One source-authored annotation to register. source_key identifies the
annotation within the source system; registration is idempotent per
(source_system, source_key), so a full-content pass re-asserting the same
key updates in place rather than duplicating.
One source column’s pinned replicate index, as the API’s register response
reports it (and as stream metadata persists it under
replicates.assignments). Sync services assign each value’s
replicate_index by looking its source column up here.
One source-authored site note to register. source_key identifies the note
within the source system; registration is idempotent per
(source_system, source_key).
Replicate-family declaration on a stream registration. The API pins each
column’s replicate_index server-side (append-only across re-registrations)
and returns the authoritative mapping as ColumnAssignments on the
register response; source_columns order is provenance, not the index.
The API requires at least two unique columns and measurement_type: "spot"
on the request.
A completeness claim over one stream: the readings sent alongside are the COMPLETE content of
the source for this stream over [from, to), read from source_rows_read source rows. The
server diffs stored content against the payload and converges (new / changed / withdrawn);
without a window the request is a bare append, exactly the old semantics.
One portal standard curve to register. source_key identifies the curve
within the source system; registration is idempotent per (source_system,
source_key).
The cadence vocabulary a reading is classified under. The wire fields stay strings, so a
producer names a value through this rather than spelling the literal and a consumer refuses
what it does not recognise instead of falling through to the default.