---
title: "Define release evidence a coding agent can hand off"
description: "Specify revision, fixtures, run identifiers, observed assertions and unresolved checks so another agent or reviewer can assess a QA result accurately."
canonical_url: "https://vostride.com/agent-library/software-factory/release-evidence-contract.md"
md_url: "https://vostride.com/agent-library/software-factory/release-evidence-contract.md"
content_kind: "guide"
topic: "software-factory"
---

# Define release evidence a coding agent can hand off

A useful QA handoff connects a concrete application revision to completed observations. It should let another agent or reviewer determine what was tested, what passed, what failed and what could not be checked. A summary such as “all looks good” leaves those decisions unresolved.

## Minimum record

For each relevant test, retain the application revision, environment URL or device profile, fixture identity, test file, run identifier and terminal status. Link the failing or decisive step and the artifacts supporting its outcome. Include external hook results when the acceptance criterion concerns persistence, delivery or file contents.

```text
Change: application revision and requirement reference
Environment: target, app build and fixture revision
Test: file path and completed run identifier
Observed: each acceptance criterion and supporting artifact
External checks: hook or harness result and record identifier
Failures: original run, first mismatch and proposed cause
Rerun: new run identifier after the relevant repair
Unverified: missing prerequisite or unobserved requirement
```

This is a report format, not a command or runtime schema. Adapt it to the team's existing pull request or release process.

## Distinguish statuses

Keep a schema validation result separate from application execution. A dry run describes planned work. A queued or running job has no final verdict. A test that failed during authentication may have no evidence about the feature it intended to exercise.

Preserve the original failure during repair. If the rerun uses a different fixture, model, browser or application build, record that difference because it can affect the interpretation. Report failed and blocked checks alongside passing ones.

## Use artifacts at their proper scope

A screenshot can establish a visible value at a particular step. It cannot prove an invoice was settled or that a downloaded file contains all expected rows. Pair visual evidence with independent checks where the contract needs them. The [recorded evidence guide](/docs/agent-qa/guides/recorded-evidence.md) illustrates actual result presentation and its limits.

[Coding-agent workflow](/docs/agent-qa/guides/coding-agent-workflow.md) · [Reporting recipes](/agent-library/reporting-exports.md)