docs: reserve VERIFIED assurance in control evaluations - #906
Conversation
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7c001ad9b0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
7c001ad to
135713d
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 135713d99b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
2 participants
What
Clarify the assurance vocabulary used in paired and ad hoc agent evaluations:
VERIFIEDfor results bound to the exact task input by an operator-authorized checker and verification record;Why
A recent no-Jacobian control pilot used uppercase
VERIFIEDin 4 of 6 answers for manual or self-run checks. The mathematics could be correct while the assurance claim crossed Jacobian's trust boundary. The maintained structured Harbor contract already enforces assurance ceilings; this closes the corresponding guidance gap for unstructured/ad hoc pilots.Impact
Documentation only. No mathematical capability, checker, benchmark task, routing policy, or existing draft PR is changed.
Closes #903.
Checks
make docs-command-checkmake docs-linkcheckusing the pinnedmarkdown-link-check@3.15.0(the environment had Node but nonpxexecutable)make test-plan BASE=origin/main→ docs lane only