The tools around the record. Each one runs inside the same boundaries as the transcript
itself, and each is listed with its limit.
Find in a transcript never leaves the page
The review pane's find runs in the browser, against the transcript the page already
holds. The query — a phrase typed about a recording's content — is never sent to any
endpoint and never enters a URL, so it appears in no server log and dies with the
page.
Limit: it is a case-insensitive plain-text match, not a pattern
language, and it matches within a line — a phrase that spans two lines is not found.
Searching across recordings is the matter search, which is a request to the server.
Matter search returns the line, not just a hit
One search runs across every transcript in a matter and returns each matched line as
an excerpt with the hits marked, grouped by recording — you read the sentence around the
term before opening anything. It works by decrypting and scanning the matter at the
moment you search, which is the same reason there is no plaintext index.
Limit: one matter at a time, inside that matter's encryption
boundary — there is no search across matters. And the match is a plain substring,
case-insensitive, not a query language.
Speakers can be merged, unmerged, and split
Two tracks that turn out to be one person can be merged; a merge is a reversible
pointer, and unmerge is the reversal. A track that holds more than it should can be
split from a chosen line onward onto a new track, and the new track records which
speaker it was split from, so that is reversible too. Nothing is destroyed either
way.
Limit: a split is your judgment, recorded as yours — the new
track starts unnamed, with no claim about who is on it, and the product still will not
split one microphone into the people it heard on its own. Channel-derived tracks cannot
be merged at all, in either direction: channel identity is separation the recording
itself established, and a merge would destroy it.
A video source plays beside the transcript
When the uploaded file is video, the review pane plays that original file next to the
transcript, and the video element becomes the pane's clock — one clock, not two, so the
line and the picture cannot drift apart.
Limit: the transcript still comes from the audio alone; nothing
is read from the picture. Video is detected from the uploaded file's name, and the
decision is presentational — an audio-only file in a video container plays as sound over
a black frame, and transcription behaves identically either way.
A matter is shared on the record
An admin on a matter can grant a colleague access to it by their email address, at a
level — read, write, or admin — and revoke it the same way. Every grant and every revoke
appends a row to an audit log recording who gave what to whom, and when. Those rows
carry no client content, which is what lets them survive even the matter's deletion.
Limit: sharing stays inside your firm — the address must belong
to a user of the firm, and nothing is emailed anywhere. Firm-wide roles do not bypass
the wall: a firm owner who needs a matter grants themselves access through the same
audited path, leaving a row that says so.
The cost is quoted before it is spent
Every recording is quoted on the upload page in the unit the price is charged in: its
audio minutes, split into the minutes the firm's remaining free grant covers and the
minutes it does not. The same arithmetic is enforced at an admission gate at the last
instant before the first paid transcription call — a job held there has spent nothing
while it waits.
Limit: no quote exists until the file has been measured, and
until then the page shows the unknown rather than a zero, because a zero that means
"not yet measured" reads as "free". The rate itself is on the
pricing page and nowhere else — this page included.
