User report · Coding tasks
Repair formal proofs while preserving reviewed specifications
Source-backed example. Practical decision guidance is proposed and untested; this is not a full implementation guide.
The problem
Changes to verified software can break proofs even when the intended specification should remain unchanged.
Agent job
Inspect proof states, edit proof code, run builds and audit assumptions after repair.
Expected output
A reviewable proof repair with checker results and premise review.
Where this framing comes from
- Core problem: Editorial purpose framing
- Agent job: Editorial purpose framing
- Expected output: Editorial purpose framing
- Relevant job roles: Editorial purpose framing
- Benefit type: Editorial purpose framing
Editorial framing and outcome limits
Describes the task’s purpose and intended output; measured benefit, safe unsupervised operation and successful delivery are not established by this framing.
Editorial synthesis based on the linked source. This framing does not add evidence of successful delivery or results.
Case-specific decision guidance · untested
Case-specific decision guidance and first checks are original proposals. The linked source supports the attributed task; this added guidance has not been executed.
Read the attributed source evidence and original caveats below ↓
Why use this
A proof repair can make a checker accept changed code while altering a definition, strengthening a precondition or weakening the theorem the team intended to preserve. The visible completion of proof obligations does not by itself show that the reviewed specification survived.
A premise-and-change review packet could distinguish proof-script adjustments from changes to the mathematical contract. Formal-methods reviewers could inspect every proposed assumption and boundary change before deciding whether an authorized checker run and merge are appropriate.
Why an agent / simpler alternative
- Useful judgment
Trace proposed proof changes to affected definitions and assumptions while flagging shifts in the statement being established.
- Simpler alternative
Repair a localized tactic script manually when the specification and dependency boundary are clear.
- Unsuitable when
No qualified reviewer can assess changed invariants, premises or theorem statements.
Proposed inputs
- Name
Proof-change description
- Owner
Formal-methods maintainer
- Minimum scope
Fictional theorem statements, invariant summaries and proposed edit categories
- Quality check
Separate proof steps from specification changes.
- Name
Reviewed contract
- Owner
Specification owner
- Minimum scope
Preserved definitions, allowed assumptions and discussion requirements
- Quality check
Identify premises requiring concrete justification.
Access and setup — proposed or unknown
- Proposed minimum
Offline symbolic descriptions and a premise-difference manifest without a proof assistant.
- Compatibility unknowns
Unknown: current runtime, proof-library versions, checker configuration and assumption-audit tooling.
- External actions requiring approval
Definition, invariant, theorem or precondition changes; separately authorize builds, repository writes and merge.
Expected deliverable and review checks
- Proposed artifact
Proof-change classification with premise differences, specification impacts and reviewer questions.
- Critical checks
Every added assumption or altered statement is highlighted explicitly.
A proposed checker pass is not treated as evidence that the original contract is preserved.
- Missing output evidence
No new workflow run or output inspection was performed for this overlay; use the canonical source-outcome references.
Failure signals and proposed responses
- Signal
A repair adds a stronger precondition without justification.
- Why it matters
The theorem may apply to fewer cases than intended.
- Proposed response
Hold the change for specification-owner review and require a concrete rationale.
- Signal
A theorem statement changes inside a supposedly proof-only patch.
- Why it matters
The scope of the established claim may shift unnoticed.
- Proposed response
Reclassify the patch and require explicit formal-methods review of the new boundary.
Effort and cost — unknown
- Setup work
Record reviewed statements and premise-audit requirements.
- Ongoing review
Inspect semantic changes and later authorized checker evidence.
- Unknowns
Unknown: build cost, library compatibility and expert review effort.
First check — proposed, unexecuted
- Status
proposed_unexecuted
- Synthetic fixture
A fictional proof-change manifest labels an added precondition and altered theorem conclusion as tactic-only edits.
- Acceptance checks
Reclassify both as contract changes.
Require reviewer justification without claiming a proof has been checked.
- Stop conditions
Proof execution, real theorem editing or correctness certification is required.
Limits
This proposed premise checklist proves nothing; it cannot establish logical soundness, checker acceptance or specification preservation.
Editorial candidate. This is source-backed reporting or a documented workflow. BotCurrent has not independently tested this task. A proposed Canadian adaptation is not evidence of a Canadian deployment or compliance certification.
Original source
https://arxiv.org/abs/2606.21167 ↗Attributed to: Shengyi Wang
User report- Original action type
- coding_action
- What the source supports
- The CertiGC experience report describes Codex editing Rocq proofs, running builds and inspecting proof states to repair invariants after mutable collector changes, followed by a premise audit and cleanup.
- Source outcome limits
- Actions and outcomes are attributed user reports, not an independent reproduction or an execution guarantee.
- Evidence attribution caveat
- Actions and outcomes are attributed user reports, not an independent reproduction or an execution guarantee.
- Source approval description
- Definition changes required discussion; additional specification preconditions needed concrete justification. A human judged the invariant and theorem boundaries.
- Named tools
- Codex; Rocq; VST; CertiGraph; rocq-mcp; make
- Model or version
- Unknown
- Publication date
- 2026-06-19
- Source last checked
- 2026-10-14
- Demonstrated in Canada
- Unknown
- Canadian availability
- Unknown
- Proposed Canadian adaptation
- Proposed: apply checker-driven repair to Canadian research code with documented theorem goals and expert review.
- Recommended approval boundaries
- Require formal-methods review for every specification, invariant and theorem change before merging.