Commit History

  1. Remove ReviewBench from Perch datasets

    Verified signature
    524dbf2
  2. Stop timed-out review containers

    Verified signature
    7c9b8fb
  3. Remove unused comment generation stages

    Verified signature
    eb0f70d
  4. Support complete Jev corpus captures

    Verified signature
    c50a2c2
  5. Evaluate Jev with source-inspected review comments (part 3)

    Verified signature
    600788b
  6. Evaluate Jev with source-inspected review comments (part 2)

    Verified signature
    3dfbd34
  7. Evaluate Jev with source-inspected review comments

    Verified signature
    dd4332f
  8. fix: investigate vague review findings

    Verified signature
    80cdbb0
  9. docs: record final Jev comparison controls

    Verified signature
    6b953e3
  10. test: compare Jev review techniques

    Verified signature
    ad852b7
  11. fix: preserve completed reviews during retries

    Verified signature
    a60bfc8
  12. fix: score source-reviewed findings without judge API calls

    Verified signature
    2b5741b
  13. Bound explanation context and respect token rate limits

    Verified signature
    3dd6560

  1. Review test and file inputs with Jev and Luna and explain findings from source

    Verified signature
    c378ec4
  2. Pin the scanner with constructor-parameter lookup fixed

    Verified signature
    93e9b49
  3. Add Perch runner for ReviewBench⁠

    Verified signature
    08b16cf
  4. Rerun SubQ with full method context (#43)⁠

    17924c1
  5. Add SubQ Snap benchmark results (#42)⁠

    5d6c95b

  1. Remove failed requests from benchmark tables (#41)⁠

    8908f2d
  2. Plot benchmark performance against inference cost (#40)⁠

    84b026e
  3. Add Mercury and Microsoft decision benchmarks (#39)⁠

    9cb9ca1

  1. Describe the benchmark as a test of SystemOne decision models (#38)⁠

    6b47824
  2. Add CC BY 4.0 license and source credits (#37)⁠

    ece8ea9
  3. Show recall against false positive rate (#30)⁠

    54dd3e9
  4. Update GPT-6 Luna benchmark results (#36)⁠

    3cfc6d5

  1. Add OpenAI Decisions benchmark results (#35)⁠

    8c88255

  1. Update completed CodeQL benchmark results (#34)⁠

    76359b1

  1. Add GLiNER-2.5-Decide benchmark results (#33)⁠

    8e97f11

  1. Show CodeQL in the bug and security benchmark tables (#32)⁠

    7767af1
  2. Compare CodeQL and Jev on matched public methods (#31)⁠

    49665ce

  1. Rank results by recall minus false alerts and rescale charts (#29)⁠

    900138f
  2. Compare benchmark recall against false alerts (#28)⁠

    820e346
  3. Add TypeLLM benchmark results (#27)⁠

    96dc6b1
  4. Remove benchmark chart from dataset card (#26)⁠

    b62ce47
  5. Update Jev and Liquid bug benchmark results (#25)⁠

    aaa090a

  1. Add Liquid d1 benchmark results (#24)⁠

    2567800

  1. Add Span-01 results and benchmark runner (#23)⁠

    286f2b3
  2. Rank benchmark tables by AP and call out always-alert security models (#22)⁠

    6f63fa7
  3. Use configured bug and security floors and 70% semantic lint cutoff (#21)⁠

    a58c648
  4. Add 50% floor results for current Perch question pack across three benchmarks (#20)⁠

    1add04d
  5. Update benchmark runners for current Perch scan questions (#17)⁠

    157d7a8

  1. Publish auditable Perch Lint Bench scores (#19)⁠

    4fa5160
  2. Add combined Perch Lint Bench results (#18)⁠

    7853b6a

  1. Clarify metrics and add inference costs (#16)⁠

    6eaa915
  2. Remove SecureBERT from benchmark rankings (#15)⁠

    3fe5a6d
  3. Rank benchmark results by F1 and highlight metric leaders (#14)⁠

    b1762b8
  4. Link Perch website from benchmark card (#13)⁠

    292c549
  5. Introduce Jev and System One classifiers in benchmark overview (#12)⁠

    1d78f32
  6. Add benchmark adapters for released models (#11)⁠

    90964f9
  7. Publish runnable public benchmark evaluator (#10)⁠

    f459705