Scope question: does A2A intend to standardise budget/audit semantics for runs, or is that out of scope?

[qlheric]

Hi — first, thanks for the spec; v1.0.0 already gives us more than most frameworks do.

Question (scope): A2A specifies task lifecycle, idempotency semantics (§3.3.1), and terminal-task behaviour. For a control plane implementing budget accounting, a hash-chained audit trail, and capability-tiered tool exposure, we had to make decisions that the spec (correctly, I think) leaves to implementations:

1. Budget: integer micro-USD with a two-phase reserve → settle, retries must not double-reserve. Is that intentionally out of A2A scope?
2. Audit: an append-only hash chain over canonical JSON so two implementations can verify each other's logs. Would A2A consider an "auditable task record" extension, or should this stay an implementation concern?
3. Vocabulary: we currently use waiting_approval and outcome_unknown. Would the A2A community prefer input_required and an explicit UNKNOWN terminal/non-terminal state?

Context (not a product pitch): we wrote a small review-able contract + fixtures and implemented it twice (Java + Python) to check cross-language consistency — https://github.com/qlheric/agent-run-semantics-contract. A related technical note on ambiguous outcomes (DOI 10.5281/zenodo.22969881) covers the execution-safety model we are trying to make checkable.

If the answer is "these are out of scope by design", that is a perfectly useful answer — it tells us whether to propose an A2A profile/extension or keep it separate.