What you see
Inside the opinion card:- Confidence factors: at most four lines, each marked as supporting (✓) or cautioning (⚠). Examples: Root-cause analysis is high-confidence; 5 supporting observations in the analysis; Open unknown: …; A competing explanation isn’t ruled out. Supports lead, caveats follow, and at least two caveat slots are reserved so weak spots are never crowded out by supports.
- Why not the obvious move: the reflex an on-call engineer would reach for (a blind rollback, a restart) and why it is not the recommended move here. It appears only when there is a genuine reflex to argue against and it is distinct from the chosen move.
Where the factors come from
The factors are derived by the backend from signals it holds, not written by the model:
“Why not the obvious move” is the one model-authored piece, kept only when it is distinct from the chosen move and from the cause tradeoff.