Best for
- Reports that come back 'cannot reproduce'
- Turning a user's angry email into an actionable ticket
- Capturing a bug properly while it is still on your screen
What you give it
- What you saw, however messy — notes, a user complaint, logs, a screenshot
What you get back
- A report with numbered reproduction steps that actually reproduce
- Expected vs actual stated so the disagreement is unambiguous
- Severity and impact assessed, evidence attached, the unknowns listed as unknowns
How it works
- Extracts the observable facts from the mess and asks only the questions that unblock reproduction.
- Hunts the pattern behind 'sometimes': what the failing cases share.
- Writes steps a stranger can follow from a clean state to the failure.
- Separates fact (observed), hypothesis (suspected) and unknown — labelled as such.
Example
You: A customer says 'the export is broken sometimes'. Make this a real bug report.
Result: After three clarifying questions: a report titled 'CSV export omits the final day when the range ends on a month boundary', five reproduction steps from a fresh account, expected vs actual with the off-by-one shown in the output, scope (every export since the timezone change), severity justified — assigned and fixed the same week.
Limits — please read
- Some bugs resist reproduction; the report then captures conditions and evidence honestly, marked 'not yet reproduced'.
- Severity is proposed from impact; your tracker's politics are yours.
- It reports; diagnosing the cause is the fixer's craft (or a debugging agent's).