How a rehearsal works

You bring one real talk: a defense, a conference talk, a pitch, a class. The coach works on recorded takes of it, one at a time, and every note it makes points at the exact moment it describes.

1. Say what you’re practising

   A short guided setup asks what the talk is, what you want to say, and what you want it to achieve. Three answers plus consent is enough to start; everything else, audience, setting, style to preserve, specific concerns, accommodations, is optional and skippable.

   Formats range from investor pitch to thesis defense to a situation you describe in your own words. Formats without a dedicated rubric yet are analyzed with the general speaking rubric and say so; interview practice is general practice, not answer-structure scoring.

2. Record a take, or upload one

   Record right in the browser, with preview, countdown, and retake, or upload a recording you already have. Takes can be up to 20 minutes and 50 MB (mp4, mov, webm, or audio files).

   If recording in your browser fails for any reason, upload always works.

3. Consent, plainly

   Before anything runs you see exactly what will happen: only the audio is analyzed, which processors touch it, and how long it is kept (30 or 90 days, your choice). Declining means no analysis, full stop.

4. The take is transcribed and measured

   The audio becomes a word-by-word transcript with timings, plus measured delivery signals: pace, pauses, fillers, emphasis, energy. Specialist readers then work through the take in parallel: content and structure (message, hook, through-line, signposting), vocal delivery, and engagement (from the language alone). When you say who the talk is for, a fourth reader checks audience fit; if you don’t, that section honestly says it was skipped rather than guessing an audience. It’s built for talks in English first.

   Nothing visual is processed, even when you record video. Anything about your audience is judged only against what you stated, never inferred.

5. Every claim passes an evidence check

   Before a note reaches you it must survive a validation gate: the timestamp has to exist, the quote has to match the transcript at that moment, the metric has to come from the measured signals, and uncertain inferences have to be labeled as such. Notes that fail are dropped and recorded, not shown.

6. You read the report next to your take

   Strengths come first: two to four specific things worth keeping, at least one of them about your style. Then at most three priorities, each with why it matters for your stated aim, the evidence, a concrete suggestion, and an exercise. Measured numbers come with reference ranges, not targets, and nothing anywhere scores you as a person.

   See a sample report built from synthetic fixture data. It’s labeled as such on the page; no real voice, no real person.

7. The next take is the point

   The report ends with a three-line plan: one focus, what to keep, and how you’ll know it worked. Progress compares your next take to your last one, on measured metrics only. It never compares you to anyone else.