Can Viva catch a student reading from a script?

Two independent things catch this, and they catch different students.

• Read-aloud detection. Speech that is being read has a different rhythm from speech that is being composed: flatter, unusually fluent, without the ordinary hesitations of someone thinking. It lives in one place, the Reading pattern row inside Speech signals, and it is EVIDENCE rather than a flag: it can take that panel to Minor signals, but it never raises the flag on the roster and never appears as a signal under the verdict. The rhythm is measured against a single enrolment sample, people read differently, and the score bands are being re-derived on a rebuilt measurement, so on its own it is not enough to put an accusation next to a student's name. Read it beside the follow-ups below, which are the stronger evidence. The comparison needs the student's own voice sample; without one the row says so instead of showing a score. Voice match is a separate panel and answers only whether the same person was speaking.
• The row always gives the most basic reason first: Needs a voice sample when the student had none at the time (their first interview is what creates it), Not enough audio for a very short conversation, and Could not be measured when a sample exists but Viva could not measure the speaking rhythm of that recording. That is a limit of the measurement, not a problem with the student or their audio, so one student can show a percentage on one attempt and Could not be measured on the next. A rhythm score with no pauses behind it would say nothing about whether answers were read. Could not be measured belongs to recordings measured before 29 September 2026. Since then pauses are measured inside the student's own speech, leaving out the time spent listening to a question, so a recording only lacks a rhythm when the student said very little, and the row then reads Not enough audio. Until the score bands are re-derived on the new measurement, a conversation measured this way shows a reading score only when both it and the voice sample carry word-timed pauses; otherwise the row is left out rather than showing a score Viva cannot stand behind.
• Word-for-word overlap. Separately from rhythm, Viva compares what the student said with their own submission and counts the spoken words that sit inside runs of eight or more words appearing in the same order in the text (six for Korean). This is text against text: no microphone, no voice sample, nothing acoustic. The Read from the submission row in Speech signals shows the share. A quarter is shown in amber; at half or more the conversation is flagged on the roster and under the verdict, because reading your own essay aloud is about as unambiguous as this gets.
• Follow-ups. This is the stronger one. A prepared answer covers the question that was expected; Viva then asks something that was not, about the specific thing the student just said. A script has no next page.
• Response time. The report compares how long the student takes to start answering with their own earlier interviews. A pause far longer than usual, followed by a smoother-than-usual answer, is reported as a pattern rather than a verdict.
• Stalls. Silences of 20 seconds or more before an answer, and every time Viva had to repeat a question, are counted on every interview. A repeat counts however the student asked for it, and also when they gave only a few words and Viva asked the same thing again. The count is always shown; it raises a flag only as a pattern: 3 or more repeated questions, or 2 or more silences. This needs no history, so it is there on a student's very first interview, with the breakdown under the count.
• Keyboard sounds. Viva listens to the student's microphone for runs of key presses while the student is not talking: six or more in a row. A single click, a mouse or a pen is not counted, and nothing is looked for while the student speaks. The Keyboard sounds heard row in Focus & environment says how many times they were heard, each moment is marked in the transcript and on the recording bar, and clicking a mark plays that moment. Taking notes is legitimate, so the report says what was heard and when, never why. A sound on a steady beat, such as a ticking clock, is recognized and left out. In two orders the panel reads Review and the conversation is flagged, on the roster and under the verdict: keyboard sounds just before or after the student asked Viva to repeat the question, or together with a pause of 10 seconds or more before the answer. Keyboard sounds anywhere else are shown and never flagged. The flag asks you to listen to the moment; it does not change the verdict or the score. It is measured on interviews from 29 September 2026; earlier reports do not show the row. A quiet or heavily noise-suppressed microphone can hide key presses, so 0× means none were picked up, not that none were made.