core/docs
Shay 38872f825a feat: ADR-0119.7 — seal GSM8K test as gsm8k_math holdout (Phase 5 substrate complete)
The 1,319 GSM8K test cases are now sealed at
evals/gsm8k_math/holdouts/v1/cases.jsonl.age, age-encrypted to the
ADR-0119.1 recipient. Plaintext never touched disk in the working
tree; only ciphertext is committed.

First honest CORE-vs-real-GSM8K measurement
  cases_total: 1319
  correct:     0
  wrong:       0   ← ADR-0114a Obligation #4 holds against external corpus
  refused:     1319
  overall_pass: True

Zero confabulation. Parser refuses what it can't grammar-handle; the
"wrong == 0" discipline survives the move from CORE-original cases
to a real public benchmark. The 0/1319 correct rate is the truthful
gap that ADR-0120's threshold work will quantify.

What landed

scripts/seal_gsm8k_test.py
  - Loads GSM8K via datasets.load_dataset("openai/gsm8k", "main")
  - Strips worked-solution prose; extracts final-answer integer/float
    after "####" (handles "2,125" → 2125 thousands-separator)
  - Reads recipient from docs/holdout_recipients.txt (single repo key
    per ADR-0119.1)
  - Encrypts via pyrage; writes only ciphertext
  - Refuses to overwrite test path with train-derived seal

evals/gsm8k_math/runner.py
  - Empty expected_unit (sentinel) skips unit-comparison; grades on
    answer value alone. Required because GSM8K answers carry no unit
    structurally. wrong-zero discipline preserved.

tests/test_adr_0119_7_sealed_gsm8k.py — 6 invariants:
  1. sealed file present + age-formatted
  2. no plaintext companion files (sibling-leak guard)
  3. decrypted JSONL matches documented schema
  4. runner against decrypted suite produces wrong==0
  5. tests skip (not fail) when CORE_HOLDOUT_KEY unset
  6. case ids match "gsm8k-test-NNNN" pattern

Defensive gitignore: plaintext patterns under
evals/gsm8k_math/holdouts/v1/ are explicitly excluded.

ADR-0114a obligation roll-up
  10/10 discharged for the gsm8k_math lane:
    #1 ✓ sealed-holdout (fab_control + GSM8K test)
    #2..#10 ✓ as before

Phase 5 status: 5.1..5.7 done; 5.8 in flight (PR #149). After 5.8
merges, ADR-0120 (first expert promotion contract) becomes
feasible.

Test plan
  - pytest tests/test_adr_0119_7_sealed_gsm8k.py with CORE_HOLDOUT_KEY → 6/6
  - pytest without CORE_HOLDOUT_KEY → 3 pass + 3 skip
  - core test --suite smoke -q → 67/67
  - CLAIMS.md regenerated (no diff)
  - HF token NEVER in repo (saved at ~/.cache/huggingface/token, mode 600)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-22 20:08:35 -07:00
..
adr feat(runtime): opt-in unified-ingest path (ADR-0090, audit Findings 6+7) (#95) 2026-05-20 21:00:27 -07:00
architecture feat(physics): add mind-physics layer — ADR-0008/0009/0010, blueprint, and operator stubs 2026-05-12 23:20:58 -07:00
curriculum docs(adr-0087): rhetorical style as selection axis + writing-chain harvester spec (#72) 2026-05-20 16:09:16 -07:00
decisions feat: ADR-0119.7 — seal GSM8K test as gsm8k_math holdout (Phase 5 substrate complete) 2026-05-22 20:08:35 -07:00
evals docs(evals): articulation benchmark preamble — discourse-planner spine 2026-05-19 12:47:38 -07:00
handoff docs(handoff): two Codex briefs — domain-pack closure cleanup + grc/he content phase II (#98) 2026-05-20 23:12:15 -07:00
sessions docs: runtime_contracts + eval_methodology + papers + session log (audit #3) (#122) 2026-05-22 13:28:03 -07:00
capability_roadmap.md feat: ADR-0113 rename expert-demoaudit-passed; reserve expert namespace (ADR-0114 GSM8K roadmap) 2026-05-22 15:36:10 -07:00
ethics_packs.md feat(adr-0034): EthicsCheck — structural surface parallel to SafetyCheck 2026-05-17 20:46:34 -07:00
EVAL_AUDIT_2026-05-20.md chore(evals): contracts + bench json + Lane B viewer + chart + audit + demo schema (#62) 2026-05-20 13:53:13 -07:00
eval_methodology.md feat: ADR-0113 rename expert-demoaudit-passed; reserve expert namespace (ADR-0114 GSM8K roadmap) 2026-05-22 15:36:10 -07:00
formation_pipeline_plan.md docs: teaching-order doctrine + refresh formation/roadmap status 2026-05-17 18:48:44 -07:00
frontier_baselines.md feat(evals): frontier structural-zero baselines for Phase 2 v1 lanes 2026-05-16 12:45:28 -07:00
gaps.md Add chain-first capability ledger and domain seeds (#97) 2026-05-20 21:33:24 -07:00
holdout_recipients.txt feat: ADR-0119.1 — seal fabrication_control holdout with age encryption (Obligation #1) 2026-05-22 17:22:46 -07:00
identity_packs.md feat(surface): ADR-0031 — score-decomposition surface (per-axis hedges) 2026-05-17 20:16:22 -07:00
pack_inventory_2026-05-21.md docs: pack inventory 2026-05-21 — strengths, weaknesses, leverage gaps (#84) 2026-05-20 19:59:18 -07:00
PROGRESS.md feat: ADR-0124 — systems_software audit-passed promotion (third successful) 2026-05-22 16:55:41 -07:00
reviewers.yaml feat: ADR-0124 — systems_software audit-passed promotion (third successful) 2026-05-22 16:55:41 -07:00
runtime_contracts.md feat: ADR-0113 rename expert-demoaudit-passed; reserve expert namespace (ADR-0114 GSM8K roadmap) 2026-05-22 15:36:10 -07:00
RUST.md Activate and verify Rust backend 2026-05-13 22:23:48 -07:00
safety_packs.md feat(safety): ADR-0032 — SafetyCheck structural surface 2026-05-17 20:25:22 -07:00
teaching_order.md feat(identity-packs): ADR-0027 — swappable identity manifold via packs 2026-05-17 19:24:39 -07:00
truth_seeking_schema.md feat(epistemic): contradiction coherence checker — CONTESTED transitions wired, last Tier 4.5 row closes 2026-05-17 10:36:48 -07:00
Whitepaper.md feat: ADR-0113 rename expert-demoaudit-passed; reserve expert namespace (ADR-0114 GSM8K roadmap) 2026-05-22 15:36:10 -07:00
Yellowpaper.md feat: ADR-0113 rename expert-demoaudit-passed; reserve expert namespace (ADR-0114 GSM8K roadmap) 2026-05-22 15:36:10 -07:00