Evaluator

Portal status: reference only. A separate internal Evaluator V7 application runs locally with the pinned Core 1.0.0-rc.1 tarball. It compares deterministic protected and unprotected application scenarios and displays frozen durability evidence replays; it does not run real business effects. This Portal and the current private-preview bundle supply no Evaluator runner, package, or hosted service.

A future general Evaluator model could compare an approved bounded Trace snapshot with declared expectations and report PASS, FAIL, or INDETERMINATE. That model is documentation, not the input contract of the existing local V7 comparison lab. V7's scenario findings do not change Core effect truth or certify a live integration.

What it adds

An application or test harness supplies a bounded projection and expectations such as “the request stopped before dispatch when the warehouse label was ambiguous” or “the verifier used the exact label identity.” PASS, FAIL, and INDETERMINATE describe the expectation comparison, not the business result.

A PASS means the projection satisfies the expectation. A FAIL means it does not. INDETERMINATE means the projection cannot decide. None is a production-readiness claim, effect truth, or permission to make a customer claim.

Example

A warehouse workflow expects one dispatch, one independent read-back, and zero redispatches when printing a shipping label. An Evaluator can fail the expectation if a test Trace shows two dispatches, even if the authoritative system eventually contains the label. Conversely, an UNKNOWN effect can be the honest lifecycle result while an expectation passes. Evaluation belongs to testing; truth belongs to verification.

The host defines expectations and supplies the approved Trace projection. The Evaluator does not rerun operations, contact a provider, inspect private provider state, or repair a record. Missing, stale, malformed, or privacy-truncated Trace should produce INDETERMINATE.