Where we start
Session 1 said the case for sketchnoting is selection, not the pen. Session 2 built the marks so selection could be fast. A claim like that is easy to nod at and hard to know you have followed. A bench is a claim you cannot nod at: it reports a number, and the number is either what you hoped or it is not.
★ What you walk out with tonight. Your own capture of a real talk at real pace, with its numbers: ideas kept out of twelve, verbatim share, words per idea. And the honest knowledge of which rung of the ladder you started on.
Every number below is computed from the box against the transcript. Ideas are matched on cue words and light up as chips so you can see which registered. Verbatim overlap uses the definition from the note-taking research: the share of your words that sit inside a run of three or more consecutive words the speaker also said consecutively. The only thing that is not a measurement is the one-line verdict, and it wears a badge saying so. And one thing is not measured at all, on purpose: the drawing. A browser cannot see your icons. It can see what they cost.
What gets measured, and which claim each number belongs to
6 min liveSeven numbers, and each one is a claim from an earlier session turned into something countable. One of them is the single finding from the handwriting literature that survived every replication, which is why it is the one the bench leads with.
LiveWhy verbatim overlap is the number that matters3 min▶
Mueller and Oppenheimer measured it as the share of a student's note words that sat inside a three-word string also found in the lecture. Laptop notes: 14.6%. Longhand: 8.8%. The effect was large (d = 0.94) and more overlap predicted worse answers on conceptual questions (beta = -0.43, p = .005). When the study was replicated in 2021 with 142 students and 88 co-authors, the quiz result did not hold, and this one did: laptop 12.97%, longhand 8.13%, g = -0.85. Telling typists not to transcribe, in the original study 2, "was completely ineffective at reducing verbatim content."
So the bench uses that exact definition. A three-word run you also typed is a run you did not think about; you were a microphone for those three words. The share of your capture that is microphone is the share that did no selecting. Under 25% and you were listening. Over 40% and you were typing.
Self-studyWhat the bench refuses to model3 min▶
There is no recall number. Session 1 laid out why: the doodling study licenses a claim about attention under a dull stream, the drawing-effect studies license a claim about encoding single words, and the handwriting literature licenses a claim about transcription. None of them licenses "this capture will be remembered X% better", and a bench that printed one would be inventing it.
There is also no drawing score, and this is the more important refusal. A browser can measure a textarea. It cannot see that your icon for legal was a beautiful pair of scales. What it can see is that while you drew the scales, the founder said four thousand bags, and the bags are not in the box. The anti-preset below is exactly that page: five lovely lines and eight missing ideas. The bench measures the cost of drawing, which is the only thing about drawing that can be measured here, and the only thing about it that matters at speaking pace.
Nine minutes, four captures, one trap
★ 26 min · everyone drivesLoad the four presets in order and read the numbers before you read the verdict. Then press play, capture the talk yourself with your session 2 icons beside you, and stop when you stop. The bench works on whatever is in the box.
The selection bench
The founder's autumn roast talk, 557 words, twelve things worth keeping. Every number except the verdict is computed from your capture against the transcript. The twelve ideas light up as they register.
The ladder below was read off this bench in a browser before this page was written. Load the presets in the same order and you will get the same numbers, because they come from the same measurement.
LiveRead the ladder: more words, fewer ideas4 min▶
- Type everything: 286 words, 84% the speaker's, ten of twelve. The typist was still on the pricing paragraph when the photos and the checklist went past. The longest stolen run is 44 words: for 44 words in a row the typist was a microphone. This is the laptop condition from the research, and it loses ideas at the end because it spends all its time at the beginning.
- Keywords only: twelve of twelve, 36 words, three per idea. Complete and useless. "2nd wk Oct" brings the date back tonight and nothing back next month. Session 5 is about the reader who is not you; this rung fails that reader entirely.
- Keyword and one line, your words: twelve of twelve, 110 words, zero verbatim. A handle to find it by, then the idea in the listener's own words. 35% of the words appear nowhere in the transcript, which is the sound of somebody thinking. This is the target rung and it cost 74 more words than the one below it.
- Draw everything beautifully: four of twelve. Every idea got an icon and a banner and the talk kept going. The bench cannot see that the icons were lovely, and neither can anyone reading the page next week. It is the honest cost of the art end of Rohde's scale, and it is why session 2 made you build the icons before the meeting.
LiveNow yours: play, capture, read10 min▶
Clear the box. Press play at 150 words per minute, which is conversational pace. Capture into the box using your session 2 vocabulary: a handle in capitals, then the idea in your words on the same line, one line per idea. When the talk ends, or when you stop, press pause and read the seven numbers before you read the verdict.
Then look at the chips. Every idea that did not light up is a specific moment you were doing something other than listening, and you know what it was. Most first attempts land between the top two rungs: eight to ten ideas, twenty to thirty percent verbatim, because the speaker's phrasing is sticky and the hand copies it. Play it again at 130 and see if the verbatim share drops. Then once at 175, because real speakers do not wait.
If you finished with time to spare, paste in the notes from a real meeting this week. The idea chips will not light, because the transcript is the founder's, but the verbatim share still means something if you have the meeting's transcript or recording: the bench is a general instrument, and the twelve ideas are only the ones keyed for this talk.
The number to bring to session 4. Your words per idea on the best run. Under four and your page will not be readable next week; over fourteen and you were writing sentences. The rung that survives a week sits between eight and twelve, and layout, which is next session, is where those eight to twelve words go so that somebody can find them.
What the numbers can and cannot tell you
3 min liveA bench that only reports what it can measure has to be clear about what it cannot. Four limits, stated on the widget and here.
| Limit | What it means for your numbers |
|---|---|
| Selection, not drawing | The bench grades what survived into text. An icon that stands for an idea counts only if you also typed its handle. On paper, the icon would count; the bench cannot see paper. |
| Cue matching on stems | "Sidamo" registers the Guji idea. A private abbreviation nobody else would read might not. The chips show which ideas registered, so a miss is visible and you can decide whether the cue or the note was at fault. |
| Three-word runs | The verbatim rule is the one the research used. A phrase reordered scores as your own words even when the vocabulary is borrowed. That is a known softness of the measure, in the papers as well as here. |
| The verdict is a rule of thumb | Its thresholds (six ideas, 25% and 40% verbatim, four and fourteen words per idea) are judgement, badged as such. The seven numbers above it are not. |
Self-studyWhy the anti-preset is a capture and not a drawing3 min▶
The honest way to represent "spent the time drawing" in a text bench is to show what was left over when the drawing was done: five lines, four ideas, and parenthetical notes about the icons that ate the other eight. Nobody who has sketchnoted a fast talk will find that page unfamiliar. The bench does not pretend to grade the icons; it grades the page they cost, which is the page the reader gets.
A Daybreak designer ran the bench three times. First run: nine ideas, 31% verbatim, because she typed the founder's jokes. Second run at 130 wpm: eleven ideas, 12%. Third run at 175, with her twenty icons on paper beside the laptop and only the handles typed: twelve ideas, 4%, 88 words. The icons did not appear in the bench at all. They appeared in the time they saved.
Before session 4
25 min- Run the bench twice more on the founder's talk, at 130 and at 175, and write down your ideas-kept and verbatim share for each. The gap between the two speeds is the gap your layout has to absorb.
- Capture one real meeting on paper with the handle-and-one-line shape, then type only the handles and lines into the bench box. The chips will not light, but the words-per-idea number will, and it should sit between eight and twelve.
- Find the longest run of the speaker's words in your real notes from this week, by hand. If it is over ten words, you were a microphone for that stretch. Write what the idea was in your own words underneath.
- Bring an agenda you know in advance for a meeting or talk next week. Session 4 draws the skeleton before the speaker starts, and it needs a real agenda.
Three questions
Clear all three and session 3 gets stamped in your passport.
1 · The "type everything" preset captured 286 words and the "keyword and one line" preset captured 110. Which kept more of the talk's ideas, and why?
Ideas are counted on cue words and the chips show which registered. The transcript preset has a 44-word verbatim run, and for those 44 words the listener was a microphone. Two ideas arrived while the microphone was still running and never made it to the page.
2 · Why does the bench measure verbatim overlap rather than recall?
Mueller and Oppenheimer's quiz result did not replicate; their mechanism result did, twice. A bench reports what has evidence behind it. The share of your words that were the speaker's is that number, and it is the number the widget leads with.
3 · The "draw everything beautifully" preset scores four of twelve. What does that number measure?
The bench cannot see the icons and says so. It sees the page they left behind: five lines, four ideas, eight gaps. Drawing is not banned; it is priced, and the price is paid in ideas. Session 2 exists so the price is low.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Session 3 cheat sheet · pin this
You know what survived and what it cost. Session 4 is where it lands: Rohde's seven page layouts, how to pick one from the agenda before the speaker starts, and the same twelve ideas laid out two ways so you can see which page a colleague could find something in.