Transcript Lake

Documentation / Examples

Keep a Lake current with the live stream

1. Goal: Follow complete source bytes as agents append them, without periodic rescans. 2. Risk: Local mutation and provider-facing reads. 3. Environment: macOS terminal with installed Transcript Lake and one existing valid Lake. 4. Preconditions: No other writ

  1. Goal: Follow complete source bytes as agents append them, without periodic rescans.
  2. Risk: Local mutation and provider-facing reads.
  3. Environment: macOS terminal with installed Transcript Lake and one existing valid Lake.
  4. Preconditions: No other writer owns the Lake; vendor transcript changes are append-only.
  5. Inputs: Existing LAKE_DATA and subsequent local agent activity.
  6. Artifacts and side effects: Appends masked canonical events and affected Oko session rows, then atomically advances only the source cursor. Vendor files remain unchanged.
  7. Steps:
sh
LAKE="/absolute/operator-owned/lake"
transcript-lake --data-dir "$LAKE" doctor
transcript-lake --data-dir "$LAKE" stream --json

In another terminal:

sh
transcript-lake --data-dir "$LAKE" status --json
transcript-lake --data-dir "$LAKE" events --limit 20
  1. Observable result: The stream prints start, then one commit line per changed source with files, failures, and ms. status reports the live process and durable cursors. Unchanged sources cause no work.
  2. Failure path: A source truncation, same-size rewrite, invalid segment, parser failure, or unreadable file logs a named source failure without advancing that cursor. Preserve non-append sources and use rebuild into an empty root.
  3. Off-switch: Send SIGINT or SIGTERM, or unload the supervising service. The process records state: stopped; retention is independent.
  4. Related operations: Query sessions, compact to Parquet, or reconstruct the Oko projection.