for benchmark models (46e065cc8e8f1a6f7f6c6e6d087d5ad3bddef18e)
- Add Jev semantic lint inference cost (a58e8af2800be0dc722c52829b8724f86eaad12b) 18:45 Rank benchmark tables by AP and call out always-alert security models (#22 ) - Rank benchmark tables by AP and call out always-alert security models (7bc45250a742d659f3f2d2cf9e063a1331102ebc) 18:25 Use configured bug and security floors and 70% semantic lint cutoff (#21 ) - Use configured bug and security floors and 70% semantic lint cutoff (b861a3edf636a0bf8d24f00240d4440a9b4e1a1d) 17:56 Add 50% floor results for current Perch question pack across three benchmarks (#20 ) - Add 50% floor results for current Perch question pack across three benchmarks (c3a7d9c73cbd13a5787cfab0416ffbdc8502f4a3) 16:26 Update benchmark runners for current Perch scan questions (#17 ) - Update benchmark runners for current Perch scan questions (6c6b5fd032e62177a5f1bf41cdd23099e38c22ea) Sep 27 21:22 Publish auditable Perch Lint Bench scores (#19 ) - Publish auditable Perch Lint Bench scores (11e019b0cbc9402c0f218de7d53c110f3912db25) 21:21 Add combined Perch Lint Bench results (#18 ) - Add combined Perch Lint Bench results (ae07d8b5842e22f3cd180f58b288f26d5161b2f9) Sep 26 20:14 Clarify metrics and add inference costs (#16 ) - Make benchmark metrics easier to read (4c7eaecdb0159f0108d029a8e00c26a57df4e054)
- Add measured benchmark inference costs (069f4fd231c188352f0a2a88fae86d14ce464518) 19:48 Remove SecureBERT from benchmark rankings (#15 ) - Remove SecureBERT from benchmark rankings (279a525d191066e434ddaeb42796c72c22cacbd9) 19:42 Rank benchmark results by F1 and highlight metric leaders (#14 ) - Rank benchmark results by F1 and highlight metric leaders (a9e4064b67eba75830802b21d008c8d8da6eb511) 05:54 Link Perch website from benchmark card (#13 ) - Link Perch website from benchmark card (bf59137a50708645322be3006ad0863546f2cbdd) 05:51 Introduce Jev and System One classifiers in benchmark overview (#12 ) - Introduce Jev and System One classifiers in benchmark overview (d4beae58ebf9edeee517225f4a81fc39cb7abe0c) 05:47 Add benchmark adapters for released models (#11 ) - Add benchmark adapters for released models (50d9cbaef4a1bfaeb8d3c8108d8dfb384add9f84) 05:38 Publish runnable public benchmark evaluator (#10 ) - Publish runnable public benchmark evaluator (a8dd27f8e3a00809e6c77bb393443cfd51a36d8d) 05:16 Simplify benchmark pair description (#9 ) - Simplify benchmark pair description (46b06cc5ccdc0f8bfa9c0eb0e16c870673163bc0) 05:15 Disable dataset viewer on results card (#8 ) - Disable dataset viewer on results card (8908175f09bc6fb024c0c7c3f932073a47a07f98) 05:14 Remove dataset caveat paragraph from results card (#7 ) - Remove dataset caveat paragraph from results card (eed556d4def8aa43b5ee5e01ccc26aa5a59f740a) 05:13 Remove protocol and checkpoint section from benchmark card (#6 ) - Remove protocol and checkpoint section from benchmark card (2a498d39eed806cf0232e1ff73c3d3cb7ae7f891) 05:11 Describe Perch method-pair benchmark and unify model tables (#5 ) - Describe Perch method-pair benchmark and unify model tables (496776fe145e8585004c9f8575a59d247befdf7b) 05:08 Link model cards and explain result columns (#4 ) - Link model cards and explain result columns (5a826be5c3d1116975609b6536e71d834931e62d) 05:05 Remove private-score limitation paragraph (#3 ) - Remove private-score limitation paragraph (1bb7783830943693efb695c151688689c7c7ec53) 05:01 Clarify pair-order and fixed-cutoff columns (#2 ) - Clarify pair-order and fixed-cutoff columns (a588920c33e09768d096279abb50281cd2062595) 04:59 Publish public and private split benchmark results (#1 ) - Publish public and private split benchmark results (eaa3d5bb74b2202915229cd04b9f712532ca0814) 04:58 initial commit