Where your deck stands, and tonight's lever
Sessions 2 and 3 took your deck from a topic label with six bullets to a sentence headline with one visual, and then stripped everything off-message. On the golden set that is 73 percent. Tonight you claim the fourth rung, and it is the last one available before session 6.
★ What you walk out with tonight. One chart in your real deck rebuilt so it argues instead of decorating, a cue on the number that decides the meeting, and a traceability audit that names every figure in the deck, its source, its date and its owner. Plus the exact sentence from Microsoft's own documentation that explains how a number drifts between your source file and your slide.
LiveThe ladder, and which rung is tonight3 min▶
The simulator you met in session 1 adds one lever at a time. The rungs are fixed, they are read out of the live model, and tonight owns exactly one of them.
| Rung | Lever | Understood in 30s | Session |
|---|---|---|---|
| 0 | a normal corporate slide | 27% | 1 |
| 1 | sentence headline that states the message | 48% | 3 |
| 2 | one visual as the evidence | 64% | 3 |
| 3 | strip anything off-message | 73% | 3 |
| 4 | cue the one number that matters | 82% | 4 · tonight |
| 5 | text clears WCAG 4.5:1 | 91% | 6 |
82 percent is the ceiling until session 6. No amount of further chart craft gets past it, because the last nine points are an accessibility fix. Knowing that now saves you an evening of polishing.
Self-studyWhat is real here and what is simulated2 min▶
Said plainly, because it matters more on a data session than anywhere else in this course. The comprehension percentages are a teaching simulation. The weights model Garner and Alley 2013 and Mayer's principles. They do not measure your room.
What is real: the contrast ratio is WCAG 2.2 arithmetic on the colours the slide is actually painted with, floored and never rounded up, and the chart in the simulator is real SVG built from the numbers it is given.
Holding that line is the whole subject of Part 3. If I dressed a model up as a measurement on a page about evidence integrity, you would be right to stop reading.
Charts that argue
7 min liveHere is the test, and it is the session 1 test wearing different clothes. Say the claim your chart makes, out loud, in one sentence. If you cannot, the chart is decoration and it is costing you the thirty seconds you needed for the ask.
LiveOne chart, one claim, matching the headline4 min▶
Session 3 gave every slide a sentence headline. A chart slide has one extra rule on top of that: the chart must prove the headline, and nothing else. If your headline says EMEA is the gap and your chart shows five regions across three quarters with a margin line, the chart is answering four questions and the room has to work out which one you meant.
- Delete every series that is not part of the claim. Not grey it out. Delete it. Mayer's coherence principle was worth nine points in session 3 for removing things, and it applies inside the chart as much as around it.
- Sort by the thing you are arguing about. Alphabetical order is a filing choice. Sorted by size is an argument.
- Say the number in words next to the chart. The room should not have to read a bar against an axis to hear your point.
A pricing deck goes in with a chart of eleven competitor price points. The headline says "we are mid-market". The client spends four minutes finding their own product on the chart, decides the axis is unfair, and the pricing decision moves to next month.
Self-studyWhy the second y-axis and the 3D pie have to go5 min▶
These are not taste rules, they are comprehension costs, and each one has a specific mechanism.
| The thing | What it does to the room | What to do instead |
|---|---|---|
| Second y-axis | Two scales invite a comparison you did not intend, and the crossover point can be moved by choosing the scales | Two charts side by side, or index both series to 100 and use one axis |
| 3D pie | Perspective makes the front slices look larger, so the chart misreports before anyone misreads it | One stacked bar, sorted, with the share you care about labelled in words |
| Rainbow legend | Every colour reads as a meaning, so nothing is emphasised and the room hunts the legend | One accent for the series in the claim, one neutral grey for everything else |
| Truncated axis | A bar chart starting at 90 turns a 2 percent move into a cliff, and someone always notices | Start bars at zero. For small real moves, use a line chart or state the delta in words |
One honest caveat: chart craft is its own discipline and this course only covers the part that decides whether your slide lands. The source map marks this session as partial coverage for exactly that reason.
A claim you can write in one sentence is a chart you can build in ten minutes. A claim you cannot write is usually four charts wearing one costume, and no tool will untangle that for you.
Cue the one number
6 min liveMayer's signaling principle says people learn better when cues highlight the organisation of the essential material. On a chart slide that sentence has a very concrete meaning: one number decides this meeting, and right now it is set in the same weight as the four numbers that do not. This is the nine-point lever, and it is the difference between a chart people look at and a chart people act on.
LiveSignaling, made mechanical4 min▶
Three moves, in this order, on every chart slide. They take about four minutes each and they are the whole rung.
Darken the deciding element, neutralise the rest. One accent on the bar, line or slice that carries the claim. Everything else goes to a single neutral grey. Not five greys. One.
Set the number larger, near the thing it describes. The deciding figure gets its own size, sitting next to the element rather than in a legend or a data table at the bottom.
Put the ask in words beside it. "78 percent of the shortfall sits in EMEA" is the finding. "Approve the EMEA rescue plan today" is the ask. The room needs both on the slide, because only one of them is a decision.
The cue is a choice about what matters, which is why no tool can make it for you. A generator does not know which number decides your meeting, and every generator will happily give all five bars a different bright colour.
Self-studyCueing without lying, and the labels that keep you honest5 min▶
There is a real line between emphasis and distortion, and it is worth being able to state it: a cue changes what the eye goes to first, it never changes what the data says. Darkening one bar is a cue. Cropping the axis so that bar looks four times taller is not.
| Emphasis | Distortion |
|---|---|
| One accent colour on the deciding bar | An axis that starts at 90 instead of zero |
| The deciding number set larger and nearer | Dropping the two data points that disagree |
| The claim stated in words on the slide | A percentage with no denominator anywhere |
| Everything off-claim in one neutral grey | Two y-axes chosen so the lines cross where you want |
Four labels keep a chart honest, and they cost nothing: what the unit is, what the period is, whether the figure is actual or forecast, and what the denominator is on any percentage. A chart missing the denominator is the single most common thing a client picks apart in the room.
Where fabricated figures actually come from
7 min liveNobody sets out to put a wrong number in a client deck. The number arrives quietly, and the mechanism is boring rather than sinister. Tonight you get the precise version, in the vendors' own words, because a vague fear of "AI hallucination" is not a control and it will not help you at the moment a client reads a figure back to you.
LiveWhat the vendors say, in their own words5 min▶
Keep these three sentences. They are the ones you quote when someone asks why you are re-checking a figure the tool produced, and they come from the vendors themselves rather than from a critic.
Read the Microsoft one twice. That second clause is the mechanism behind a number that drifted between your source document and your slide. The tool was never promising to carry the figure across unchanged. It was promising to summarize, transform and reason over it, which are three operations that can each move a decimal point.
And note what Anthropic's line implies about workflow rather than quality: the human review step is not an optional polish pass, it is named in the documentation as the thing that makes the output shippable.
Self-studyThe traceability discipline, in one rule6 min▶
One rule, and it is short enough to hold in your head during a build: every figure on a slide points at a cell or a line in a source you can open, and if it does not, it comes off the slide.
"Comes off the slide" has two honest exits. Either you find the source and cite it, or you cut the figure and say the qualitative version out loud. Both are fine. What is not fine is a number sitting on a slide with nobody able to say where it came from, because in a client deck that number is now a professional claim.
| On the slide | What the audit needs | Verdict |
|---|---|---|
| "Revenue down 8 percent" | Finance pack, tab Q3, cell B14, dated, owner named | Cite it in the notes |
| "Roughly a third of users churned" | No file, remembered from a call | Cut it, or ask the owner today |
| "Market growing 14 percent" | A model summary of an analyst PDF you have not opened | Open the PDF, or cut the figure |
| "Costs up 8.2 percent" | Same source as the slide before, which said 8 | Pick one, fix the other |
That last row is the one that actually happens. Two slides, one source, two roundings, and a client who spots it. The fix is a single pass, not a policy.
A quarterly readout carries an 8 percent decline on slide 4 and an 8.2 percent decline on slide 9. Same source, two roundings, four weeks apart. The client asks which is right, and the next twenty minutes are about the deck's credibility rather than about the plan. Nobody fabricated anything. One figure just lost its address.
Rebuild one chart, then audit every number in the deck
★ 22 min · everyone buildsTwo things tonight, and the second one is the one that will still be paying you back in a year. First rebuild one real chart from your deck so it argues. Then run a traceability audit across the whole thing: source, date, owner, for every figure on every slide.
List the figures. Walk the deck slide by slide and write every number in one column. Percentages, currency, counts, dates, headcount. A twelve-slide deck usually holds between twenty and forty.
Find each source. Next to each figure: the file, the tab, the cell or the line. Not "the finance pack". The cell. If finding it takes more than a minute, that is a signal, not an inconvenience.
Mark the orphans. Any figure with no openable source gets flagged. Expect a few, and expect at least one of them to be a number you have said out loud in a meeting already.
Cut or cite. Every orphan takes one of two exits. Cite it in the speaker notes with source, date and owner, or cut it and say the qualitative version. There is no third exit.
Pick the chart that matters most. Not the prettiest one. The one on the slide where the decision happens. Write its claim in one sentence before you touch the chart itself.
Name the relationship, then fix the form. Comparison, trend, composition or distribution. If the current chart form does not match, change the form. This is usually the moment a pie becomes a sorted bar.
Delete every series that is not in the claim. Out loud, for each one: does the claim need this? If not, it goes. Do not grey it, do not shrink it, delete it.
Cue the deciding number. One accent colour on one element, everything else in one neutral grey, the number set larger, the ask in words beside it. That is the nine points, and the golden set moves 73 to 82.
Add the four honest labels. Unit, period, actual or forecast, and the denominator on every percentage. Thirty seconds each, and they close the most common question from the floor.
Run the audit on the whole deck. The four moves in the pipeline above. Put source, date and owner in the speaker notes, not on the slide. Session 6 turns those notes into what you actually say.
If you are presenting this week. Skip the chart rebuild and do the audit. A deck with plain charts and traceable numbers survives a hostile room. A deck with beautiful charts and one orphan figure does not.
Before session 5
25 min- Write the claim for every chart in the deck. One sentence each. Any chart you cannot write a claim for is either merged into another chart or cut.
- Cue one number per chart slide. One accent, one neutral grey, the figure larger, the ask in words. Nothing else changes.
- Finish the traceability audit. Every figure in the deck: source, tab, cell, date, owner. Orphans cited in the notes or cut. Count how many orphans you found and remember the number.
- Recompute three totals from the source. Not re-read. Recomputed. If any of the three disagrees with the slide, that is tonight's most valuable finding.
- Decide which tool you are building in. Session 5 is the real build, and it goes faster if you already know whether you have a Copilot Designer licence, a Claude Pro seat, or neither. Check before class.
Three questions
Clear all three and session 4 gets stamped in your passport.
1 · Cueing the one number is which rung of the ladder, and what is it worth?
Signaling is rung 4 and takes the golden set from 73 to 82. 82 is the ceiling until session 6, where the WCAG contrast lever takes it to 91. Rung 1 is the sentence headline, and it is the biggest single lever in the model.
2 · Microsoft says Copilot is optimized for which of these?
Microsoft's own wording, and it is the mechanism behind a drifted number. The same documentation says Copilot "cannot understand meaning or evaluate accuracy", so the accuracy check was never on offer in the first place.
3 · A figure on slide 7 has no source you can open. What happens to it?
Two exits only. Softening the wording keeps the professional claim while hiding that you cannot back it, which is the worst of the three options in a room that asks follow-up questions.
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. Certificates, videos and assessments stay with the official providers.
Session 4 cheat sheet · pin this
Your deck now argues and every figure has an address, and you have been building it by hand. Session 5 is the real build in whatever tool you actually have a licence for, and it opens with the trap that wastes the most time: the feature you are trying to use may simply not be switched on for your account, and nothing on screen will tell you why.