The questions a campus asks.

No. OralIQ never outputs an authorship verdict, an AI-use probability, or the sentence "the student wrote this". It proposes an understanding score against the instructor's rubric and files a record. The instructor sees two columns, both hers: the artifact grade and the understanding score. Where they agree, nothing happens. Where a strong artifact meets a weak understanding score, the divergence is shown with the transcript and no inference about the cause.

No. The quotient in the name is of understanding of your own work, measured against your instructor's rubric, never of the person. Every question is about a choice the student made in the submission in front of them. Nothing is scored that could be answered without having made the work, and nothing about the student is scored at all.

We project about half of them pass, and we will publish the measured number rather than hide it. The questions ask why a choice was made and what breaks if it changes; the follow-up is grounded in the answer just given, at speaking pace; and a confederate who understands the material has done the learning. The check is a weight on the grade with a record, not a lie detector.

In the first version: an individual submission of authored work that Canvas receives as text entry or a file upload. Essays, papers and memos above a minimum length, documents as PDF, DOCX or text, and code as source files or an archive. Later, by phase: group assignments, with every member checked on the whole shared artifact; problem sets and quantitative work; slide decks, notebooks and design files; resubmissions; non-English work; and discussion posts in a lighter form, one short spoken defense per module. Never, by design: quizzes and exams, the student's own audio or video recordings, on-paper work, attendance, labs and clinical performance, the quality of creative work, and identity verification as a product.

A consent block opens every recording and sits on the first page of the record, the same block everywhere. A student who declines takes the typed accommodation. No voiceprint and no face are ever stored.

The check is started by the student inside a window of days, so it is the opposite of a cold call. Each answer has ninety seconds, extended time is a setting filed from the letter, and there is no clock on the whole check. Typed is the Section 504 method, an interpreter seat is available, and the settings are logged on the record. A check run without a filed letter's setting is free.

A detector reads the artifact and asks nothing, and the detector vendors' own guidance bars its score as the sole basis for a case. OralIQ asks the student, proposes on the instructor's own rubric, and files a record the instructor signs. It sells beside that line, not against it.

$4 a paper check, and a card by assessment type that tops out at $20 for a capstone defense, under the $20 you pay for a live-proctored exam. You buy credits in $10,000, $25,000 or $50,000 blocks in any month, February included, and top up in any month. A department can pay per use monthly. Your annual quote is built from your own prior year's assignment mix in Canvas, and you pay for every analysis your instructors did not reject with a reason.

Because the product never posts a grade. Canvas holds the proposal as a provisional grade until the instructor selects, publishes and posts it. The acceptance rate will be published after the first measured term, with the share accepted unread, and every write sits in your own grade change log.

Then that course runs transcript-only, the mode faculty already accept. The proposal is a course-level choice, and the signature is always a person's.

No one, at any level. From the weekly paper to the capstone defense, the check is a second, documented examiner beside the humans in the room, with its own score broken down by section, and a person signs.

Proctoring verifies a sitting; OralIQ verifies understanding, and they are different objects. A webcam watches a two-hour exam; an essay or a program is made over two weeks, and nobody watches two weeks. Proctoring also asks the student nothing and scores no rubric. Where a campus has exams, it keeps its exam line; OralIQ is the instrument for everything else.

They get a low understanding score and whatever artifact grade the work earns. That is a grade, not an accusation, and nothing on the record says anything about how the work was made. Where a campus wants evidence for a conduct case, the divergence plus the transcript, contestable word by word, is what the office will have to read, and the conclusion is the office's under its own policy, never the product's.

OralIQ runs under your existing Canvas data agreement, with thirty days' notice before any change of sub-processor, and training permission is off by contract. Two vendors see the words. Deepgram transcribes the speech with model training opted out on every request, so audio is held there only long enough to transcribe it. Anthropic's Claude models write the questions and score the answers under commercial terms that bar training on the content and delete it from the vendor's systems within thirty days. The recording is held by OralIQ for one year with a legal hold on any request; the transcript sits on the submission as an education record the student can read and contest.

A contested understanding score is re-assessed by a person, the instructor's own oral or a second attempt, at no charge to the student. Bottom-band proposals are marked for review and cannot be selected without opening the transcript. The overturn rate will be published after the first measured term.