Skip to content

Session Metadata

.scribekit/session.json, beside the transcript. It is ScribeKit's bookkeeping: the transcript is the user's document, and losing or failing to parse this record never makes the transcript unusable.

Lifetime

Written before a meeting begins, and updated only after its transcript has been flushed and closed — and after any retained recording has been finalised. That ordering is the durability guarantee: a record saying a meeting completed is a claim about files that have already closed.

A completion that cannot be recorded leaves the session recorded as unfinished rather than falsely as finished.

What it records

Where the session stands: its identity, its MeetingState, its times, the applications captured, and the recognition language. It holds no transcript text.

It also records where the meeting's audio came from, as captureMode: "applications" for App Audio, "microphone" for a Microphone meeting. For a Microphone meeting the source name is written as Microphone (<input name>); the input device's identifier is never written. captureMode is additive and optional, like every field added after the first release of the record: a record written before it existed — every v0.1.0 session — decodes with it absent and is read exactly as before, and the schema version is unchanged.

MeetingState is the persisted domain lifecycle, with explicit transition rules. The runtime's MeetingRuntimeStatus — which has a failed case the persisted enum does not — is a presentation derivation and is not what is stored. See Meeting Lifecycle.

The distinction the states carry matters: an unexpected capture termination is recorded as interrupted, not completed, because a completion additionally claims that the meeting ended because someone ended it. See Failure Semantics.

Transcript seal

transcriptSeal is evidence about the transcript, recorded once, when the meeting closes:

"transcriptSeal" : {
  "algorithm" : "sha256",
  "byteCount" : 4182,
  "digest" : "9f2c…"
}

It is written after the transcript has been flushed and closed — and after any recording has been finalised — by reading the closed file back in 64 KB chunks, and it goes into the same atomic replacement of session.json that records how the meeting ended. There is therefore no moment at which a record says a meeting closed while carrying a checksum for some other version of its file. It holds no transcript text, and nothing is added to the Markdown.

It is additive and optional, like captureMode: records written by v0.1.0 and v0.2.0 decode with it absent and are read as unverified, and an earlier build ignores it. The schema version is unchanged. It is absent, too, while a meeting is still open, when ScribeKit stopped before the meeting closed and recovery marked it interrupted, and when the closed file could not be read back — in which case the completion is still recorded and the meeting is not failed for it. Absence is never treated as damage.

ScribeKit never updates a seal to match a transcript that has changed, never rewrites a transcript to match its seal, and never sends either anywhere. See Transcript integrity for what the check guarantees.

Reading it back

  • A session marked in progress at launch is offered for recovery.
  • A damaged or newer-format record is reported and left exactly as it is. ScribeKit never repairs, rewrites or deletes one.
  • A directory with a ScribeKit transcript and no record is listed as a legacy meeting, with only the facts the transcript itself states.
  • A record must exist for a meeting to have an identity that notes and reviewed marks can attach to.

None of that is load-bearing for the transcript, which reads perfectly on its own.