No. LLM evaluation scores a model output against a rubric, a dataset, or a judge model. Outcome verification reads the real system after a side effect and decides whether the outcome happened. LLM judgement is never completion evidence.

AspectLLM evaluationOutcome verification
QuestionIs the output good?Did the action reach the promised state?
EvidenceText, a rubric, a judge modelProvider state, correlated to the operation
TimeBefore or after a run, offlineAfter each action, in production
ResultA scoreA verdict and a signed receipt
Trust in the modelThe judge is a modelA model report is E0 and never sufficient
Failure it catchesA bad answerA false completion claim
the promptthe agenttool callStripe200 pendingthe agent reads the status code and stops“Done. The customer has beenrefunded $142.”an agent assertion is evidence level E0E0 is never sufficient on its ownNobody compares the claim with the record, so nobody finds the difference.
The gap that outcome verification closes, and no other tool reads.
In words
  • A person asks the agent to perform the action.
  • The agent calls the provider and reads a success response.
  • The agent reports the task as done.
  • The provider state is still not the promised outcome.

When is LLM evaluation the right tool?

Use LLM evaluation for answer quality, tone, and regression testing of prompts. Use outcome verification when an agent makes a side-effecting call and must not say "done" until the provider proves it.

Questions developers ask

Can an LLM judge decide that a refund succeeded?

No. A judge reads text. The refund state lives in Stripe. Provely reads Stripe with the credential that you give the verifier, and compares the state with the contract.

Does the Integration Compiler use an LLM?

Yes, for semantic analysis and contract synthesis, bounded by provenance and conformance gates. Every semantic claim cites a source. The runtime uses no LLM.