Which people must attend a root-cause session for a tracked defect, and what does an analysis based only on the defect record produce?

## Attendance follows the evidence, not the org chart An analysis is only as good as the evidence it can put on the screen while the argument is live. So the invitation list is derived, not chosen: write down the links you expect the explanation to need, and for each one name the person who can **show** the thing that settles it. Everybody else is optional, and most of them are a cost. That inverts the usual instinct, which is to invite by title and team. Seniority settles no link. Assignment on the defect record says who will fix it, which is a different question from who saw what happened. The one test that matters is: *if this link is disputed, can this person produce the thing that ends the dispute in the next two minutes?* ## Who can settle what | The link you expect to need | Who can settle it | | --- | --- | | what the change actually did | whoever wrote it, or reviewed it closely | | what the failure looked like and how it reproduces | whoever reported it and can run it again | | what the data or configuration was at the time | whoever owns that store or that setting | | why the check that should have caught it did not | whoever runs or maintains that check | | how the consuming side used the interface | someone from that side, not a proxy for them | Keep the room small. Every person present who cannot settle a link turns the session into a status meeting, and status meetings converge on the most senior person's hypothesis. Three people with evidence and twenty minutes beat ten people with opinions and an hour, reliably. ## What an analysis built from the defect record alone produces - **The chain the record's own wording already implied**, restated with more confidence than the record had. The wording of a report is itself a hypothesis, and a session with nothing else in the room adopts it. - **Links nobody present could have falsified**, because nobody present observed the thing the link describes. - **Actions assigned to a team that was not there** to say whether they are feasible, which is how an action item dies quietly. - **A written outcome the actual evidence holders would not sign.** They read it later, disagree privately, and do not bother to correct it. - **Damage to the practice itself.** Once people have sat through one session that produced a confident guess, the next one is skipped, and the technique gets blamed for the room. ## Preparing so the session can be short 1. **Confirm it reproduces**, or state plainly at the top that the analysis rests on a single unrepeatable observation. Analysing an unreproduced defect with a full room is the most expensive way to guess. 2. **Attach the evidence to the record beforehand**: the reproduction steps, the data or configuration as it stood, the change history around the relevant area, and the result the check returned. If it is attached, the session argues about meaning instead of hunting for material. 3. **Circulate the failure being explained in one sentence.** Ambiguity about what is being explained is the second-largest time sink after missing evidence. 4. **Ask each invitee which link they can settle.** If the answer is none, they do not need to be there, and saying so is a kindness rather than an insult. 5. **End every link with the same question — who can show that?** If nobody in the room can, the link is recorded as open with an owner, and the session moves on rather than talking around it. ## When the person who can settle a link cannot attend Get the evidence, not the person. Send one precise question — the data as it stood at that timestamp, the result that check returned on those inputs, the history for that area — and hold the link open until the answer lands. A short written answer to a specific question is worth more than a full room and a general discussion, and it costs the other person minutes instead of an hour. For a defect that crosses teams, this is usually the practical shape: a small session for the links your own team holds, plus two or three written exchanges for the links it does not. The underlying principle is simple enough to state in one line, and it is what separates an explanation from a story: **an analysis can only conclude what someone in it was able to show.** Everything else in the outcome is a hypothesis, and hypotheses belong in the record labelled as hypotheses.