A deliverable is one generated report about one lead product — a tracking plan, a business case, an engagement playbook. Each product has at most ONE deliverable per type: generating again reuses that slot and replaces the content, so "regenerate" and "generate" are the same action on the same address.
A deliverable moves pending → generating → ready | failed. A failure always carries a named reason: step_failed (an AI step errored), no_output (the run produced nothing usable), spending_cap_exceeded (the account's AI budget ran out — raise the cap and regenerate), or stale_swept (the run died and was cleaned up after 2 hours — regenerate to retry). A ready deliverable has a 24-hour regeneration cooldown; a failed one can be retried immediately.
Every paid step of a generation writes a ledger row (tokens, model, cost), and the deliverable records the exact process and prompt versions that produced it — its provenance is auditable on the report page.
Related: Process · Share link · Spending cap