# Write and review complete independent expected cases

Supply intended outcomes from the source, freeze the complete set and keep case approval separate from release approval.

Source: https://help.smartfact.ai/manuals/automation/independent-cases

Review edition: 25 October 2026

Roles: author, reviewer

Expected cases describe the complete result the team intends. Calculate or establish the answer from the process source before comparing the candidate. The rule under test never supplies its own independent expectation.

## Save complete intended outcomes

In Check expected outcomes independently, use the product-specific controls. Supply every applicable output, reason, selected rule, component line, destination or other result the case schema requires. A single pass flag can miss an incorrect detail in the complete native result.

Include ordinary successful work and meaningful refusals. Cover missing or contradictory information, exact identities, relevant thresholds and states that should block use. Keep the independent engineering or process basis with the example so the reviewer can reproduce the expected answer.

## Respect the product's case lifecycle

Save the complete draft before freezing examples or preparing its release. Freeze the complete set rather than treating one edited row as all expected evidence. Where an expert-approved FactorySpec model carries frozen cases, changing them requires a new family draft and expert approval.

Use the required independent case review for products that separate it from release review. The reviewer must satisfy the product's contributor rules. A person changing expected results cannot become independent merely by switching roles.

## Investigate proof differences honestly

Run permitted verification against the exact candidate and frozen cases. Compare complete expected and actual results. A mismatch can expose a candidate error, an incorrect expectation or an unresolved source interpretation. Resolve that question from the agreed source instead of copying actual output into the expected field.

For source-validity cases, include ages below, at and above the limit, withdrawal, supersession, incompleteness, missing observations, expiry and tighter native restrictions. Other products need their own meaningful boundaries.

Follow [Understand release evidence](/manuals/automation/proof-and-release). Passing finite cases demonstrates the supplied checks. It cannot prove an unstated intention or certify a physical object that has not been independently observed.