core/scripts/generate_claims.py
Claude 9cf6a164d5
feat(evals,chat,tests): PR-12 — the Wilson re-count. 25 licensed bands become 4 (R-13)
Rank 5 on the docket, and the first item this arc whose expected outcome is
LOSING capability.

conservative_floor is a one-sided Wilson lower bound, and Wilson assumes
independent trials. CORE's pipeline is deterministic, so replaying an identical
case is not a second trial — it is the same trial with a guaranteed outcome.
The deduction sealer folded the raw practice corpus and counted every replay,
so bands recorded 720/720 committed whether that was 720 decisions or 28
decisions seen 26 times each. conservative_floor(720,720)=0.990868 cleared
theta_SERVE=0.99; conservative_floor(28,28)=0.808413 is not close.

RESULT: 25 licensed bands -> 4. Twenty-one demoted.

Survivors: en_conditional_chain, en_disjunctive, en_verb_fact,
en_verb_universal — the only bands whose corpus holds enough independent cases.

What the demotion did and did not do, stated exactly. NO ANSWER CHANGED. No
answer became wrong. `wrong` stayed 0 across all 25 bands. The engine is
exactly as correct as it was. What changed is the CLAIM attached to the answer:
21 bands now serve the same sound conclusion prefixed with "(reasoned, but I
haven't yet earned a verified track record on arguments of this shape)". The
reasoning did not get worse; the boast did.

The producer is hardened, which is the half that stops this recurring.
seal_ledger now refuses outright — assert_sealed_evidence_distinct compares the
ledger about to be written against the corpus's distinct-case count and raises
BEFORE anything is written. The curriculum sealer has carried this guarantee
since ADR-0264 R9; the deduction sealer had no equivalent, which is exactly how
the exposure arose. Catching it at seal time rather than in an audit matters:
an audit finds a padded ledger after it is committed, trusted, and gating a
live flag, and unwinding that then needs a ruling — it needed one.

A HOLLOW GUARD OF MY OWN, caught by sabotage rather than trusted. The first
version compared all_gold_problems() with distinct_gold_problems() — two
functions that agree with each other by construction, and neither of which is
what build_ledger folds. Reverting build_ledger to the raw corpus walked
straight past it AND re-sealed the inflated artifact. Same defect class the
ledger already suffered: a guard checking something adjacent to the thing it
protects. It now takes the BUILT LEDGER as input, because the invariant is
about the artifact. Re-running the identical attack: refused, wrote nothing.
Pinned as test_sealer_refuses_a_ledger_that_counts_replays.

A SECOND MISS, caught by the gate rather than by me. My blast-radius scan
grepped for deduction_serve_license / deduction_serve_ledger / 720 and found
nine files. It missed tests/test_construction_inventory.py, which asserts
deduction SURFACE STRINGS and never mentions the ledger — so a pattern search
over the wrong noun could not see it. The full gate found it. Recorded because
"I scanned for the blast radius" is worth exactly as much as the scan's key,
and mine was the wrong key.

That test is the #138 fabrication defect pin, and it is updated WITHOUT being
weakened. Both surfaces now carry the disclosure prefix, and the fabrication
survives untouched — which is a sharper finding than before: a disclosure
prefix looks like it might cover the hazard, and it does not. The user is still
told they said "furthermore". Exact-match assertions kept on the full strings
so neither the fabrication nor the licence state can drift unnoticed, plus an
isolating assertion that the only difference between clean and leaked recitals
is the fabricated premise.

Four records corrected in the same PR, each of which would otherwise have gone
on asserting capability that no longer exists:
  - CAPABILITY_LEDGERS' manifest note ("25 bands at 720/720 wrong=0")
  - the published CLAIMS.md claim — caught immediately by G-22's own pin, which
    is that fix earning its keep three days after landing
  - test_volume_honesty.py's module docstring, which said the exposure was
    "NOT silently repaired ... changing it is Shay's ratification, not a
    test's". R-13 IS that ratification, so the line went stale by being acted on
  - the exposure inventory's framing: it now pins the repair, not the gap

test_volume_honesty's sealed-vs-producer assertion is strictly stronger than it
was. It compared sealed committed counts to the producer's committed counts and
both were the inflated 720 — the assertion held perfectly while the number it
agreed on was the wrong number. Two records agreeing is not evidence that
either is right. It now pins committed == distinct, plus a non-vacuity check.

Nine tests encoded the unearned expectation and were updated to the honest one.

Closes G-19 and H-1.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wcw2pnMBwyvmNyQg4uPEt4
2026-07-28 07:42:58 +00:00

284 lines
9.7 KiB
Python

"""Generate CLAIMS.md from the capability ledger and pinned lane SHAs.
CLAIMS.md is a single-page, auditable list of every CI-pinned
capability claim CORE currently makes. It is **mechanically derived**
from two ground-truth sources:
- ``core.capability.ledger_report()`` — domain ratification rows
(Tier 1: "this domain is reasoning-capable").
- ``scripts.verify_lane_shas.PINNED_SHAS`` + ``LANE_SPECS`` —
per-lane pinned report SHA-256 (Tier 2: "this lane's report bytes
hash to X under deterministic replay").
The generator is **deterministic**: same inputs → byte-identical
output. ``tests/test_claims_md_is_current.py`` regenerates the file
into a temp path and asserts byte equality with the on-disk copy.
"""
from __future__ import annotations
import argparse
import hashlib
import sys
from dataclasses import dataclass
from pathlib import Path
REPO_ROOT = Path(__file__).resolve().parent.parent
CLAIMS_PATH = REPO_ROOT / "CLAIMS.md"
# Import via path so this script remains runnable without installing the
# package.
sys.path.insert(0, str(REPO_ROOT))
from core.capability import ledger_report # noqa: E402
from scripts.verify_lane_shas import LANE_SPECS, PINNED_SHAS # noqa: E402 # type: ignore[import-not-found]
# ADR → human-readable lane purpose. Keep this in lockstep with the
# ADR docs themselves; the generator fails fast if any lane lacks a
# mapping here, which surfaces drift before CI does.
_LANE_ADR: dict[str, tuple[str, str]] = {
"reviewer_registry": (
"ADR-0092",
"Reviewer registry schema validates + bootstrap entry self-seals",
),
"miner_loop_closure": (
"ADR-0095",
"Miner-sourced proposals route through single reviewed teaching path",
),
"curriculum_loop_closure": (
"ADR-0104",
"Curriculum-sourced proposals route through single reviewed teaching path",
),
"domain_contract_validation": (
"ADR-0093",
"All ratified packs satisfy the 9 ADR-0091 contract predicates",
),
"fabrication_control_summary": (
"ADR-0096",
"Phantom endpoints / cross-pack non-bridges / sibling collapses refuse",
),
"demo_composition": (
"ADR-0098",
"Demos compose from shipped modules; no parallel mechanism",
),
"public_demo": (
"ADR-0099",
"Public showcase runs deterministically under 30s; all claims supported",
),
"math_teaching_corpus_v1": (
"ADR-0131",
"Math teaching corpus replays deterministically; all chains pass exit criterion (correct_rate=1.0, wrong=0)",
),
"deductive_logic_v1": (
"ADR-0206",
"Propositional entailment scored against an independent truth-table oracle; dev+holdout+external 716/716 correct, wrong=0, refused=0",
),
"curriculum_serve_v1": (
"ADR-0262",
"Flag-gated curriculum serving answers exam questions from a subject's RATIFIED chain corpus only; untaught facts return UNKNOWN (anti-recall probes enforced), wrong=0",
),
"deduction_serve_v1": (
"ADR-0256",
"Flag-gated deduction serving decides real English/member/fused/verb/existential arguments end-to-end, wrong=0 across all splits; 4 of 25 shape-bands hold an earned SERVE licence on DISTINCT evidence (R-13 re-count, 2026-07-28) and the other 21 decide with disclosure",
),
}
# Domain → ratifying ADR. The ledger row already carries provenance
# inside each contract (e.g. "adr-0101:reviewed:2026-05-21"), but we
# pin the canonical ADR id here so the table reads cleanly even if a
# pack later carries a different provenance prefix.
_DOMAIN_ADR: dict[str, str] = {
"systems_software": "ADR-0101",
"mathematics_logic": "ADR-0097",
"physics": "ADR-0100",
"hebrew_greek_textual_reasoning": "ADR-0102",
"philosophy_theology": "ADR-0085",
}
@dataclass(frozen=True, slots=True)
class DomainRow:
domain: str
status: str
adr: str
pack_count: int
open_gaps_count: int
@dataclass(frozen=True, slots=True)
class LaneRow:
lane_id: str
adr: str
purpose: str
report_path: str
pinned_sha: str
def _collect_domain_rows() -> list[DomainRow]:
report = ledger_report()
rows: list[DomainRow] = []
for entry in report["domains"]:
domain = entry["domain"]
if domain not in _DOMAIN_ADR:
raise RuntimeError(
f"domain {domain!r} missing from _DOMAIN_ADR mapping in "
f"scripts/generate_claims.py — add its ratifying ADR id."
)
rows.append(
DomainRow(
domain=domain,
status=entry["status"],
adr=_DOMAIN_ADR[domain],
pack_count=len(entry["packs"]),
open_gaps_count=len(entry["open_gaps"]),
)
)
# Deterministic ordering: by ADR id (lexicographic), then domain.
return sorted(rows, key=lambda r: (r.adr, r.domain))
def _collect_lane_rows() -> list[LaneRow]:
rows: list[LaneRow] = []
for spec in LANE_SPECS:
if spec.lane_id not in _LANE_ADR:
raise RuntimeError(
f"lane {spec.lane_id!r} missing from _LANE_ADR mapping in "
f"scripts/generate_claims.py — add its ADR id and purpose."
)
adr, purpose = _LANE_ADR[spec.lane_id]
pinned = PINNED_SHAS.get(spec.lane_id)
if not pinned:
raise RuntimeError(
f"lane {spec.lane_id!r} has no entry in PINNED_SHAS — "
f"run scripts/verify_lane_shas.py --update first."
)
rows.append(
LaneRow(
lane_id=spec.lane_id,
adr=adr,
purpose=purpose,
report_path=spec.report_relative,
pinned_sha=pinned,
)
)
return sorted(rows, key=lambda r: r.adr)
def _render(
*, domain_rows: list[DomainRow], lane_rows: list[LaneRow]
) -> str:
"""Render CLAIMS.md as a deterministic markdown string.
Two tables. No timestamps, no git revs, no wall-clock — anything
volatile would break the "regenerate and compare bytes" test.
"""
lines: list[str] = [
"# CLAIMS",
"",
"<!--",
"AUTO-GENERATED by scripts/generate_claims.py.",
"Do not hand-edit. Run `python3 scripts/generate_claims.py` to",
"regenerate after any change to the capability ledger or pinned",
"lane SHAs (scripts/verify_lane_shas.py).",
"-->",
"",
"Every row below is mechanically derived from in-tree state and",
"verified by CI. Tier 1 rows come from `core.capability.ledger_report`;",
"Tier 2 rows come from `scripts/verify_lane_shas.py`'s pinned SHAs.",
"",
"## Tier 1 — Ratified domains",
"",
"Each row asserts: the domain's packs pass all nine ADR-0091 contract",
"predicates, declared operator chains meet the ≥8 / ≥3-intent floor,",
"the reviewer registry resolves the primary reviewer, and the ledger",
"status predicate evaluates to `reasoning-capable` with no open gaps.",
"",
"| Domain | Status | ADR | Packs | Open gaps |",
"| --- | --- | --- | --- | --- |",
]
for row in domain_rows:
lines.append(
f"| `{row.domain}` | {row.status} | {row.adr} | {row.pack_count} | "
f"{row.open_gaps_count} |"
)
lines.extend(
[
"",
"## Tier 2 — Pinned lane reports",
"",
"Each row asserts: running the lane's runner produces a JSON",
"report whose SHA-256 matches the pinned value below. Mismatch",
"is a CI failure (`.github/workflows/lane-shas.yml`).",
"",
"| ADR | Lane | Purpose | Report path | Pinned SHA-256 |",
"| --- | --- | --- | --- | --- |",
]
)
for row in lane_rows:
lines.append(
f"| {row.adr} | `{row.lane_id}` | {row.purpose} | "
f"`{row.report_path}` | `{row.pinned_sha}` |"
)
lines.extend(
[
"",
"## Verification",
"",
"```bash",
"python3 scripts/verify_lane_shas.py # verify all Tier 2 SHAs",
"core test --suite full -q # exercises Tier 1 invariants",
"```",
"",
]
)
return "\n".join(lines)
def render_claims() -> bytes:
domain_rows = _collect_domain_rows()
lane_rows = _collect_lane_rows()
text = _render(domain_rows=domain_rows, lane_rows=lane_rows)
return text.encode("utf-8")
def main(argv: list[str] | None = None) -> int:
parser = argparse.ArgumentParser(description="generate CLAIMS.md")
parser.add_argument(
"--check",
action="store_true",
help="exit non-zero if CLAIMS.md differs from regenerated bytes",
)
parser.add_argument(
"--output",
type=Path,
default=CLAIMS_PATH,
help="write to this path (default: repo-root CLAIMS.md)",
)
args = parser.parse_args(argv)
payload = render_claims()
sha = hashlib.sha256(payload).hexdigest()
if args.check:
if not CLAIMS_PATH.exists():
print(f"FAIL: {CLAIMS_PATH} does not exist; run without --check")
return 1
on_disk = CLAIMS_PATH.read_bytes()
if on_disk != payload:
print("FAIL: CLAIMS.md is stale — regenerate with:")
print(" python3 scripts/generate_claims.py")
print(f" on-disk sha256: {hashlib.sha256(on_disk).hexdigest()}")
print(f" computed sha256: {sha}")
return 1
print(f"OK: CLAIMS.md is current (sha256: {sha})")
return 0
args.output.write_bytes(payload)
print(f"wrote {args.output} (sha256: {sha})")
return 0
if __name__ == "__main__":
sys.exit(main())