Skip to content

benchmarks: add exact Navier–Stokes polynomial certificate - #595

Merged
morluto merged 8 commits into
mainfrom
agent/conjecture-navier-stokes-polynomial
Aug 7, 2026
Merged

benchmarks: add exact Navier–Stokes polynomial certificate#595
morluto merged 8 commits into
mainfrom
agent/conjecture-navier-stokes-polynomial

Conversation

@yuelgrace1810-ops

Copy link
Copy Markdown
Collaborator

Summary

  • adds the independent Harbor task navier-stokes-polynomial-certificate
  • derives it from retained Unresolved Conjecture row C-006 and the Clay Navier–Stokes status source
  • asks for one exact non-irrotational affine steady flow with quadratic pressure
  • independently replays incompressibility, both polynomial momentum residuals, rational canonicalization, and vorticity
  • keeps the assurance ceiling at CHECKED and explicitly forbids any global 3D regularity conclusion

Benchmark design

  • Family: Assurance
  • Primary objective: exact symbolic PDE certification
  • Provisional difficulty: Hard; the agent must coordinate two polynomial residual identities, divergence, nonzero vorticity, canonical rational coefficients, and a strict scope boundary
  • Quality score: 87/100
  • Shortcut audit: rejects the zero field, irrotational fields, copied residual labels, noncanonical fractions, false VERIFIED claims, and input/evidence tampering
  • Alternative answers: the verifier accepts any bounded rational affine/quadratic certificate satisfying the equations, not only the Oracle rotation
  • Assurance boundary: one exact 2D steady polynomial example; no evidence about global 3D Navier–Stokes smoothness

Portfolio value

This adds exact PDE residual replay, which is distinct from the existing bounded graph, cuboid, point-set, and algebraic certificate workflows. It is intentionally a finite assurance task rather than an open-problem solver.

Validation

  • focused adversarial tests: 4 passed
  • Harbor selected-task preparation/check: passed
  • Ruff format/lint: passed
  • git diff --check: passed
  • Docker Client/Server: 29.6.2
  • task and verifier images built
  • both executed with --network none
  • direct clean-room Oracle reward: 1.0 on every applicable component
  • native Harbor 0.20 macOS exact-task Oracle remains blocked by the known no-network capability-detection defect; GitHub Linux exact-task Oracle is required before readiness

This PR is independent and targets current main.

morluto commented Aug 7, 2026

Copy link
Copy Markdown
Owner

@codex

@morluto
morluto marked this pull request as ready for review August 7, 2026 09:05

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 426dbfaea1

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".

Grace Lee Rui Yue and others added 3 commits August 7, 2026 18:08
- Remove arbitrary evidence size cap by passing max_bytes=None
- Declare input-binding exception in tests/verifier_contract.json and add
  it to the Dockerfile COPY list
- Decouple mathematics, evidence, scope, and assurance from contract validity
- Emit complete zeroed diagnostic shape (protocol, input_binding,
  mathematics, evidence, scope, assurance, false_certification) in the
  exception handler instead of aggregate-only keys
- Derive coefficient bounds from /tests/input.json instead of hard-coding
  them
@morluto
morluto force-pushed the agent/conjecture-navier-stokes-polynomial branch from 426dbfa to c0a15fa Compare August 7, 2026 10:18

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c0a15fafac

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".

morluto and others added 4 commits August 7, 2026 18:39
…hecksum

- Bound _raw_submission() with is_regular_bounded_file
- Remove undocumented evidence size limit per benchmark policy
- Refresh Dockerfile checksum label
…act and lint

Co-authored-by: morluto <morluto@users.noreply.github.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 96bd1357cb

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".

@morluto
morluto merged commit 86bdb1b into main Aug 7, 2026
33 checks passed
@morluto
morluto deleted the agent/conjecture-navier-stokes-polynomial branch August 7, 2026 11:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants