out-of-scopelisted

# Out-of-scope — fix it now, or hand the next session a landmine ## The stance Two contract items point in opposite directions across one case, and neither is wrong. - **CLAUDE.md item 8** — on audit, review or structural work, file one issue per distinct concern rather than fixing out-of-scope items in the same pass. - **CLAUDE.md item 3** — never close a turn reporting in-scope work as deferred-by-choice. Item 3 governs work *inside* the scope. Item 8 governs work *outside* it. Between them sits work that is genuinely out of scope **and** whose deferral costs the next session more than fixing it costs this one — and that case had no rule, so it was settled by whoever held the keyboard. This skill makes it decidable. The call is scored by the `decision-matrix` engine against six weighted dimensions, the winner is taken, and a **deferral** is written into a committed store carrying the spec that produced it. That last part is half the value: a deferral today is invisible tomorrow unless the reasoning travels with it. The tracker is where this repository has watched findings go to be re-discovered late — one burn-down found 31 of 43 examined issues already fixed on `main`, each re-established from context by somebody who did not have it. Rigid skill. The trigger predicate, the six criteria, the tie-break and the refusals are not judgment calls. The scores are. The whole resolution is recorded in ADR-0168, with the two forks it turns on in DEC-0112 (where the record liv