What happens to audio the system could not transcribe

The failure that makes a transcript untrustworthy is not a wrong word. It is a missing stretch that the page does not admit to. If thirty seconds of a call produced no transcript and the document simply continues, nothing on the page tells you to go back to the audio — and the absence of a sentence reads exactly like the sentence never being said.

Pincite Audio prints a marker where the words would have been.

What the marker looks like

An untranscribed span renders as a single bracketed line carrying the range it covers:

[NO TRANSCRIPTION RETURNED 01:01:01–01:02:05]

A gap is a fact about one channel. Where a recording's channels were recorded separately, the marker names the one it belongs to, using the same designation the speaker lines carry:

[NO TRANSCRIPTION RETURNED ON CHANNEL 2 01:01:01–01:02:05]

Where a recording has no channel designations the marker names none, because a channel number would be a coordinate that appears nowhere else in the transcript. It never names a party: which party is on which channel is not something the recording establishes.

Timestamps are floored to the second and zero-padded, and the hour field is not truncated — a gap two hours into a recording renders as 02: rather than starting over. A span of zero length cannot render at all; the renderer rejects it rather than emitting an empty marker.

The string is produced by one shared function. It is not assembled independently by each surface, which is what keeps the same gap from reading two slightly different ways depending on where you meet it.

Where it appears

On the transcript you read in the app, and in the data an assistant receives. Not in a PDF: a recording with a gap of this kind is held for review, and no PDF is produced for it at all.

That is a deliberate line rather than an omission. A recording the system cannot fully account for does not become a finished export; the gap marker is how you learn that while you are reading, and the export simply does not exist until the recording is accounted for. Gaps are stored with the transcript rather than applied as formatting at export time, so the surfaces that do show one are reading the same stored rows through the same renderer instead of each deciding for itself what to show.

Silence is not a gap

A recording is mostly not speech. Pauses, hold time, the gap between a question and an answer — if every one of those printed a marker, the page would be more marker than transcript and you would stop reading them.

So detected silence is recorded as a benign cause and does not render a notice. Every other cause does. The distinction is stored, not inferred at display time, and the full set of causes is pinned on the format page, which is checked against the code rather than maintained by hand.

What the marker does not tell you

It records that no transcript was produced for that span. It does not tell you what was said there, and it does not claim the span was unintelligible — only that nothing came back for it.

That is the honest limit of the mechanism, and it is the reason the marker carries a timestamp range: the range is what lets you go to the audio and listen to exactly that stretch, rather than re-reading a document that has already told you everything it knows.