core/evals
Shay 283680f110 feat(adr-0044, adr-0045): domain ethics pack + long-context comparison
ADR-0044 — Medical / clinical ethics pack (worked-example domain pack).
Ships packs/ethics/medical_clinical_ethics_v1.json with six commitments
partitioned across all three remediation tiers:
  - refuse: no_dosing_recommendation, no_emergency_triage_authority
  - hedge:  defer_diagnosis_to_clinician, surface_evidence_grade
  - audit:  disclose_no_clinician_relationship, respect_patient_autonomy

Ratified end-to-end through scripts/ratify_ethics_pack.py (PACK_IDS
extended).  Production-mode load via load_ethics_pack succeeds.
ChatRuntime composition includes universal safety floor + every medical
commitment.  tests/test_medical_clinical_ethics_pack.py (8 tests) gates
file existence, sealed report, disjoint refusal/hedge lists, and
pack-swap visibility (default pack does NOT carry medical commitments).

ADR-0045 — Long-context recall: CORE vs transformer baselines.
Adds evals/long_context_cost/comparison_runner.py with a deterministic
needle-in-a-haystack measurement at N ∈ {100, 1_000, 10_000, 100_000}.
CORE recall = 100% at every tested N by exact cga_inner scan.

Paired with frozen citations of published transformer NIAH numbers in
evals/long_context_cost/baselines/transformer_long_context.json:
Claude 2.1 (200k, 50%), GPT-4 Turbo 128k (~71%), Gemini 1.5 Pro (99.7%),
NVIDIA RULER (varies).  Each citation carries source + url.

The two components measure different inputs (synthetic versors vs NL
needles) and are not directly comparable benchmark-for-benchmark.  The
comparison is at the architectural level — exact-scan recall vs
attention-based probabilistic recall.  Scope and limits documented in
the ADR.  tests/test_long_context_comparison.py (5 tests) gates schema,
CORE recall == 100%, and baseline citation presence.

CLI integration: two new demo targets with study-grade preambles.
  - core demo pack-measurements          (ADR-0043 — wired)
  - core demo long-context-comparison    (ADR-0045)
README + docs/PROGRESS.md cheatsheets updated.  docs/decisions/README.md
index extended with ADR-0044 + ADR-0045; pack-layer chain title now
"ADR-0027 through ADR-0045".

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-17 22:31:47 -07:00
..
adversarial_identity docs(identity): empirical finding — fix #3 needs upstream ingest-gate work 2026-05-16 14:23:20 -07:00
articulation_of_status feat(epistemic): realizer-side closure — refusal_calibration + articulation_of_status graduate 2026-05-17 10:12:59 -07:00
audit_tour feat(adr-0042): audit-tour demo — pack-layer story in four scenes 2026-05-17 22:06:45 -07:00
calibration fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
classical_literature_ood fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
cognition feat(evals): Phase 0 — benchmark methodology lock-in and eval framework 2026-05-15 22:36:53 -07:00
compositionality feat(compositionality): compose_relations operator lifts lane 68.8% → 100% 2026-05-16 22:44:06 -07:00
contradiction_detection feat(epistemic): contradiction coherence checker — CONTESTED transitions wired, last Tier 4.5 row closes 2026-05-17 10:36:48 -07:00
cross_domain_transfer feat(algebra): null-preserving versor_apply path + un-skip 2 invariant tests 2026-05-16 21:40:37 -07:00
discourse_paragraph feat(compositionality): compose_relations operator lifts lane 68.8% → 100% 2026-05-16 22:44:06 -07:00
elementary_mathematics_ood fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
english_fluency_ood fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
forward_semantic_control docs(cli): self-explanatory demos — preambles + per-directory READMEs 2026-05-17 16:39:50 -07:00
foundational_biology_ood fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
foundational_physics_ood fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
grammatical_coverage fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
hebrew_fluency feat(phase5): land 5.2–5.7 — six new fluency lanes, parallel sweep 2026-05-16 20:59:31 -07:00
identity_divergence feat(adr-0043): Phase-2 pack measurements — claims → numbers 2026-05-17 22:19:24 -07:00
inference_closure feat(algebra): null-preserving versor_apply path + un-skip 2 invariant tests 2026-05-16 21:40:37 -07:00
introspection feat(phase3): core/cognition/explain.py — close Gap 3 introspection 2026-05-16 15:09:48 -07:00
koine_greek_fluency feat(phase5): land 5.2–5.7 — six new fluency lanes, parallel sweep 2026-05-16 20:59:31 -07:00
long_context_cost feat(adr-0044, adr-0045): domain ethics pack + long-context comparison 2026-05-17 22:31:47 -07:00
monotonic_learning feat(evals): v3 lanes — monotonic-learning passes, adversarial-identity reveals gap 2026-05-16 13:42:47 -07:00
multi_agent_composition fix(gaps): close G1+G2+G3 + identity vocab + pipeline safety-stub honour 2026-05-16 21:21:06 -07:00
multi_step_reasoning feat(algebra): null-preserving versor_apply path + un-skip 2 invariant tests 2026-05-16 21:40:37 -07:00
provenance feat(evals): parallel runner + adversarial-identity v2 2026-05-16 13:10:26 -07:00
refusal_calibration feat(adr-0043): Phase-2 pack measurements — claims → numbers 2026-05-17 22:19:24 -07:00
reports feat(adr-0023): Forward Semantic Control proof evidence — Accepted 2026-05-17 12:55:19 -07:00
results feat(adr-0043): Phase-2 pack measurements — claims → numbers 2026-05-17 22:19:24 -07:00
sample_efficiency feat(phase4): sample-efficiency v1 — first quantitative-curve lane 2026-05-16 15:39:28 -07:00
symbolic_logic feat(evals): v2 lanes for calibration and symbolic-logic 2026-05-16 13:17:41 -07:00
teaching_injection_resistance feat(epistemic): truth-seeking schema audit — 3 leaks closed, 4 new lanes, 3 new invariants 2026-05-17 07:27:41 -07:00
zero_code_domain_acquisition feat began creation of zero code domain acquisition. did not complete yet. 2026-05-16 06:31:01 -07:00
__init__.py
baseline_runner.py feat(evals): frontier structural-zero baselines for Phase 2 v1 lanes 2026-05-16 12:45:28 -07:00
CLAIMS.md feat(bench): bench cost — $/1000 turns + latency, with disclosed assumptions 2026-05-17 10:53:08 -07:00
cognition_cases.jsonl feat: vault recall index, Rust versor parity, cognitive pack expansion 2026-05-15 15:34:39 -07:00
framework.py feat(evals): Phase 0 — benchmark methodology lock-in and eval framework 2026-05-15 22:36:53 -07:00
holdout_runner.py feat(evals): Phase 0 — benchmark methodology lock-in and eval framework 2026-05-15 22:36:53 -07:00
metrics.py
parallel.py feat(evals): parallel runner + adversarial-identity v2 2026-05-16 13:10:26 -07:00
run_cognition_eval.py