A local blob file that failed to unwind after a host write failed. Names the
blob so a host learns which files are left behind, not just that some were.
Everything one active Circle contributes to staging its device’s next Circle
acknowledgement: the exact control and epoch the live projection derives from,
the access authority that seals the acknowledgement, and the retained bootstrap
coverage the projection was seeded from (None for a founder/source device).
The joining device’s own journal of in-flight join attempts, kept in its own
SQLite file under the stores root because the Store database it is joining
does not exist yet while the attempt runs.
Schema info for all synced tables: maps table name to column indices. Built
once before an apply and moved (owned) into the conflict closure, which must
be 'static.
Changeset bytes paired with the exact synced schema that validated their row
identities. Construction is the one identity-validation boundary; apply only
accepts this type, so callers cannot parse once for classification and then
parse the same bytes again during mutation.
How far a gated root’s make-remote transition has got, as its durable
intent records it. The write id a publication carries is bookkeeping the
transition owns, so it is not part of this.
Staging audience-move blobs into the spool needs the connected staging
owner replication composes. The capture transaction names only this port;
the returned rollback closure is consumed on failure.
Apply bytes to conn, resolving conflicts (premerge + row arbitration),
building the TableSchema from tables once. A convenience wrapper over
resolve_and_apply_changeset_with_schema for callers that apply a single
changeset and don’t already hold a schema (tests, snapshot round-trips).
The stored bytes of one operation’s objects, supplied alongside the
operation that names them so the database can install the bytes and their
owner claims at the same durable boundary.