Command reference
oko index sync
Brings this person's transcript catalogue in the shared Oko database up to what Transcript Lake holds. The text comes from Lake, already masked; changed sessions are read newest first. Identical retained Lake events count once when building searchable conversation text and classifying recorded repairs. Each rewritten session also has its frustration moments recorded.
Invocation#
oko index sync [--limit N]Inputs and options#
- --limit bounds how many sessions are considered; 0, the default, is every one.
Output and state effect#
- Writes session and repair rows, then prints how many Lake sessions, changed sessions, rows and frustration moments it read or wrote. Repeated exact Lake events do not become extra repairs, duplicate conversation text or inflated operator-message counts.
Refusals and safe exits#
- A non-numeric --limit (exit 2); the Transcript Lake refusals of `oko lake sources`; the shared database refusals of `oko database check`; `repair session selection could not be encoded: <error>`; `upsert <table> on (<keys>), <n> rows: <database error>` identifies a failed write rather than claiming an empty catalogue.
- When the one-off model judgment of a batch fails, that batch's sessions are still written and every later session is still catalogued; the sync then exits unsuccessfully with `oko index sync wrote <n> catalogue rows, but <m> of them have no one-off judgment, so their one-off repairs are not recorded and the next sync judges them again: [<session ids>]: one-off judgment for session <id>: one-off model judgment failed: <the model route's answer>; …`, one entry per failed batch. Those sessions keep an old one-off version, so the next sync judges them again.