ms-ai-architect/scripts/kb-eval
Kjell Tore Guttormsen 2935039f44 feat(ms-ai-architect): R7.2 stempling — 9/11 pass born-verified etter re-judge-gate; 2 diskvalifisert → flagged (R11) [skip-docs]
Re-judge-gate (frossen v3.1, 11 blinde Opus-4.8-xhigh subagenter, live MS Learn-fetch,
G6 Layer A dekket alle fetches). Kvalitetsgate på irreversibelt steg: born-verified stempel
krever TO uavhengige alt-grounded pass.

9 ENIGHET (alle claims grounded) → stemplet Verified 2026-07-24 / judge-v3.1:
  apim-azure-front-door-ai, apim-vs-direct-access-comparison, cost-tracking-apim-policies,
  developer-portal-ai-apis, multi-region-ai-gateway-design, streaming-support-apim,
  versioning-ai-api-endpoints, data-versioning-lineage, real-time-streaming-ai

2 UENIGHET (≥1 non-grounded) → pass→flagged, IKKE stemplet (R11 work-list):
  - security-hardening-ai-gateway: #2 source_silent («20+ security policies» usourcet),
    #7 not_grounded R8 (feature = Prompt Injection Protection, ikke «Prompt Shields»;
    indirekte injection-deteksjon ikke dekket)
  - etl-vs-elt-ai: #2 not_grounded R8 (Gold-lag = curated for reports/dashboards, ikke ML-features)

Gaten fanget gull/judge-run-uenighet (jf. gull-friskhet kan invertere adopsjon). Ledger:
94 records, R7.2 9 pass / 40 flagged (validateManifest valid). Stempel = kun Verified/
Verified by-header (body byte-identisk, per-fil invariant-sjekket, atomicWrite). Suite 947/947.
2026-07-24 20:14:45 +02:00
..
data feat(ms-ai-architect): R7.2 stempling — 9/11 pass born-verified etter re-judge-gate; 2 diskvalifisert → flagged (R11) [skip-docs] 2026-07-24 20:14:45 +02:00
lib feat(ms-ai-architect): G3 gull-intern-konsistens-lint + G4 build-instruks-policy — §8 gap-lukk [skip-docs] 2026-06-30 09:55:15 +02:00
apply-skill-op.mjs feat(ms-ai-architect): Spor D steg 1 — apply regenererer hel-korpus score-cache post-mutasjon (TDD) [skip-docs] 2026-06-23 19:51:04 +02:00
build-gold-set.mjs feat(ms-ai-architect): G3 gull-intern-konsistens-lint + G4 build-instruks-policy — §8 gap-lukk [skip-docs] 2026-06-30 09:55:15 +02:00
build-judge-payloads.mjs feat(ms-ai-architect): R7.1 — build-judge-payloads får --claims <path> (per-batch korpus-manifest, TDD; default = bakeoff-manifestet uendret) [skip-docs] 2026-07-18 11:16:22 +02:00
build-sample-frame.mjs feat(ms-ai-architect): Fase 0 steg 1 — deterministisk sample-frame (45 volatil + 10 kontroll) + TDD [skip-docs] 2026-06-26 14:18:53 +02:00
compute-base-rate.mjs feat(ms-ai-architect): Fase 0 steg 3 — base-rate-rapport + GATE: BYGG Fase 3 (scoped/hybrid) [skip-docs] 2026-06-26 17:15:54 +02:00
detect-skill-lifecycle.mjs feat(ms-ai-architect): Sesjon 15 — B2 K10 søsken-scope-ikke-overlapp 2026-06-20 11:59:59 +02:00
eval.mjs feat(ms-ai-architect): Spor 3 Port 3 (CT5) — deterministisk sourcedness erstatter LLM-samplet K8 [skip-docs] 2026-06-29 15:10:14 +02:00
extract-judge-claims.mjs feat(ms-ai-architect): S1 judge bake-off harness + forhåndsregistrert gate (TDD) [skip-docs] 2026-06-26 20:10:58 +02:00
judge-claim-prompt-v2.md feat(ms-ai-architect): S1 v2 targeted iteration — GATE PASS (recall 84.2%, precision 84.2%) [skip-docs] 2026-06-26 21:23:50 +02:00
judge-claim-prompt-v3.1.md feat(ms-ai-architect): G5b gull-friskhets-spot-sjekk LUKKET — 4 v3-FP re-adjudert mot live, ALLE stale gull (v3 flagget korrekt), baseline løftet v3 P89.7/R92.1 → P100/R92.9/0FP; v3.1 forfattet (ren recall-hardning, FP-vakt droppet) [skip-docs] 2026-06-30 13:43:52 +02:00
judge-claim-prompt-v3.md feat(ms-ai-architect): judge-claim-prompt-v3 (R1-R7 mot 8 feilmoduser) — G4 lukket, G1 v3 forfattet [skip-docs] 2026-06-30 10:14:31 +02:00
judge-claim-prompt.md feat(ms-ai-architect): S1 judge bake-off harness + forhåndsregistrert gate (TDD) [skip-docs] 2026-06-26 20:10:58 +02:00
judge-pass-manifest.mjs feat(ms-ai-architect): per-fil judge-pass-manifest (appendJudgedFile + pendingFiles resume-skip, data/ git-tracked, TDD) [skip-docs] 2026-07-04 23:17:01 +02:00
judge-prompt.md feat(ms-ai-architect): Spor 3 Port 3 (CT5) — deterministisk sourcedness erstatter LLM-samplet K8 [skip-docs] 2026-06-29 15:10:14 +02:00
lint-gold-consistency.mjs feat(ms-ai-architect): G3 gull-intern-konsistens-lint + G4 build-instruks-policy — §8 gap-lukk [skip-docs] 2026-06-30 09:55:15 +02:00
plan-skill-op.mjs feat(ms-ai-architect): Sesjon 20 — B3 create_skill (dvelende, konstruktiv op) [skip-docs] 2026-06-20 21:37:19 +02:00
ref-file-audit.py docs(ms-ai-architect): ref-KB audit — verifisert ground truth + reproduserbart audit-skript 2026-06-26 00:25:39 +02:00
run-judge-bakeoff.mjs feat(ms-ai-architect): S1 v2 targeted iteration — GATE PASS (recall 84.2%, precision 84.2%) [skip-docs] 2026-06-26 21:23:50 +02:00
score-skill.mjs feat(ms-ai-architect): Spor D Steg B — skill-quality lifecycle-gate + score-cache (TDD) 2026-06-23 18:26:29 +02:00