[renmengye]

Closes #489, as the PI decided on that issue.

There was no explicit one-idea rule; panel judges inferred "bundling" from the unsupported-claim and measurement-gap checks. This PR states the rule directly:

• verifier.py (VERIFY_SYSTEM_PROMPT): a composite change is acceptable when the report gives each component's isolated measurement and the combination's under the same protocol, and the combination is what the kernel measured. A bundle with no isolated measurement for each component stays blocked, and is named in both the measurement-gap check and the blocking list. The judge checks the report's evidence and re-runs nothing.
• guidance_pack.py (AUTHOR), beside the one-hypothesis rule: a combination whose parts each fall below the improvement floor may ship together if the PR reports each part's isolated measurement and the combination's.
• One test for the verify brief; CHANGELOG line.

Gate: ruff check, ruff format --check, mypy, pytest (3069 passed, 7 skipped).

Built by DeepSeek-V4.1-Flash; reviewed by Claude.

[github-actions]

Round 1 — reviewed head 0eb78394 — reviewer hermes/gpt-5.6-terra.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: nothing blocking — 1 advisory note.

1 finding attached to the lines below.

The bare-bundle rule can be read as blocking only bundles with no isolated measurements at all.

[github-actions]

Round 2 (re-run on the same head) — reviewed head 0eb78394 — reviewer summarizer:hermes/gpt-5.6-terra over coverage+credentials+deployment+general+lifecycle+prose.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: 2 blocking, 0 advisory.

2 findings attached to the lines below.

2 blocking findings recorded. No findings were deduplicated or rejected: the coverage test failure and credentials trust-boundary bypass concern different files and claims. Deployment, general, lifecycle, and prose reported no findings.

[github-actions]

Round 3 — reviewed head fc148f47 — reviewer hermes/gpt-5.6-terra.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: 1 blocking, 1 advisory.

2 findings attached to the lines below.

The composite-evidence path can accept forged component measurements and can hide genuine later measurements.