# Runs, results and attempts

A **test run** executes selected definitions. Each **test result** pairs a definition snapshot with its retained attempts and comments. An **attempt** is one execution, with its own actions, outcome, timestamps and evidence.

## Preserve history

Retrying appends another attempt instead of replacing the earlier failure. A report can still select that earlier attempt after a later pass. The policy chosen by the project determines the summarized result; default policy helpers require an explicit choice of `latest`, `any-failure`, or `passed-on-retry`.

Do not infer a clean history from the latest green result. Read attempts and the selected aggregation policy together.

## Comments are independent

A result comment has a type and importance independent of outcome. Default types are problem, optimization, suggestion and context; default importance is low, medium, high and critical. Projects can configure both scales. An explicit setting allows new comment types, while importance stays within the configured scale.

A comment can describe a problem discovered during a passing test or context for a result that has not run. Its basis distinguishes observations, hypotheses and context. Promotion preserves that distinction; an incidental problem does not inherit the intended test's expectation or verification.

## Proof and completion

Missing required evidence makes an attempt incomplete, even when it observed a real failure. That failure can still support a bug report. Optional missing evidence produces a warning. Hook diagnostics describe setup/cleanup; they do not replace behavioral proof.

The runner persists checkpoints so finished and partial observations remain available if execution fails. Raw checkpoint/attachment output is private execution state until reviewed export.

Use [review and promotion](../guides/review-results-and-create-bug-reports.md), [selection/retries](../guides/select-and-run-tests.md), and [format reference](../reference/report-and-bundle-formats.md) to work with this history. [Illustrative results](../../examples/test-results.md) demonstrate the distinction without claiming a live run.