8000 Pilot A: protected refutation and derived disproved status · Issue #41 · Mnehmos/MathOS · GitHub
[go: up one dir, main page]

Skip to content

Pilot A: protected refutation and derived disproved status #41

Closed

Description

@Mnehmos

Goal

Complete the first real Pilot A path through MathOS's existing protected interfaces for the false source claim “Every prime number is odd.” The protected candidate must retain an exact witness 2, publish only a kernel-checked refutation of the claim's logical negation formalization, and derive disproved only after a separate v2 fidelity review.

This issue advances Pilot A steps 1–6 and the Phase 3 authoritative-refutation criterion. It does not claim completion of Pilot A or MathOS 1.0.

Required behavior

  • Ingest the exact source statement through mcl source create.
  • Normalize it as a universal claim over natural numbers through mcl claim create.
  • Retain a no-import Lean 4.32.0 module with:
    • an explicit checked witness theorem that 2 satisfies the committed prime predicate and is not odd;
    • a theorem whose exact type is the logical negation of the normalized universal claim;
    • no axioms or forbidden source tokens.
  • Create the formalization with immutable claim_polarity: negation and the exact theorem type/declaration identity.
  • Run the real local elaboration and audit services, then prepare publication_request/1 with outcome: refutation.
  • Rebuild, dependency-check, and axiom-audit the exact module in the protected publication workflow.
  • Preserve the existing attestation, staging, controlled ingestion, receipt, and atomic authority boundary unchanged.
  • Promote exactly one accepted authoritative lean_kernel_refutation for the exact formalization version; no caller-authored or workflow Boolean may grant authority.
  • Show open/no_current_verified_fidelity before review.
  • Have the distinct protected-fidelity-reviewer submit a closed v2 review of logical_negation against the exact source/claim/formalization chain.
  • Derive disproved with one exact refutation witness after review; never persist or accept a truth-status mutation.
  • Retain the before/after status, fidelity, authority, request/report/closure, Lean outputs, and attestation evidence in the workflow artifact.
  • Keep the standalone non-authoritative True isolation smoke intact.

Tests and adversarial checks

  • The committed Pilot A module compiles with the pinned local Lean toolchain and #print axioms reports no axioms.
  • Existing publication outcome/polarity mismatch checks continue to reject a proof request for a negation formalization.
  • Existing altered/missing closure, CAS, attestation, currentness, role-separation, and fidelity-relation checks remain green.
  • bash -n passes for publication scripts.
  • Formatting, warnings-denied Clippy, all Rust tests, legacy Python regressions, and git diff --check pass.
  • PR exact-head Linux, Windows, Lean, storage, and legacy jobs pass.
  • Protected post-merge publication passes and its downloaded artifact is independently audited for exact hashes, closure membership, refutation authority, fidelity relation, and derived status.

Documentation

  • Update the implementation status and release checklist with only evidence actually produced.
  • Update publication-boundary documentation to distinguish the retained standalone True smoke, historical proof authority, and the current Pilot A refutation candidate.
  • Record exact PR-head and post-merge run, commit, tree, artifact, and semantic identities in GitHub and the implementation ledger.

Non-goals

  • Counterexample-package schema and API.
  • Atomic repaired-claim creation and research.repairs edge.
  • Proof of the repaired claim.
  • Pedagogy generation, release export, RL export, the remaining pilots, or the 1.0 acceptance command.

Activity

  1. Mnehmos commented on Jul 21, 2026

    @Mnehmos
    OwnerAuthor

    Implementation candidate published as PR #42.

    Exact head 1a5755fa815ca187c594e1d4db0c488d0a5e327c, tree 8945fd4746c1f9d9d536dd14d17854da3c9d0bdd.

    Local evidence is green: pinned Lean compiles the exact no-import module; a real Windows mcl CLI playtest completed source → universal claim → negation formalization → elaboration → axiom-free audit → refutation request; selecting proof for the same formalization failed MCL_PUBLICATION_OUTCOME_MISMATCH. Formatting, warnings-denied Clippy, 148 Rust tests, real Lean integration, 39 legacy regressions, Bash syntax, Actionlint, and whitespace checks pass.

    Protected authority and derived disproved remain pending exact-head CI, merge, protected-main publication, and independent artifact audit.

  2. Mnehmos commented on Jul 21, 2026

    @Mnehmos
    OwnerAuthor

    Protected merged-tree audit — passed

    Canonical merge:

    • commit 18abdd23025cc245f0b5722a8676e5b3f1bb2237
    • tree 8945fd4746c1f9d9d536dd14d17854da3c9d0bdd
    • main CI 29860589159 — all five jobs passed
    • protected publication 29860588969, job 88735987761 — passed
    • artifact 8507149183, publication-boundary-18abdd23025cc245f0b5722a8676e5b3f1bb2237
    • archive 89,176 bytes, GitHub and independent SHA-256 7172aa87bccedf6a694299d076e6279061cd541cf5d4b8625d01bb7e0ba3b87f

    Independent replay matched all 78 ZIP members, parsed all 54 JSON files and 25 JSONL closure entries, and rehashed every fixed role. All eight stderr files are empty. The retained module exactly matches the committed LF-normalized Git blob, has SHA-256 1e011925f7444a635c263725bf5f7dfc6fabdcd17c6f7e97135b22a534cc5af3, declares no imports, retains two_is_prime_and_not_odd, and both local/protected axiom outputs report no axioms.

    Exact lineage:

    • source 019f861a-16f5-7891-a15e-d86bb82cb6f3 / 89410bfeab715c5045558e7e628de8c860a3b0bb262151d9ddc5bcd4e232b61c
    • claim 019f861a-1706-7981-9f4f-0ed1231ef72c / 739c66af9a18a882b5c988bb3f4eb658c87c24a4a0e2f559572b39448a67d1a7
    • negation formalization 019f861a-171a-7b22-abe8-45cd917bc797 / 1f9621d15a83cbf3c201cd6d3cc739f4624cee59fdc103cf603b94f86ebac0b9
    • request 513ed6ddb73294d03e8225298ef1f94b123434b52fa65da3540ead4a8ca88e7d
    • report 0ab1aa3c7e46814b06d235987caa0b0fae7cf146ff717697ceafcb8ff625a106
    • closure 85a72971c8715c7998492143a0e63db18fae06da7283d1d8be313c66d036f340
    • bundle 2f82f4c28c76736ce65597608f33b9ddf719a6bc4dc213e80b92c8008b5aca9f
    • stage 1ec7bbd2054173bba8cb310e66fbfc7dcff47171416ffd80ce0b02034197461d
    • raw verification 089af2f16be942bbb58995df745165ce4bba2f4a6273fd7955446974484045df
    • receipt 46d88e197a040be6ce851fb28f980aee1f4dfda7714e304e5dd6e807405aaa52

    Authority evidence 019f861a-4b93-7a11-98e7-9e39aaa18192 / 52484265b0cae459cca257b95cbec4f00033987929ff7394b3db3ffa547d9225 is exactly lean_kernel_refutation / accepted / authoritative, bound to the formalization and full receipt closure. Before fidelity, status is open/no_current_verified_fidelity. Distinct reviewer protected-fidelity-reviewer then records v2 logical_negation request 1bedad0a3f33b69b913c4979c52f9ca202781646c3f11f2f54a448872de61af5, report b0a7b0f297ac397b47422cb14680ce7dcb0d42dcd7090d76259967b4b4c49872, and evidence 019f861a-4c13-7932-9257-83c145aa28eb / 8d809395a76a5c23b126053d6b70e40a398d346a876b821ee25d9cbcf2119bfc. The second read is disproved with one exact refutation witness.

    The attestation binds the exact protected workflow, commit, ref, run, GitHub-hosted runner, repository ID 1305399818, owner ID 193347153, report digest, and Rekor timestamp 2026-07-21T19:14:47Z. No retained Boolean says authoritative:true; exactly one authority_class is authoritative.

    Issue #41 is complete. Pilot A remains incomplete: counterexample packaging, immutable repair, repaired proof, pedagogy, release, and RL export are still open.

  3. Mnehmos commented on Jul 21, 2026

    @Mnehmos
    OwnerAuthor

    Evidence-ledger follow-up: PR #43 merged as 364bad572cfd362e41492a6eded2250ee4ecbefb. Post-merge CI run 29861826124 and protected publication run 29861826102 passed; artifact 8507636126 independently reproduces the refutation-authority, logical-negation fidelity, and open -> disproved derivation on the ledger merge. The exact proof/refutation evidence checklist item is now recorded as satisfied. Pilot A itself remains open at counterexample packaging, immutable repair, repaired proof, pedagogy, and export.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0