Inspect the outputs and their limits

Reviewed Cases

Open a reviewed task to compare captured outputs, check the conditions, and see where the conclusion stops.

What you do
Compare captured AI outputs against the task and evidence.
Typical time
5–10 minutes to inspect one case.
What you take away
A bounded decision you can verify and revisit.
Open a reviewed example

Published now

Evidence you can inspect and revisit.

Output ComparisonReviewed · v1

Can you audit the budget math?

The remaining amount after fixed costs is $771. Contingency is $115.65, leaving exactly $655.35 for ads. Strong answers show the calculation without rounding early.

  • 3 immutable output snapshots
  • 2 evidence-register entries
  • Limitations and review-due date visible
Inspect the Evidence
Decision RecordReviewed · v1

Would you expand this pilot yet?

Strong answers distinguish observed counts from inference, avoid generalizing five interviews to all users, and recommend a bounded next step with retention and usage evidence.

  • 3 immutable output snapshots
  • 2 evidence-register entries
  • Limitations and review-due date visible
Inspect the Evidence
Experiment ReportReviewed · v1

Does this conversion result prove an improvement?

Strong answers say the observation does not prove causality, identify small samples and time/traffic confounding, and recommend a randomized concurrent test before a durable rollout.

  • 3 immutable output snapshots
  • 2 evidence-register entries
  • Limitations and review-due date visible
Inspect the Evidence
How cases are reviewed

Only cases that completed the Reviewed gate appear here.

  1. Draft
  2. Evidence checked
  3. Reviewed
  4. Published
  5. Review due

Published evidence formats

Only formats with evidence appear here.

Additional formats will appear only after a real build or investigation satisfies the same publication contract.

Output Comparison1 published

Real outputs from the same bounded task, compared against task-specific criteria.

Experiment Report1 published

A variable, fixed conditions, captured runs, results, and a bounded conclusion.

Decision Record1 published

Why one method, model, tool, or stack fit a stated set of constraints.