Skip to content

v0.2.0 — judgment layer: EVAL_PLAN + RELATIONS + RISKS

Choose a tag to compare

View all tags
@gasyoun gasyoun released this 02 Jul 09:09
v0.2.0
0bafcf7

The three judgment-tier plans queued in .ai_state.md, authored by Fable 5 (claude-fable-5, 02-07-2026). The P1 execution session is now fully gated.

  • EVAL_PLAN.md — ungameable quality gates: 8 anti-gaming ground rules; G-SEG 200-form stratified segmentation gold (incl. out-of-DCS contamination holdout + calibration rule); G-RENDER adversarial goldens (accented PWG key2, -L{lnum} homonyms, densest MW <ls> cards, FINDINGS §36/§39 traps); G-SALT parity tolerances vs csl-apidev's agni/indra/ka envelopes; G-SCAN page-truth beyond HTTP 200; every USE_CASES.md Accept: line as a named test T-UC1…T-UC13.
  • RELATIONS.md — ecosystem diplomacy: Meyer permission ask drafted (7 self-digitized indices off-limits without written yes); Cologne framing ("kosha serves your Salt standard"); Ambuda/vidyut + C-SALT/CCeH give-backs; binding upstream-vs-track-3 table; contact registry.
  • RISKS.md — pre-mortem R1–R12: @data_version airtight only under 4 new commitments (incl. Zenodo mirroring moved up from P7); scan-link page-truth; csl-sqlite lag surfacing; bus factors; license geometry.

Full changelog: CHANGELOG.md

Assets 2
Loading