benchmarks: audit perfect-cuboid finite scope - #592
Conversation
![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: 31cbc21c32
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".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 195d7f1bae
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".
…ormat Fix root causes flagged across conjecture-probe verifier PRs: 1. Mathematics, evidence, scope, and assurance are now computed independently of contract validity. A correct result no longer reports mathematics:0.0 when an unrelated envelope field is invalid. 2. All advertised scoreable assurances (UNVERIFIED, COMPUTED, CHECKED) receive credit, not just CHECKED. 3. False-certification detection reads the raw submission JSON before strict schema validation can discard a malformed VERIFIED claim. 4. The evidence media type is corrected from text/plain to application/json, and the required payload shape is published in the public contract so agents can discover the JSON schema. 5. Reverted unrelated whitespace churn across 26 symbolic-coordination verifier_support copies and formatter-only rewrites of two reliability verifier files that changed Harbor task digests without changing behavior.
195d7f1 to
86085bf
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 86085bfdae
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".
…cksum - Bound _raw_submission() with is_regular_bounded_file to prevent reading oversized submission files before strict validation - Refresh Dockerfile checksum label to match updated verifier code
Co-authored-by: morluto <morluto@users.noreply.github.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 28fbbceb66
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Review comment fixes1. Decouple mathematics from strict envelope validationProblem: When a submission reported the correct Fix: In Added test 2. Wire the new verifier leaf into the benchmark plannerProblem: Fix: Added a Added test Dockerfile checksumUpdated |
3 participants
Summary
C-012Benchmark contract
Shortcut and duplication audit
A class label or one public Euler-brick example is insufficient. Every submission must provide all four radicands and exact roots/null results for all twelve cases. Case order is free, but IDs must be complete and unique. The workflow is distinct from the merged bounded Vizing probe and existing Diophantine witness searches because it specifically tests semantic separation and assurance calibration.
Validation
Passed locally:
git diff --check--network noneNative Harbor 0.20 on macOS reaches task selection and then rejects Docker Desktop's no-network capability detection with the known
network_mode='no-network' is not supportederror. The policy was not weakened and the Docker socket was not mounted. This draft awaits the GitHub Linux exact-task Oracle before it is considered complete.Repository-wide static formatting currently reports unrelated pre-existing formatting drift in public-reproduction and symbolic-coordination files outside this branch; no unrelated files are changed.