Your benchmark is only as trustworthy as its verifier.

Fourteen stages around one evidence boundary.

The workflow is one source lock, twelve task-level validation stages and one submission-level evidence seal. The regression tests qualify the Tbench engine; they are not hundreds of checks claimed against your task.

All fourteen stages apply to a single-task audit: one source lock, twelve task-level stages, and one evidence seal. There is no minimum batch size.

1. 01Source lock and provenanceSource integrityRequireddeterministicOnce per audit

   Binds the repository revision, task tree digest, tool version and audit policy.
2. 02Contract and scaffold auditStatic architectureRequireddeterministicOne task

   Runs the registered rules against structure, metadata, instructions, launchers and verifier contracts.
3. 03Requirement traceabilityRequirement mappingRequireddeterministicOne task

   Maps graded requirements from the instruction through the solution to verifier assertions.
4. 04Novelty and contamination screenCorpus integrityRequireddeterministicOne task

   Compares the task with a pinned corpus rather than treating a single task as vacuously unique.
5. 05Oracle and no-op calibrationContainer calibrationRequiredcontainerOne task

   Runs positive and negative controls in isolated images.
6. 06Pass@2 precheckDifficulty calibrationRequiredmodelOne task

   Runs the fixed two-attempt gate and separates solver failure from infrastructure failure.
7. 07Verifier boundary auditVerifier rigorRequiredcontainerOne task

   Challenges false accepts and false rejects across invalid and boundary-case solutions.
8. 08Fresh-input witness verificationIndependent re-executionRequiredcontainerOne task

   Re-executes the deliverable on fresh seeded inputs.
9. 09Mutation sensitivityMutation rigorRequiredcontainerOne task

   Introduces controlled defects and measures whether the verifier kills them.
10. 10Multi-model difficulty probeDifficulty evidenceRequiredmodelOne task

    Preserves panel disagreement and empirical difficulty evidence.
11. 11Reward-hacking challengeAdversarial resistanceRequiredcontainerOne task

    Attempts leakage, protected-path access and shortcuts that earn reward without doing the task.
12. 12Pass@5 frontier gauntletEmpirical gauntletRequiredmodelOne task

    Records each valid attempt, model, environment and observed outcome.
13. 13Grounded coherence reviewAutomated reviewAdvisorymodelOne task

    Adds an advisory second opinion without overriding deterministic evidence.
14. 14Evidence seal and submission decisionAttestation and registerRequiredaggregateOnce per audit

    Binds stage records to exact digests and issues or withholds the submission result.