How to work through it

01

Open with the tested configuration

Name the production question, interface, visible model or template, date, region, account context, prompt version, input hashes, controls, and attempt budget. Explain why the chosen brief represents a real editorial risk. If a field was unavailable, write unavailable rather than inferring a default from marketing material or a different interface.

02

Show the complete attempt ledger

List every attempt in order with its preserved output, criterion-level observations, failure code, reviewer, and final disposition. Separate technical errors, instruction drift, visual variation, rights concerns, factual problems, and editing burden. A contact sheet may support review only when it uses actual authorized outputs; this page supplies the method and does not fabricate example media.

03

Conclude at the scale of the evidence

State what was observed for this prompt and configuration, where reviewers disagreed, which failures recurred, and which follow-up test would reduce uncertainty. Disclose the attempt count and selection process beside any summary. Do not convert a small successful run into a permanent capability statement, model ranking, customer outcome, or promised rate.

  • Observed fact
  • Reviewer interpretation
  • Unresolved question
  • Retest trigger