Transcript Standards
An Anticipated Transcript is legal work product, so we hold it to written standards — built by immigration attorneys, applied to every hearing, and visible on the page rather than buried in fine print. There is no single "perfect" transcript — even a careful human transcriber exercises judgment — so what matters is that every call we make is visible and checkable against the audio. Here is exactly what we do, what we mark, and what we never do.
We don't ask you to trust the transcript — we make it checkable. Every line carries a timestamp that links to that exact moment of the hearing audio, on the specific microphone channel that captured it. Reading and verifying are one click apart, which is a stronger guarantee than any accuracy percentage anyone could quote you. And those same time marks let you verify the official transcript when it arrives — every quotation, ours or the court's, is one click from the audio that proves it.
See it in the live demo →Transcription has always involved discretion. A human transcriber decides whether to keep every "um" and false start, how to render crosstalk, and who was speaking when the room is ambiguous — reasonable transcribers make different calls, and no two transcripts of the same hearing come out exactly alike. We replicate careful human practice as closely as we can; the difference is that wherever the recording leaves room for judgment, we make the call visible and link the exact moment so you can check it.
Courtroom audio is imperfect: people mumble, talk over each other, and trail off. When a reading involves a genuine judgment call — a word corrected from context, an ambiguous speaker, reconstructed speech — that line is rendered in bold gray and carries an inferred mark:
The recording at that moment is ambiguous; context makes the reading near-certain — so we render it, and we tell you we made the call. Click the timestamp and judge it yourself. Routine fixes with only one right answer (formatting, obvious vocabulary) are not marked; only judgment is.
Spoken citations become written citations. "Two-twelve A six A little i" renders as section 212(a)(6)(A)(i); "the 589" renders as the I-589. Your transcript should read like the record you'd cite, not like a phonetic puzzle.
We follow official-transcript practice for interpreted testimony:
The full multi-channel recording is always preserved — the transcript renders the record; it never replaces it.
Speech recognition tends to tidy human speech — dropping an "uh," smoothing a false start. We treat the audio as ground truth and preserve the speaker's actual words wherever the recording supports them. And where automated transcription has quietly normalized something, the click-to-hear link is your remedy: the audio is always one click away, which is precisely why we build the transcript around it.
Hear an example at 00:25 of the demo →