Audits the notes and the timing of a deck that already exists, on the rule that the slide, the spoken word and the notes each carry a different thing.
We have not measured this one. It is published untested, and the part a capable assistant already produces unprompted is the obvious advice: do not read your slides, rehearse, leave time for questions. That is not the content.
The content is a contract and some arithmetic. The contract says the slide carries what must be seen, the spoken word carries the argument, and the notes carry what the speaker will forget, with the corollary that a notes field duplicating the slide is a notes field that will be read aloud. The skill then enumerates what actually belongs in notes, which is a specific list of six things most notes fields contain none of: the transition sentence into the next slide, the number you will be asked for that is one level below the number on the slide, the source and date of every figure, the thing not to say, the question this slide always provokes with its answer, and the cut marker.
The arithmetic is more pessimistic than the arithmetic people do. A slide with a headline and a sentence is not the same length as a slide with a chart the audience has never seen, because the first ten to fifteen seconds after an unfamiliar chart appears are spent reading it and anything said in that window is not heard. It also budgets a talk backwards from the end and protects the ask first, on the basis that the ask is the thing that always gets cut and is the only reason the meeting happened.
Who it is not for. A short internal update to a team who already know the material, where the deck is a shared reference and nobody is presenting to anybody. This is for a talk with a booked slot and something at stake.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier B with a Tier A spine. One real deck with populated notes, audited twice. The objective half is countable from the file: figures with no source in the notes, slides with no transition sentence, absence of a cut plan, and predicted runtime against the booked slot. The quality of a rewritten notes field needs blind pairwise preference, since there is no single correct version.
The awkward part is that half of this is countable and half is judgement. Countable: how many figures on the slides have a source recorded in the notes, how many slides have a transition sentence, whether a cut plan exists at all, and whether predicted runtime fits the booked slot. Those can be scored from the file by somebody who never saw the skill.
The other half is the quality of a rewritten notes field, where there is no single correct version and the only fair method is a blind preference between two runs. And the deeper problem is that the real outcome, whether the talk lands and finishes on time, is only observable by delivering it twice, which is not a test anybody can run at reasonable cost. So the honest position is that the countable spine is a proxy for the thing that matters, and we would rather say so than dress the proxy up.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.The strongest alternative is a recording. Present the talk to an empty room, record it, watch it back at normal speed with a stopwatch. That produces your real per-slide times, exposes every place you had to look at the slide to remember what came next, and finds the filler words. It takes as long as the talk plus the talk again, and it is better than any audit of the file.
The presentation applications have both halves of this built in. One feature records real per-slide durations; another listens and reports pace, filler words and whether you are reading the slide. They are free and they are already installed.
Where a file like this earns its place is on the deck you are handing to somebody else, or on the deck you have already rehearsed and which still overran, or when the slot was just cut from thirty minutes to twenty and something has to go. Those are budgeting and documentation problems rather than delivery problems, and delivery practice does not solve them.
--- name: speaker-notes-and-timing description: Audits the speaker notes and the timing of a finished presentation. Covers the division of labour between what the slide shows, what the speaker says and what the notes hold, the six things that belong in a notes field, realistic per-slide timing arithmetic including why an unfamiliar chart costs several times a sentence slide, budgeting a talk backwards from the end so the ask is protected, a cut plan decided in advance for minus five and minus fifteen minutes, where to take questions, and what happens to a deck whose meaning lives in notes that get stripped when it is forwarded. This skill should be used after a deck is built and before it is rehearsed, handed over or shared. --- # Speaker notes and timing audit This skill reviews a deck that already exists. It generates nothing: no slides, no presentation file, no rewritten deck. It reads the slides, the notes and the booked slot, and reports on what the notes carry and what the talk will actually take. ## The contract Three channels carry a presentation, and each one carries a different thing. Confusing them is the single cause behind most bad talks. - **The slide carries what must be seen.** The chart, the photograph, the number, the thing words cannot do. If it works as a sentence, it should be a sentence you say, not a slide. - **The spoken word carries the argument.** The causal chain, the reason this follows from that, the so-what. Arguments live poorly on slides and well in speech. - **The notes carry what the speaker will forget.** Not the argument, which you know. The details that evaporate under pressure. The corollary is the rule that matters most: **notes that duplicate the slide are notes that will be read aloud.** Under pressure a speaker reads what is in front of them, so if the notes restate the bullets, the delivery becomes the bullets spoken slowly. That is the worst available outcome, because an audience reads a slide in about a fifth of the time it takes to say it, so they finish, wait, and then stop paying attention. You have converted a presentation into a slow audiobook of a document they have already read. A second corollary: **notes are not a script.** A full script is unreadable at a glance, cannot be resumed after an interruption, and produces flat delivery because written English and spoken English are different languages. The one exception is the opening sentence, which is worth having verbatim because the first fifteen seconds are the ones nerves take. ## What belongs in the notes, enumerated For each slide, in this order: 1. **The transition sentence into the next slide.** One sentence that gets you from here to there. Transitions are the first thing to fall apart under pressure, and a talk that lurches between slides reads as a list rather than an argument. This is the single highest-value line in any notes field and it is almost never written down. 2. **The number you will be asked for.** One level below the number on the slide. If the slide says revenue grew 40 percent, the note carries the absolute figures, the comparison period, and the split that somebody will ask about. Being unable to produce the next number down is what makes a confident slide look thin. 3. **The source and date of every figure on the slide.** Where it came from, when it was pulled, and who owns it. This is what turns a challenge into a two-second answer, and it is also what lets a colleague present the deck without inventing provenance. 4. **The thing not to say.** The unannounced product, the customer who has not agreed to be named, the number that is not public until Thursday. Writing it down as a prohibition is more reliable than remembering it, and it survives the deck being presented by somebody else. 5. **The question this slide always provokes, and the answer.** Every deck has two or three slides that reliably draw the same question. Record it and answer it in one line. If the question is asked three times out of four, consider answering it on the slide instead and reclaiming the interruption. 6. **The cut marker.** Whether this slide comes out at minus five minutes, at minus fifteen, or never. Written in the notes because that is where you will be looking. Two optional additions that repay their space: an **elapsed-time marker** (*you should be at six minutes here*), which turns a clock glance into a decision, and for the final slide, **the ask written out verbatim**, because the ask is the sentence people most often improvise and least often improve. ## Timing arithmetic, done properly **A working rate.** Plan at **130 words per minute** for technical content delivered to an audience taking notes, and **150** for a familiar talk to a friendly room. Conversational English sits roughly in that band. Two adjustments matter more than the base rate: people speed up under nerves, commonly by 10 to 20 percent, and **rehearsing in your head runs at roughly twice speaking speed**, which is why silent rehearsal timings are worthless and always optimistic. At 140 words per minute, a twenty minute talk is about 2,800 spoken words. That is a useful anchor when you are deciding whether the material fits at all. **Slides are not equal, and the one-minute-per-slide rule is wrong for the two types that dominate a real deck.** - A headline with one sentence: **20 to 40 seconds**. - A slide with three or four short points you actually discuss: **60 to 90 seconds**. - **A chart the audience has not seen before: 90 seconds to three minutes.** The mechanism is specific: the first ten to fifteen seconds after an unfamiliar chart appears go to reading it, and anything you say in that window is not heard. So a chart costs the reading time plus the explanation time, and you should either stay silent while they read or say only the headline finding, then explain. - A table anyone is expected to read: longer than a chart, and usually a sign the table belongs in an appendix. - A live demonstration: **budget double** what it takes when it works, because it will not. **Rehearsal underestimates live delivery, structurally.** A clean solo rehearsal does not contain the laptop failing to connect, the host's introduction, the room settling, the person arriving late, the question in the middle, the joke that lands and costs twenty seconds, or the moment you decide on the spot to explain something again. **Add 20 to 30 percent to a clean solo rehearsal to get a live estimate**, and more if the audience is senior, because senior audiences interrupt. ## The time budget, allocated backwards Start from the end of the slot and work back. For a thirty minute booking: - **3 minutes** lost to starting late and connecting the machine. This is not pessimism, it is the observed default. - **1 minute** for the host's introduction. - **8 minutes** for questions, if you want real questions rather than one polite one. - **2 minutes** of buffer, because the buffer is what stops a small overrun becoming a scramble. - **Which leaves about 16 minutes of speaking against a 30 minute booking.** Most people plan thirty minutes of content for a thirty minute slot, which is why most talks overrun. Halving your expectation feels wrong and is roughly correct. **Protect the ask first.** Reserve the final two minutes for what you want to happen next, and budget them before anything else, not after. The ask is the reason the meeting exists and it is always the thing that gets compressed, so it should be the one item in the budget that is not available for compression. **Cut from the middle.** The opening frames the talk and the ending carries the ask. The middle is evidence and detail, and the middle is where a talk can lose a third of its length while keeping its shape. A talk cut from the end becomes a description of a problem with no request attached, which is worse than a talk that made its case with two examples instead of five. ## The cut plan, prepared while calm Decide in advance which slides come out at **minus five minutes** and which come out at **minus fifteen**, and mark them in the notes. Two rules: - **Cut whole slides, not sentences.** Trimming sentences means speaking faster, which degrades every remaining slide and is visible to the audience as panic. Removing a slide is invisible. - **Decide when calm.** A cut chosen at minute twenty-two of a twenty-five minute slot is a cut chosen badly. A cut chosen at your desk is a design decision. The minus five list is usually two or three supporting examples and one nice-to-have chart. The minus fifteen list is a different talk: the two slides that make the argument, plus the ask. Write both down. Being handed a shortened slot is common and it is survivable if the shorter version already exists. ## Questions, and where they go **Taken throughout:** better engagement, the audience stays with you, and you find out immediately when something has not landed. The cost is that the timing becomes unpredictable and you can no longer guarantee reaching the ask, which is the thing you cannot afford to lose. **Taken at the end:** predictable and protects the ask. The cost is that the question which would have changed how you argued the whole thing arrives after you have finished arguing it, and the room may have been confused for fifteen minutes. **Decision rule.** - Small room, senior audience, decision being made in the meeting. → **Throughout.** They will interrupt regardless, and pretending otherwise loses you control of the frame rather than gaining it. - Large room, fixed slot, one-to-many, recorded. → **At the end**, stated up front so people hold their questions rather than wondering. - **You cannot tell.** → **Two designated points**: after the problem statement, and at the end. Say so in the first minute. This keeps the timing predictable and still catches the confusion early, which is where most of the value of interruption sits. Whichever you choose, the notes are what make a question answerable: the source and date of each figure, the number one level down, and the recurring question with its answer. Add one more note on any slide that has ever caused an argument: the objection, and the shortest honest response. ## The forwarded deck problem **Notes are frequently stripped or ignored when a deck is shared.** A PDF export defaults to slides without notes. Many web viewers do not display them. Some conversions drop them. And most importantly, people forward the PDF, not the source file, so the notes were never in the artefact that travelled. The consequences run both ways. - **Anything essential must be on the slide or in an appendix.** A deck whose meaning lives in the notes will be misread by everyone who was not in the room, and that is usually more people than were. Put the evidence, the caveat and the source on a slide after the final one, where it is out of the way during delivery and present in the file afterwards. - **Notes are not private.** They travel with the file, and anyone who receives the deck can read them. A note reading *do not mention the restructure* travels too, and so does a note about a named customer. Before sending, read every notes field with the assumption that it will be read by the least sympathetic person on the distribution list. **Which deck you are building changes what the notes have to hold**, and the read deck against the presented deck is set out in full in the deck narrative audit skill. The consequence for the notes field is the only part that belongs here: when you cannot tell which one you are building, which is the normal case, build the presented version and add an appendix carrying what the notes would have said, rather than compromising into a dense middle that fails at both jobs. ## Rehearsal as a procedure Rehearsal is usually an intention. Make it a procedure with an output. **Pass 1: out loud, alone, timed per slide.** Out loud is non-negotiable, since silent rehearsal runs at roughly double speed. Change nothing during the pass. Afterwards, look at the per-slide times and find the three slides that overran their budget. **Cut those slides. Do not resolve to speak faster.** **Pass 2: on the real display, standing, with the clicker.** This is where you discover the chart that is illegible at projection size, the animation that needs a click you had forgotten, the video that does not autoplay, and the colour that vanishes on the room's screen. Time it again; the number will be different. **Pass 3: with one person who is instructed to interrupt.** Record every question they ask. Each one becomes either a notes line, an appendix slide, or a change to the slide itself. **What to record after each pass:** total time, per-slide time, the three worst overruns, every question asked, and every point where you had to look at the slide to remember what came next. That last one is the most diagnostic thing on the list, because it marks exactly where a transition sentence is missing. **What to change:** one class of thing per pass, so you can tell what helped. Pass 1 changes the length. Pass 2 changes the slides. Pass 3 changes the notes. The application will do the mechanical half for you. One feature records real per-slide durations, and current versions include a coaching feature that reports pace, filler words and whether you are reading off the slide. Both exist on Windows and Mac and cost nothing. ## The audit Run these six checks against the finished deck. 1. **Read the notes without the slides.** The argument should survive. If the notes are a list of nouns and the argument is not recoverable, nothing has been recorded and the deck cannot be handed to anybody. 2. **Read the slide headlines without the notes.** In sequence they should read as a summary of the talk. If they read as a list of topics, the structure problem is upstream of the notes. 3. **Every figure has a source and a date** in the notes of the slide it appears on. 4. **The final slide's notes contain the ask, written out.** 5. **The cut markers exist**, for both minus five and minus fifteen. 6. **Nothing in any notes field would embarrass you if the file were forwarded**, which it will be. ## Worked example A slide in a quarterly review. The slide shows a line chart with the headline *Support ticket volume doubled in Q2*. All figures in the notes below are invented for this example. **The notes field before:** > Ticket volume doubled in Q2. Chart shows the trend over six months. Talk about why this happened. Mention hiring. That is four lines and it contains nothing the slide does not. It will be read aloud, in a room where everyone has already read the headline. *Talk about why this happened* is a note to do the hard part later, and *mention hiring* will be recalled as a word rather than a point. **The notes field after:** > **Open:** they can see it doubled. The interesting number is that first-response time only moved by eleven minutes. > > **Figures:** 2,140 tickets in Q2 against 1,050 in Q1. Pulled from the support platform on the 3rd, owned by the support lead. Q1 excludes the migration week. > > **One level down:** the increase is 1,090 tickets, and 60 percent of it, 654 tickets, is two customers who went live in April. The other 436 is underlying growth on a base of 1,050, which is 41.5 percent. > > **Always asked:** are we understaffed? Answer: not yet, first-response time held. The point at which it stops holding is roughly 2,800 tickets in a quarter, which the current trend reaches in Q3. > > **Do not say:** the enterprise contract is not signed, so do not name the customer. > > **Cut at minus five:** yes, fold into one sentence on the next slide. > > **Transition:** so the volume is not the problem yet, and the date it becomes one is the thing I want a decision on. **Verdict on the change.** The second version contains nothing that is on the slide. It carries the number one level down, the source with its date and its caveat, the recurring question with a specific answer, a prohibition, a cut decision, and a transition sentence that turns a status update into a request. It is also longer to read and shorter to say, which is the correct direction. The first version would have been read aloud; the second cannot be, because none of it is prose. ## Failure modes **Notes that restate the slide.** They get read aloud, the audience finishes the slide before you do, and the talk becomes a slow recital of a document already read. **Notes as a full script.** Unreadable at a glance, impossible to resume after an interruption, and delivered flat because written English is not spoken English. **Timing the talk in your head.** Mental rehearsal runs at roughly double speaking speed, so the estimate is always about half the truth and always in the dangerous direction. **One minute per slide.** Correct for one slide type and badly wrong for both types that dominate a real deck, so the estimate is wrong by more the more evidence you are showing. **Cutting from the end.** The ask goes, and a talk that made its case and asked for nothing is a talk that changed nothing. **No cut plan.** The cutting decision arrives at minute twenty-two, under pressure, and resolves into speaking faster, which degrades every remaining slide. **Questions throughout with no reserve.** You reach slide nine of fourteen with two minutes left, and slides ten to fourteen were the argument. **Confidential material in the notes.** The file is forwarded, the notes travel with it, and the sentence you wrote to remind yourself not to say something out loud is now in writing. **Essential content only in the notes.** Everyone who reads the forwarded deck sees a set of assertions with no evidence and no caveats, and concludes the work is thinner than it is. **Rehearsing a different talk from the one you will give.** Skipping the demo, skipping the video, skipping the introduction, and then being surprised by four minutes that were always going to happen. ## What this skill does not do - It cannot hear you. Pace, filler words, whether you are reading the slide and whether the room is with you are delivery questions, and a recording or another person answers them better than any file. - Its timing numbers are planning estimates, not measurements of you. After one recorded rehearsal, replace them with your own per-slide times and never look at the estimates again. - It does not know your audience, and whether a chart costs forty seconds or three minutes depends entirely on how familiar they already are with it. - It does not write or fix the talk. Whether the argument holds, whether the evidence supports it and whether the ask is the right ask are upstream questions with different answers. - It does not design slides. Density, chart honesty, layout and legibility are separate problems, covered separately. - It cannot make the slot longer, and there is no arrangement of notes that fits thirty minutes of material into sixteen.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.