Interface StartEvaluation

Overall preflight startability: every supplied initial-field row is valid and every declared start requirement is satisfied. outcome describes only the predicates; structurally invalid rows can therefore produce allowed: false with outcome: 'satisfied'.

Supplied rows startInstance would reject before writing. Empty means every row targets a workflow-scope, input-sourced declaration exactly once. Render these as corrections, not an authorization verdict.

Required input entries initialFields doesn't fill yet — the rows a startInstance now would throw RequiredFieldNotProvidedError about.

Three-valued, BINDABILITY-AWARE verdict over the start predicates: 'unsatisfied' is a definitive no for these inputs (disable and explain), 'unevaluable' means the verdict can't be decided yet — either a null operand reached an ordered comparison, or a predicate reads an entry initialFields doesn't supply (StartEvaluation.unboundReads non-empty). One exception outranks the provisional rule: any definitive requirement refusal keeps the aggregate 'unsatisfied' even while other reads are unbound. Keep the affordance enabled on 'unevaluable'; startInstance still enforces the final verdict, where absence is final rather than provisional.

One result per declared readiness node, in author order.

The $fields entries the predicates read that initialFields doesn't supply — "fill these to decide", including a singleSubject node's implicit subject read. Non-empty makes undecided requirements 'unevaluable'; a separate definitive refusal still keeps the aggregate outcome 'unsatisfied'. Empty when the definition declares no requirements.