The baseline
Three counts, from one transcript, taken twice twelve weeks apart. Everything else in the course is judged against the change.
Count one: disfluency rate per hundred words. Fillers, meaning "um", "uh", "er". Repeats, meaning a word or phrase said twice in succession. Restarts, meaning an abandoned sentence begun again. Count them, count your total words, divide, multiply by a hundred.
The reference figures: 5.97 per hundred words overall in the conversational corpus, 7.00 for the speaker carrying the planning load, 4.93 for the responder. Do not treat these as targets. They are context for whether your number is remarkable, and mostly it will not be.
Count two: time to the point. For each substantial contribution, how many words before a listener could state what you were saying. Not what you were talking about: what you were claiming. Mark the word where your point becomes recoverable. If it never does, record that.
Count three: structure present or absent. For each contribution, mark whether it had a recognisable shape a listener could follow, or whether it was a sequence of true statements in the order they occurred to you. Binary. Do not grade it.
What people find in the first transcript is almost never the disfluency count. It is how far into a two-minute answer the actual claim appears, and how often it never appears at all. That discovery is worth the whole exercise, and no amount of feeling how the answer went produces it.
A note on the binary decision in count three, because it is borrowed from the only training programme that has measurably raised charisma in a controlled trial.
In that programme, two trained coders assessed speeches for the presence of twelve specific tactics. The decision was deliberately binary: "either a CLT was appropriately demonstrated or it was not, regardless of the frequency". They achieved 85.03 per cent agreement across 72 observations, with κ = .67 (Antonakis, Fenley & Liechti, 2011).
The design choice is the lesson. They could have counted metaphors. They chose not to, because frequency invites a coder to trade quantity against quality and the reliability collapses. Present or absent is a question two people can answer the same way after twenty minutes of training.
Apply the same discipline to your own transcript. You are not rating how good your structure was. You are marking whether a shape was there. When you do this in week fourteen you will make the same binary call, which is what makes the two counts comparable.
One practical caution about transcription. Automatic transcription tools delete disfluencies by default, because they are built to produce readable text. A cleaned transcript will show you a fluent speaker who does not exist. Either turn that setting off, transcribe by hand, or count the fillers by ear from the audio while reading along. This trips up nearly everybody who attempts Exercise 0.A the fast way.
You transcribe your recording using an automatic tool and your disfluency rate comes out at 0.4 per hundred words, far below every reference figure. What is the most likely explanation?
Why does this level ask for a binary judgement on structure rather than a rating out of five?
Notes are kept with your account, alongside your progress and your gate claims. The lesson itself is readable without one.
This lesson has a tool
Open it and get your draft reviewed. Drag-and-drop tools need a wider screen; the review works anywhere.