Compare two equivalent runs to see what a system change actually did: which findings appeared, which resolved, and how the decision-grade scores moved. The comparison page does the arithmetic; this guide keeps the inputs honest.
Everything downstream depends on the pair being comparable — same target, same template, same scenario set. If scope changed, name the difference before interpreting anything.
1. Pick two runs
- Open /runs and choose an earlier run as the baseline and a later run as the candidate.
- Open the comparison page to inspect decision-grade score changes, evidence coverage, and findings that are new, persistent, or resolved.
- Use runs with the same target, template, and scenario set. If scope changed, describe that difference before interpreting the result.
Pick an earlier run as the baseline and a later one as the candidate, then open the comparison. Runs with different targets or scenario sets produce numbers that flatter whatever narrative you came in with — describe scope changes explicitly instead.
2. Read the verdict
- Start with the API comparability verdict; incompatible or partial evidence yields no numeric delta.
- Review new, persistent, and resolved findings individually; the labels describe differences between these runs, not every possible behavior — this two-run view is operational replay, not an A/B experiment.
- For an eligible Starter+ run, export the comparison with the evidence pack so another reviewer can trace the scope and observed change.
Start from the API comparability verdict: incompatible or partial evidence yields no numeric delta, by design. Read new, persistent, and resolved findings individually — the labels compare these two runs, not every possible outcome. For an eligible Starter+ run, the comparison can travel with the evidence pack.