Verity

Transcripts.

For educators — what was taught, against what the machine says was taught.

Your students are studying from AI summaries of your lectures. The summary grades itself, and nobody checks it against the lecture. Verity does. You submit the transcript and the summary made of it; every claim in the summary is tested against every segment of the transcript — not just the passage it quotes — and the verdicts come back sealed: SAID, NOT SAID, SAID OUT OF CONTEXT, CONTRADICTED LATER IN THE SAME RECORD.

Start an intake Watch a sample run How claim-checking works
1 · Intake 2 · Validated 3 · Transformer — proposals 4 · Gate — verdicts 5 · Output — sealed

Sample run — a two-minute lecture on the water cycle, and the AI summary a student brought to class.

✓ supports  ·  ✕ contradicts  ·  ◑ context  ·  · no relation  — columns are the transcript’s segments; rows are the summary’s claims.

The run

Five stages, in the open

  1. Intake. The transcript and the summary arrive together and are segmented — the transcript into timestamped spans, the summary into its individual claims. Rendered as a matrix: claims down the side, segments across the top.
  2. Validated. Before anything reads anything, the pair is bound: both documents hashed, the scope fixed — this summary, this transcript, nothing else. What was submitted can never quietly become something else mid-run.
  3. The transformer reads. For every cell of the matrix, the model proposes a relation — supports, contradicts, context, none — and points at the exact lines it is standing on. Proposals, with receipts. A claim is tested against the whole record, including the parts after the passage it quotes.
  4. The gate decides. The verdicts are not the model’s. Rules over the proposals return one verdict per claim: SAID — NOT SAID — SAID, OUT OF CONTEXT — CONTRADICTED LATER IN THE SAME RECORD.
  5. Output, sealed. The finished matrix — every verdict traceable to the segments that decided it — is sealed as a record. Yours to keep, forward, or attach to the syllabus.

Scope

What you send, what you get

  • You send A transcript (lecture, seminar, recorded lesson — text, captions, or a document) and the AI-generated summary, study guide, or notes being checked against it. Both stay yours; runs are private and the record is not published.
  • You get The verdict matrix: every claim in the summary with its verdict, and the exact transcript segments behind each one — including the contradiction the summary stepped over, quoted in full.
  • Also checked Public-record transcripts — earnings calls, hearings, interviews — against the quotes drawn from them. Same matrix, same verdicts; see claim-checking.
  • Not this Not a grader of students, and not a plagiarism hunt. The machine’s summary is what’s on trial — never the student who brought it, and never you.

Pricing

Negotiable — by letter, like everything here

There is no price sheet for transcript checking, deliberately. A single course, a department, and a district are three different shapes of work, and education budgets are not enterprise budgets. Runs are scoped through a short letter — which courses, which transcripts, how many summaries, what turnaround — with the price fixed in the letter before the first transcript moves. If the scope is wrong, who signed what is already on the record. Write first; the number comes out of the conversation, not out of a funnel.

Intake

Submit a run

Runs are taken by mail while the service is young. Fill in what you have; the button composes the intake email — nothing is sent until you send it.

Or write directly: pilot@verityverdict.com