Skip to content
Checklists/8 checks

Loop Validation Checklist

Use this after a loop runs. The goal is not to write a long report; it is to capture enough evidence that the next decision is grounded.

0 / 8
Worked review · maintained by TianMingAI · 2026-07-21

A completed example

A bug-fix loop returns a polished summary and green CI. Inspection shows that the original regression test was marked as skipped, so the visible success signal no longer measures the behavior the task was meant to repair.

Pass

Expected evidence was named before the run

The plan required the original regression test to fail before the change and pass afterward.

Stop

Actual evidence matches the expected evidence

CI is green only because the regression test is skipped; there is no passing result for the original behavior.

Stop

The result avoided forbidden shortcuts

Skipping or weakening tests was an explicit boundary and the diff violates it.

Partial

The next decision follows the evidence

The implementation change may still be useful, but it cannot be accepted until the real regression is restored and exercised.

Decision

Classify the run as failed despite green CI. A polished summary and passing aggregate status cannot substitute for the specific evidence defined before execution.

Next action

Restore the regression test, reproduce the original failure on the clean baseline, and rerun the smallest repair with a checker that compares test files and test counts.

Why this checklist matters

Validation turns a run into evidence. Without a written expectation and captured result, teams tend to remember the impressive output and overlook partial failures, hidden shortcuts, or a lack of measurable progress.

How to use this checklist

Write the hypothesis and expected evidence before the run begins.
Capture the actual command output, artifact, or review result without rewriting it to look cleaner.
Compare expectation with reality and choose one explicit next action.

Decision after the check

Stop when the evidence fully supports the outcome. Adjust one assumption after a partial result, and escalate rather than rerun when the same blocker or shortcut appears again.

Review questions

What changed because of this run?
Which evidence would make you reverse the current conclusion?
Is another iteration likely to teach something new?

Related templates