0015. Filler sources fetch on their own
- Date: 2026-08-03
- Status: accepted
- Supersedes: “there is no unattended crawler”
Context
Adding a source did nothing until someone pasted a URL or approved a pull, so an operator had to return by hand to keep a catalog growing. The earlier rule against an unattended crawler existed for a real reason: unattended fetching can fill a stranger’s disk.
Decision
Registered, enabled sources are checked on their effective schedule (filler.fetch.every, with
nullable per-source overrides) and new items download without anyone asking. The concern survives
as limits rather than a prohibition: per-source and per-provider counts per check, a catalog
ceiling, and a storage governor with a soft allowance and a hard host reserve. Everything fetched
arrives held, and a limit that is reached is reported.
Consequences
- Clips arrive because a source was added, not because someone keeps asking.
- Bulk backfill of a large collection stays a pull, where a human approves the plan.
- “Nothing new arrived” must always be explainable from live state.