Choose One Decision
Start with a question such as: “Do we have enough evidence to run another small trial?” Do not make a meeting about choosing a universal winner. Pick the actual task, an acceptable result and one unresolved concern.
This exercise uses fictional candidates and invented inputs. It does not claim a result for any named AI service.
Narrow The Saved View
Suppose you saved five candidates, but only two are currently in Testing. In your shortlist, select that stage. Leave the date view at All if both tested and not-yet-tested candidates matter. A missing test date is an information gap, not evidence that the tool failed.
Open Export, backup and restore, choose Current view only under Brief candidates, and check the count. Download the brief. Keep private notes excluded unless you have reviewed them for the intended recipient. The scope in the file states the selected stage and date view. Other saved candidates stay in your browser; they are not removed.
If no candidates match, change the filters or choose All saved candidates. Do not change a candidate’s decision stage just to make it appear in a file. A backup remains a separate, full-list JSON file for restoration, not a focused meeting handout.
Bring Evidence, Not Just Checked Boxes
For each candidate, prepare three short lines in an approved location:
| Question | Fictional example |
|---|---|
| What did we test? | Three invented requests containing missing dates. |
| What did we observe? | One proposed deadline was presented as confirmed. |
| What remains open? | Does an explicit unknown-field instruction prevent that error on the same inputs? |
Keep the original input and output available for human review. Do not convert a stage, date or checklist into a quality score. If one candidate has not been tested, say so rather than filling the gap with an AI-generated verdict.
Leave With One Next Check
The useful outcome may be another bounded test, a permissions question, or a decision to stop. Record what you actually decided and the next question in your private note. Update the local test date only after a real test.
Downloading the brief does not mark announcements read or complete checks. The file is a snapshot, not a shared live workspace, and reopening it later does not refresh vendor information. Check relevant official notices before reusing an old conclusion.
Use the sharing checklist before sending the file. When another test is needed, follow the small retest exercise and preserve the earlier result separately.