Background work and change feeds
The API starts schedulers for accounting, purchase invoices, mail intake, imports and customer notifications. These services carry out work that is separate from a single browser interaction. Domain services own the actual processing rules. A scheduler being configured does not establish that every pending business operation has completed.
Schedulers and enablement
Section titled “Schedulers and enablement”The API application registers scheduler startup and shutdown in backend/server.py. Individual services determine their configuration and work selection. The detailed scheduler inventory records declared cadence, enablement, overlap prevention and tenant selection. Runtime services describe where these processes execute and which component is responsible for their availability.
Schedulers and enablement lists current registrations, declared cadences, component gates and progress tracking.
Intent, processing and receipts
Section titled “Intent, processing and receipts”Transaction boundaries distinguish an atomic database intent from later processing and provider receipts.
A recorded intent, a processing attempt and a confirmed external result are separate stages. Each workflow must define the records that prove these stages and how to handle partial failure. External recovery cannot infer a successful provider change from a scheduler heartbeat or from an HTTP request being accepted.
Intent, processing and receipts compares claims, leases, deferred work and the evidence needed for completion.
Change feeds and checkpoints
Section titled “Change feeds and checkpoints”The public MCP change feed uses a caller-owned checkpoint and paginated continuations. It reports authorized change identities and operations rather than providing a whole-company export. Replays and resynchronization follow the returned contract. Permissions remain in force when a cursor or resource reference is reused.
Change feeds and checkpoints documents caller-owned cursors, replay, resynchronization and per-source coverage.
Stalled work and diagnosis
Section titled “Stalled work and diagnosis”A healthy process and a successful periodic run do not prove that every business record progressed. Diagnosis needs the specific operation, pending state and last confirmed result. Health and readiness provide component signals; workflow detection and recovery require their own scenario evidence and remain part of operational follow-up.
Stalled work and diagnosis separates liveness, completed ticks and item-level recovery.