Precision and Type: Alignment, Distribution, Grids, Guides, Typography

Presentation track, branch pr-mechanics (Slide Mechanics) · Node PR1.02 · ~10 focused hours · Mastery gate ≥ 85%

A senior partner looks at a slide for about a second and says it needs tidying up. Asked what specifically, she cannot say. The analyst who built it spends forty minutes changing the colours, which were never the problem, and hands back a slide that gets the same verdict.

Without knowing it, she was looking at eleven different left edges. The slide holds twelve elements. Their left edges are at 0.52, 0.55, 0.58, 0.60, 0.62, 2.95, 5.48, 6.72, 8.05, 12.10 and 12.55 inches. That's eleven distinct values where the design intends six. The top edges are just as scattered. There are eleven of those too, and four cards that were meant to be a row stand at three different heights, 1.88, 1.90 and 1.92. Of the 48 edges on that slide, 45 don't touch a guide at all.

None of those errors is visible on its own. A card 0.02 inches out of line is 1.44 points, which is smaller than the space between two lines of caption text. The eye picks up the aggregate instead, and it picks it up fast. Human vision resolves the offset between two collinear edges at a much finer scale than it resolves anything else in the visual field. That sensitivity is called vernier acuity, and it's the reason a misaligned slide reads as sloppy before anyone has processed a single word.

Then the reader draws a conclusion you didn't intend. Somebody who didn't check whether the boxes lined up probably didn't check whether the numbers tied either. It isn't a fair inference. It's close to universal, though, and it's cheap to prevent. Rebuild that same slide, with the same twelve elements carrying the same content, and every one of its 48 edges lands on a guide. Once the grid exists, you build the slide by typing in positions. Nobody drags anything.

You're learning two families of decision here, and they solve different problems. Geometry decides where things go: the frame, the columns, the gaps, and the operations that put an object on a line and keep it there. Typography decides how rank is expressed. That means how many sizes, how far apart, how long a line runs before it stops being readable, and how small type can get before someone twelve metres away can't read it at all. Both run on arithmetic you can do in advance. That's what makes them teachable in ten hours, when taste isn't.

As you work through the sections below, test this claim. Almost every "make it look more professional" instruction comes down to one of six numbers: the margin, the column width, the gutter, the number of type sizes, the ratio between them, and the line length. Fix those six and the slide is competent. Whether the message is right is a separate question, and other branches own it. No amount of precision will rescue a slide whose message is wrong. Precision only guarantees that a right message survives the trip to the reader's eye.

Learning objectives

You can:

1. Compute a column grid from a canvas size, a margin and a gutter, give the exact column width in inches, points and English Metric Units, and state which block spans land on column lines and which do not.
2. Score a slide's alignment by counting edges that miss the grid inside a stated tolerance, express it as a rule score, and predict how much the score moves when a specific element is fixed.
3. Apply the six placement operations in the correct order (span, size, align, distribute, cross-align, lock), and explain what breaks when distribute runs before sizes are equalised.
4. Choose a margin and defend it against both the extra area a tighter margin buys and the amount a projector's overscan removes.
5. Identify the three cases where optical alignment beats mathematical alignment, compute the correction each one needs, and state the size of that correction in points at a given type size.
6. Build a four-level type scale from a base size and a ratio, check every adjacent pair against a minimum ratio, and reject a fifth level with the arithmetic rather than by feel.
7. Rank the four devices that carry hierarchy (space, size, weight, colour), quantify the contrast each one supplies, and diagnose a slide that is using colour to do a job the type scale should have done.
8. Size type for a room: given a screen width, a viewing distance and a point size, compute the angular size of a capital letter in arcminutes and say whether it clears the working threshold.
9. Set a line length and a leading that keep a text block inside the readable measure, and compute the block's height so it fits the live area you have.
10. Run a five-second glance test on your own slide, score it against your stated message, and use the result to decide whether the fix is typographic or editorial.

Prerequisites & connections

Builds on. PR1.01 owns the question of where a formatting decision is stored: the master, the layout, the placeholder, the theme font and the theme colour. Everything below assumes you can put a decision on a master and have it inherited, because a grid enforced by hand on 40 slides isn't enforced at all. PR0.02 owns the tool itself: views, sections, file hygiene and the twenty shortcuts, including the ones that show and hide guides. PR0.01 owns what comes before any of this, which is takeaway, audience, decision, medium, artefact.

Feeds forward. PR1.03 takes the geometry taught here and applies it to things that move: shapes, images, icons, groups, animation and the eight-item ship checklist. PR2.02 takes the colour role sketched here in one section and gives it the full data-visualisation treatment: data ink, pre-attentive attributes, palettes, direct labelling and accessibility. PR2.03 turns the glance test from a typographic check into a slide-level standard. PR3.01 builds diagrams on the same column grid, so a process flow and a chart share their edges. PR5.01 to PR5.03 assume a competent grid and a working type scale as the floor under their archetypes, and the capstone in PR7.01 grades against it.

Five neighbours own material that touches this one, and the boundaries are deliberate. PR2.02 owns colour and decluttering as data-visualisation craft, including the accent formula, palette type and the accessibility check. The one colour section below says only enough to keep the geometry sections sound, then hands the subject over. PR2.01 owns chart form choice, so nothing here argues about bar against line. PR1.01 owns masters, layouts and templates, so the grid below is specified as numbers and storing those numbers belongs there. PR1.03 owns shape mechanics, grouping and the selection pane, so the sections below place rectangles and say nothing about how to draw them. M10.03 owns the written analyst deliverables. Line length and type scale apply there too, but at a reading distance of arm's length. On a slide it's nine metres, and that changes every number.

Core teaching content

4. The frame: one margin, obeyed

A margin is a single number that every element on every slide respects. Its value matters much less than its consistency. Any of these is defensible:

Tightening from 0.50 to 0.30 inches buys 7.6933 square inches, which is 9.60% more live area. That's a real gain, and it's usually not worth taking. The reason has nothing to do with aesthetics.

Projectors and conference-room displays crop. Overscan of about 2.5% per edge is common on older equipment and on any signal path that includes a scaler. Take 2.5% of a 13.3333 inch width and you lose 0.3333 inches horizontally, and 0.1875 inches go vertically. A 0.5 inch margin survives that with 0.1667 inches of clearance horizontally and 0.3125 inches vertically. A 0.25 inch margin doesn't. It's 0.0833 inches short horizontally, so the leftmost characters of a full-width title go off the edge of the screen. That happens in the room, in front of the audience, after the deck looked perfect on the laptop. It's a horrible moment, and a completely avoidable one. Overscan behaviour varies by device and by cable path, so treat 2.5% as a planning assumption and test the actual room where you can.

So half an inch is the working default here. Put it on the master as the four outermost guides, and stop making the decision again on every slide. One rule matters more than the number. Nothing crosses the margin. Not a logo, not a page number, not a chart that's 0.1 inches too wide, not a full-bleed image unless full-bleed is a deliberate layout in the template with its own rules.

10. Colour as a role, briefly

Colour gets one section here and a full treatment later. It's worth stating the boundary, so you know what's missing.

For geometry and type, colour has two jobs on a slide. It marks the thing the reader should look at, and it recedes for everything else. Use one accent on the single element carrying the message, with everything else in a neutral grey. That formula is enough to keep the sections above sound. It isn't enough to build a chart palette, choose between sequential and diverging schemes, or handle a colour-vision deficiency, and all of that belongs to the chart-craft branch.

Two constraints do apply straight away, because they interact with type. Contrast between text and its background needs to clear 4.5 to 1 for normal-size text and 3 to 1 for large text. The Web Content Accessibility Guidelines define large as 18 point and above, or 14 point bold and above (WCAG 2.2, checked September 2026; confirm, since the guideline version advances and the thresholds can move). Grey body text on a white slide fails that ratio at values people reach for casually. A mid-grey at around 60% lightness won't pass. Reach for a darker grey. Making the text bigger is the wrong fix.

The second constraint is the diagnostic from section 7, restated as a rule. If removing colour from a slide destroys its hierarchy, colour was doing the type scale's job. No amount of palette work will fix a slide whose ranks exist only in hue.

11. White space, and what it is actually measuring

White space is usually described as breathing room. That's true and useless, because it gives you no way to tell whether a slide has enough. Two measurements help more.

The first is the live area share. With a 0.5 inch margin, the live area is 80.17% of the canvas and the margin band is the other 19.83%. That 19.83% isn't available for content under any circumstance, so white space stops being a preference and becomes a boundary.

The second is more surprising. Add up the bounding-box area of every element on the opening slide before the rebuild and it comes to 55.1564 square inches, or 55.16% of the canvas. Do the same on the rebuilt version and you get 61.0500 square inches, or 61.05%. The rebuilt slide covers more of the canvas, and it reads as far less crowded. That surprises most people the first time.

That result rules out the naive theory that white space is the amount of empty canvas. White space depends on whether the empty canvas is organised: whether the gaps take a small number of distinct values, whether elements that belong together are closer than elements that don't, and whether the outer frame is intact. The before slide has eleven distinct left edges and card gaps ranging over 8.64 points, so its emptiness is spread around as noise. The after slide has six distinct left edges, five distinct top edges and three identical 0.20 inch gaps. Its emptiness has a structure.

So the instinct to make space by shrinking everything is wrong twice. It reduces the ink without organising the gaps, so the slide looks small instead of calm. And it pushes type below the size the room needs. You buy space by removing elements and grouping the survivors. Scaling them down doesn't buy any.

13. Three traps built into the defaults

Each of the three most common mechanical failures in a deck comes from the tool's default behaviour. Nobody chose any of them.

Centred text is the first. It has a ragged left edge by construction, so every line begins at a different horizontal position, and the alignment everything above depends on is gone. A centred paragraph of four lines adds four different left edges to a slide that was trying to have six. Centring is right for a single line, for a title over a symmetrical object, and for a quote slide where the block is the only thing present. It's wrong for bullets, for card contents, for table cells other than short labels, and for any block of more than two lines.

The eight-font deck is the second, and nobody chooses eight. It happens like this. A theme font is set once and then overridden locally, slide by slide, each time for a reason that made sense at the time, until the deck holds eight sizes across three weights and 24 distinct type roles. The count that matters is distinct (family, size, weight) triples. Do actually count them on a deck you inherit, because the number is always higher than the owner's estimate. The repair is the reset control and the theme fonts, and you'll find both on the master.

Shrink-to-fit is the third, and it's the most damaging default in the product, because it silently turns a content problem into a typographic one. Overfill a 24 point placeholder and the text steps down through roughly 24, 22, 20, 18 and 16 point until it fits. Now two slides that both use the "body" role carry 24 point and 16 point type. That's a ratio of 0.6667, and the smaller text has 44.44% of the apparent area of the larger. At a 9 times projection, that's 0.7 inches of lost cap height on the screen. Worse, the deck's type scale is now undefined. A reader can't infer rank from size, because paragraph length is setting the size. If you change one setting after reading this, make it this one.

The fix in every case is upstream. Turn autofit off in the master's placeholders, so overfilling produces visible overflow. That overflow is a clear signal that the slide has too many words. Then cut the words, split the slide, or move the material to the notes, and let the type scale stay a scale.

{
  "skill": "Auditing a slide's geometry and type scale against a stated grid before a human review",
  "use": "After you have written down your grid (canvas, margin, gutter, column count) and your four type sizes, hand an assistant the element list from your slide as a table of positions and sizes and ask it to compute which edges miss the grid, what the alignment score is, and which type sizes are off the scale. Use it as a fast arithmetic pass on your own specification; never let it choose the grid or the scale.",
  "tools": ["Chat assistants (Claude, ChatGPT, Gemini)", "A spreadsheet holding the element table", "python-pptx or a similar file-reading library for extracting real positions from a deck"],
  "prompt": "My grid is a 13.333 x 7.5 in canvas, 0.5 in margins, 12 columns, 0.2 in gutters, and a vertical rhythm of 0.25 in from the top margin. My type scale is 14 / 18 / 24 / 32 pt. Here is my element table with name, x, y, width, height in inches and point size: <TABLE>. Do four things. 1) List every edge that is more than 0.02 in from a column line or a rhythm line, naming the element and the edge. 2) Compute the alignment score as 1 minus misses over four times the element count. 3) List every point size that is not on my scale. 4) List the distinct left-edge values and the distinct top-edge values. Show the arithmetic for the score. Do not suggest a different grid or a different scale.",
  "verify": "Recompute the column line positions yourself from the margin, gutter and column count before trusting a single flagged edge, because an assistant that mis-derives the column width will flag correct elements and pass wrong ones, and both errors look identical in fluent output. Check two of the flagged edges by hand against the position dialog in the file itself. Then check the count the other way: the number of distinct left edges it reports must equal what you get by sorting your own table, and if the two disagree the assistant has rounded somewhere you did not ask it to. Any figure that reaches a slide, including an alignment score quoted in a rubric self-assessment, is yours to reproduce.",
  "diy": "You must be able to derive a column width from a canvas, a margin, a gutter and a column count with a calculator, and to compute the angular size of a capital at a stated distance, because both decisions get made in rooms where nobody is going to open a chat window. The gate asks for the arithmetic rather than the answer for that reason: an analyst who can only recognise a bad grid cannot specify a good one, and specifying is the job the mechanics branch exists to teach.",
  "market": "IN"
}

Common mistakes & how experts think differently

1. Aligning to the slide when you meant to align to the selection. The two commands are side by side, and their results differ by the width of the margin. You can spot it when a group has jumped to the canvas edge and now hangs outside the frame. An expert checks the align-to setting before every operation. Or they fix the anchor object first and align everything else to the selection, which makes the setting irrelevant.

1. Distributing before equalising size. Distribute produces equal gaps between whatever it's given, so unequal boxes get equal gaps and the width difference reads as an emphasis you never intended. The four cards in the worked example carry a 0.07 inch width range that survives any amount of distributing. The order in section 3 exists to prevent this. People get it wrong because distribute is the operation that visibly improves the slide, so it gets run first.

1. Nudging by eye and calling it optical alignment. Three cases justify an optical correction, and each has a size you can compute. Everything else is drift. An expert can name the case and state the correction in points before touching the object. A slide with four unexplained nudges has a grid that the next person to open the file can't maintain.

1. Using colour to build a hierarchy the type scale failed to build. Reach for a second colour to make a sub-head look like a sub-head, and you've diagnosed a scale with too many levels or too little contrast between them. Test it by converting the slide to greyscale. If the ranks vanish, the repair belongs in the type scale, and doing it there fixes the printed handout and the colour-blind reader at the same time.

1. Setting type size for the screen you're building on. An 18 point capital subtends 24.44 arcminutes on a laptop at arm's length and 15.04 arcminutes at the back of a 30 foot room, a factor of 1.62. Everything is legible while you build. An expert asks how big the room is before choosing the body size, and treats 45 foot rooms as a different medium altogether.

1. Letting the placeholder decide the line length. A full-width text block at 18 point runs 98.7 characters, a third beyond the readable maximum. The failure shows up as re-reading, and nothing looks wrong on the slide. The expert habit is to fix the body-text span once, at eight columns on this grid, and to treat a text block wider than that as a layout error.

1. Treating autofit as a convenience. Shrink-to-fit is how a deck with a type scale becomes a deck without one, silently, one overfilled placeholder at a time. Experts switch it off so that too much text produces visible overflow. Overflow is information. A font that shrinks without telling you isn't.

1. Checking alignment at fit-to-window zoom. At the zoom level where the whole slide is visible, a 0.02 inch break is a fraction of a screen pixel, and you can't see it. Nobody catches misalignment by looking harder. You catch it by running an alignment check, by using guides that objects snap to, or by counting distinct edge values. Look harder at 100% zoom and you'll find the breaks that were already obvious and miss the ones that matter in aggregate.

1. Retyping a total in a formatted table. Hand-typed totals turn up in tables somebody has been styling. The four regional figures sum to 6,300.0, and the printed total says 6,300.0 because it's the sum. An expert treats every printed aggregate as a computed cell and re-derives it after any edit to the rows. A table that doesn't tie destroys the credibility of the slide it's on, and of the two slides after it.

1. Skipping the glance test because it needs a second person. The check that predicts whether a slide will work takes two minutes and one colleague. The reason it gets skipped is social, since the time cost is tiny. Experts trade slides with someone for five seconds each. That costs four minutes for a pair, and it catches the failure no machine rule can see: a technically immaculate slide whose message nobody can state after looking at it.

Worked examples

Every figure below comes from one computation, and the Verify block at the end restates each one at full precision. All companies are synthetic composites. Positions are in inches on the 13.333 by 7.5 inch canvas, and the grid throughout is twelve columns with 0.5 inch margins and 0.2 inch gutters, which gives a column width of 0.844444 inches. Keep a calculator handy, because you'll want to check a few of these yourself.

Worked example 1: Twelve elements, forty-eight edges (India, ₹)

The setting. Meghdoot Logistics (synthetic) runs contract logistics across four regions. FY25 revenue is ₹6,300.0 crore against ₹5,688.0 crore in FY24, so the group grew 10.7595%. By region, FY25 against FY24 in ₹ crore: West 2,417.0 against 2,071.0, South 1,905.5 against 1,702.0, North 1,284.5 against 1,210.0, East 693.0 against 705.0. The slide has one message to carry. West produced 56.54% of the group's ₹612.0 crore increase, and East is the only region that shrank.

The wrong version. One slide, twelve elements, all placed by dragging. The title box starts at x = 0.62 and the kicker line beneath it at 0.55. Four KPI cards start at 0.52, 2.95, 5.48 and 8.05, with widths of 2.30, 2.28, 2.35 and 2.30 and tops of 1.90, 1.88, 1.92 and 1.90. The chart starts at 0.58, the callout box at 8.05, the source note at 0.60, the logo at 12.10 and the page number at 12.55, and a draft stamp is at 6.72. The title reads "FY25 Regional Performance".

Audit it edge by edge against the grid, at a tolerance of 0.02 inches:

Three of 48 edges land, and all three by accident. Card 3's right edge is 0.0189 inches from column 8's left line. The callout's top happens to be at 3.50, which is a rhythm line. And the draft stamp's right edge is 0.0144 inches from column 8's right line. The score is 1 − 45/48 = 0.0625.

Two more counts describe the damage better than the score does. The slide has eleven distinct left-edge values and eleven distinct top-edge values. The logo and the page number both run to 13.00 and cross the 12.8333 right margin, while the callout box stops at 12.65 and stays inside it. The top and bottom margins fare worse. The title's top edge at 0.35 and the logo's at 0.30 are above the 0.50 line, and the source note, the page number and the draft stamp all end past 7.00. So five of the twelve elements are outside the frame.

The corrected version. Nothing is deleted and nothing is added. Every element gets a span and a rhythm line:

All 48 edges land, so the alignment score is 1.0. Distinct left edges fall from eleven to six and distinct top edges from eleven to five, and nothing crosses the margin any more.

The title changes from "FY25 Regional Performance" to "West added ₹346 crore of the group's ₹612 crore; East is the only region that shrank". That moves the title-is-the-message rule from 0.25 to 0.95. The type scale drops from eight sizes to four. The chart form doesn't change. A sorted bar across four regions was already the right form for a comparison message, and the chart-choice rule scores 1.0 on both versions.

The two Linter passes, with the seven rules scored:

The before slide fails the 0.80 gate. The after slide clears it, and would clear the 0.90 crown gate too. The composite moved 0.5332, a relative improvement of 135.85%, and the machine mean alone moved 0.5554.

Read it. The score measures whether the slide's construction was deliberate. That's a precondition for quality and no substitute for it. Notice that chart choice scored 1.0 before the rebuild. The author knew which chart to draw and didn't know where to put it, and that's the profile the mechanics branch exists to fix. Notice also what would break the result. Suppose the four KPI cards were cut to three and the layout kept four columns of content. The 2.9333 inch card width would be wrong and the whole row would need respanning, because the grid is a system and not a set of remembered coordinates.

Worked example 2: Two hundred and forty-seven words (US, $)

The setting. Northwind Analytics (synthetic) is a US software company. Its Q3 internal update slide carries 247 words of continuous prose in eight sentences, the longest of them 63 words, in a single text box that autofit has stepped down to 16 point. Here's the real content. Net revenue retention finished at 104% against a 112% plan. New logo acquisition ran ahead of plan and expansion revenue ran below it. Gross retention was unchanged. Two enterprise renewals slipped to Q4, Q4 pipeline coverage is 2.8 times against a 3.0 target, and full-year guidance is unchanged.

The wrong version. The title says "Q3 Business Update". The body is a paragraph beginning "In Q3 we continued to make progress against our stated objectives across the commercial organization, with new logo acquisition tracking modestly ahead of the internal plan while expansion revenue within the installed base came in somewhat below where we had hoped", and it continues in that register for seven more sentences. Type sizes on the slide run 11, 14, 16, 18, 20, 24, 28 and 40 point, eight distinct sizes, with a smallest adjacent ratio of 1.1111. The hierarchy rule scores 0.0 on both halves.

Two arithmetic facts describe what that costs. At 200 words a minute of silent reading, 247 words take 74.1 seconds. A five-second glance covers 16.67 words, so the slide is 14.82 glances deep. And with eight type sizes, the reader has no way to tell which of the eight things on the slide is the point, because a size difference of 11% isn't a rank.

The corrected version. Same facts, 46 words, seven lines and a two-line title:

Expansion, not new logos, is what missed: NRR came in at 104%
against a 112% plan                                        [32 pt]

New logos ahead of plan                                    [18 pt]
Expansion revenue below plan                               [18 pt]
Gross retention unchanged                                  [18 pt]
The gap: two enterprise renewals slipped to Q4             [24 pt]
Q4 coverage 2.8x against a 3.0x target                     [18 pt]
Full-year guidance unchanged                               [14 pt]

Four type sizes: 14, 18, 24 and 32 point, with adjacent ratios of 1.2857, 1.3333 and 1.3333 and a hierarchy score of 1.0. The 24 point line carries the cause, and it's the only line promoted above body size. So the rank is unambiguous without a single change of colour.

The word count falls from 247 to 46, a cut of 201 words, or 81.38%. The after slide is 18.62% of the before slide's length. Reading time falls from 74.1 to 13.8 seconds, which saves 60.3 seconds per reader per slide. In glances, the slide goes from 14.82 to 2.76.

The numbers also get stated as differences instead of levels, and that's what makes them checkable. NRR at 104% against a 112% plan is a miss of 8.00 percentage points, or 7.14% in relative terms. Coverage at 2.8 times against a 3.0 target is 6.67% short. On a $40.0 million quarterly quota, 3.0 times coverage means $120.0 million of pipeline against the $112.0 million held. So the gap is $8.0 million of pipeline to build, and that's a number you can hand somebody as a target.

The glance test. Both versions were shown for five seconds each to a synthetic cohort of 24 readers, who then wrote down the slide's one message. The before slide was read correctly by 5, which is 20.83%. The after slide was read correctly by 21, which is 87.50%. The gain is 66.67 percentage points, and 4.2 times as many readers got it. (These are synthetic training figures built for this example, not a published experiment. Run the protocol on your own slide and expect different numbers.)

Read it. The failure here was editorial, and the repair was only partly typographic. Cutting 201 words isn't a typography decision, and no type scale would have rescued the original. The type scale made the surviving 46 words rank themselves, so a reader with five seconds spends them on the right line. Notice what the rebuild doesn't claim. It doesn't say the quarter was good. The action title says expansion missed. That's worse news than "Q3 Business Update" implied, and it's easier to act on.

Worked example 3: Distribute, in the wrong order (India, ₹)

The setting. The four KPI cards from Worked example 1, before the rebuild. Their widths are 2.30, 2.28, 2.35 and 2.30 inches. Their left edges are 0.52, 2.95, 5.48 and 8.05, and their tops are 1.90, 1.88, 1.92 and 1.90. They carry group revenue of ₹6,300.0 crore, growth of +10.76%, West's share of 38.37%, and three of four regions growing.

The wrong version. Select all four, run distribute horizontally, and stop. The gaps become equal, and the row still looks wrong.

Measure what distribute actually fixed and what it didn't. Before, the gaps are 0.13, 0.25 and 0.22 inches, with a mean of 0.20, a standard deviation of 0.0624 and a range of 0.12 inches. That's 8.64 points between the widest gap and the narrowest. Distribute sets all three to 0.20 and drives that standard deviation to zero. It leaves the widths alone, at a standard deviation of 0.0299 inches and a range of 0.07 inches. It leaves the tops alone too, with their range of 0.04 inches, or 2.88 points. And the row still spans 9.83 inches inside a 12.3333 inch band, so it floats inside the frame with no relationship to it.

Now the reader sees four boxes of visibly different widths at perfectly equal spacing, and that reads as though the width differences were intended. The third card is the widest at 2.35, so it looks like the important one. It isn't. It carries West's 38.37% share, the third of four facts.

The corrected version. Run the six operations in order.

Now the row spans 12.3333 inches instead of 9.83, the four cards are 2.9333 inches wide each, and all sixteen edges of the group land on guides. Steps 1 and 2 are what make step 4 mean anything. Distribute over equal boxes inside a fixed span produces the only arrangement possible, so arithmetic can check it and your eye doesn't have to.

Read it. Distribute is the operation that most visibly improves a slide. So it gets run first, and the resulting slide is hard to debug. A row of equal-gap unequal boxes passes a casual distribute check and fails the alignment score, because the score checks all four edges of every object, while the distribute check only looks at gaps. If your row looks wrong and the gaps are provably equal, check the widths and then the tops, in that order.

Worked example 4: A room 45 feet deep (US, $)

The setting. Cascade Industrial (synthetic) holds an investor day in a hotel ballroom. The screen is 10 feet wide. Seating is seven rows of 24 at 15, 20, 25, 30, 35, 40 and 45 feet from the screen, so 168 seats. The deck was built on a laptop and carries 32 point titles, 18 point body text, 14 point chart axis labels and 11 point source footnotes.

The wrong version. Ship it as built. A 10 foot screen against the 13.3333 inch slide is a magnification of 9.0. A capital letter takes up about 0.7 of the em box, so the angular size of a capital at each size and each row follows directly. Against the 15 arcminute working threshold:

So 120 of the 168 seats, which is 71.43%, can't read the chart axis labels. Seventy-two seats, 42.86%, can't read the body text. Only the titles reach every seat. The 14 point axis labels stop at 23.40 feet in a room that's 45 feet deep, a shortfall of 21.60 feet.

The corrected version. You have three repairs available, and they aren't equivalent.

Raise the type. The smallest size readable from the last row is 26.93 point, so 28 point is the smallest round size that serves the room. Body text at 28 point holds about two fifths of the words that fit at 18 point, because the type grows on both axes at once. That forces the editorial cut a room this size needs anyway. Titles stay at 32 or move to 40.

Change the room. Eighteen point serves everyone out to 30.08 feet, so a room 30 feet deep works with the deck as built. Add a second screen halfway down the room and the furthest viewer is at 22.5 feet. There the smallest readable size is 13.46 point, and the whole existing scale clears.

Change the screen. Magnification is linear in screen width, so the same 18 point text reaches 30.08 feet on a 10 foot screen, 36.10 feet on a 12 foot screen, 42.11 feet on a 14 foot screen and 48.13 feet on a 16 foot screen. A 16 foot screen fixes the room without touching the deck.

Our recommendation combines two of the three. Ask for the room dimensions before you build, size the body text for the back row, and treat any content that can't survive at that size as handout material. It doesn't belong on the screen.

Read it. The trap here is that everything looked fine during the build. Take an 18 point capital on a laptop screen 13 inches wide, viewed at 24 inches. It subtends 24.44 arcminutes, which is 1.62 times what the same letter subtends at the back of a 30 foot room, and about 2.4 times what it subtends at 45 feet. You can't catch this by looking at the screen you're building on, and nothing replaces asking how deep the room is. The numbers can't tell you whether the material belongs on a slide at all. A chart whose axis labels have to be 28 point is a chart with too many categories, and sometimes enlarging the type is the wrong repair.

Worked example 5: Three corrections, and no fourth (India, ₹)

The setting. One slide from the Meghdoot deck carries three elements whose mathematical centre isn't their visual centre. A status pill 0.40 inches tall holds the word AUDITED in 24 point capitals. A play triangle 0.60 inches wide is set inside a 1.20 inch circle marking a video link. And a KPI marker pairs a 0.36 inch circle with a 0.36 inch square.

The wrong version. Centre all three by bounding box, which is what every alignment command does, and accept the result.

The corrected version. Each case has a computable correction.

Start with the capitals. A typical sans face puts its ascender 0.75 em above the baseline and its descender 0.21 em below, so the box is 0.96 em tall and its centre is 0.27 em above the baseline. A line of capitals has no descenders, and its visual centre is at about 0.35 em. So box-centring puts the capitals 0.08 em too low. At 24 point that's 1.92 points, or 0.0267 inches. Inside a 0.40 inch pill that's 6.67% of the pill's height, and once you've seen it you can't miss it. The fix is to raise the text by 1.92 points. Better still, set the pill's internal top and bottom insets, so the correction is stored in the shape style and nobody has to apply it by hand.

Then the triangle. A right-pointing triangle 0.60 inches wide has its bounding-box centre 0.30 inches from the flat edge and its centroid at 0.20 inches, because the mass is at the base. Centre it by bounding box and the visual mass ends up left of the circle's centre. The correction is 0.10 inches, which is 7.2 points and 8.33% of the 1.20 inch circle's diameter.

Last, the circle and the square. A circle whose diameter equals the square's side has π/4 = 0.785398 of its area. At 0.36 inches the square covers 0.1296 square inches and the circle 0.1018, which is 21.46% less. To match the areas you'd need a circle 12.8379% wider, at 0.4062 inches, and that looks like a mistake in the other direction. A 2% overshoot to 0.3672 inches leaves the circle 18.29% smaller in area. A 3% overshoot to 0.3708 inches leaves it 16.68% smaller. Either reads as balanced. The eye stops complaining long before the areas agree.

Read it. That's three corrections on one slide, 0.1267 inches of movement in total once the capitals rise 0.0267 and the triangle shifts 0.10. Each one comes from a calculation. None of them is a judgement call. The habit that matters most here is the negative one. A fourth nudge, made because something "looked off", is how a grid stops being maintainable. The next person to open the file can't tell which offsets are corrections and which are drift, so they either preserve every error or flatten every correction. If you can't name the case and state the number, put the object back on the line.

Worked example 6: A block that has to fit (US, $)

The setting. A Northwind slide needs a 120 word explanatory paragraph above a chart. The live area is 12.3333 by 6.5 inches. The title band runs from 0.50 to 1.25, and a 0.25 inch gap follows it, which leaves 5.50 inches of height for the block and everything below it. Take 6.1 characters a word, including the space, and that gives 732 characters.

The wrong version. Drop the paragraph into a full-width text box, since the slide is wide and the box defaults to the placeholder. Twelve columns is 12.3333 inches. At 18 point that runs 98.67 characters a line, so the block takes 7.42 lines, rounded up to 8. At 1.35 leading the line box is 24.30 points and the block is 2.70 inches tall.

On height alone that's the best of the options, and it's the one to reject. Ninety-eight characters is a third beyond the 45 to 75 range that Robert Bringhurst gives for a single column in The Elements of Typographic Style. You pay for it on the return sweep, where the eye has to travel 12.33 inches back and find the next line down. Readers lose their place and re-read, and none of it shows up as anything visible on the slide.

The corrected version. Four candidates, measured:

Set across eight columns at 18 point with 1.35 leading, the block is 4.05 inches tall. That leaves 1.45 inches for the chart, and 1.45 inches isn't a chart. Tighten the leading to 1.20 and you recover 0.45 inches, which gets you to 1.90. Still not a chart. Six columns is readable and 5.40 inches tall, which leaves nothing at all. Drop to 14 point across eight columns and the height comes down to 2.3625 inches, but the measure goes up to 83.89 characters. So it fails on the other axis, and 14 point is a caption size that won't survive a large room.

The block doesn't fit. That's the finding, and the repair is editorial. Cut the paragraph to 46 words, which is 5 lines and 1.6875 inches at eight columns and 18 point, a word reduction of 61.67%. Put the remainder in the speaker notes or the appendix.

Read it. Running the arithmetic turns an argument about taste into a fitting problem with no solution. Otherwise somebody would have proposed 14 point, somebody else would have proposed tighter leading, and both proposals would have looked reasonable, because each fixes the height. The table shows that every option that fixes the height breaks either the measure or the room. That leaves the word count as the only variable with room to move. Notice where the method stops, though. It tells you the block has to shrink, and it says nothing about which 74 words to cut.

Verify

Every figure printed above, at full precision, in the order it appears.

Canvas and grid: slide 13.333333333333334 × 7.5 in, area 100.0 sq in, aspect 1.777777777777778, EMU 12,192,000 × 6,858,000. 4:3 legacy area 75.0 sq in; widescreen gain 0.3333333333. Content width 12.333333333333334; content height 6.5; column width 0.8444444444444444 in = 772,160.0 EMU = 60.8 pt. Live area 80.16666666666667 sq in = 0.8016666666666667 of the slide; margin band 0.19833333333333325. Spans: 12 columns 12.333333333333334; 8 columns 8.155555555555557; 6 columns 6.066666666666666; 4 columns 3.977777777777778; 3 columns 2.9333333333333336; 2 columns 1.8888888888888893. Four 3-column cards at x 0.5, 3.6333333333333333, 6.766666666666667, 9.9, last right edge 12.833333333333334. Five equal cards 2.3066666666666666, on no column line. Vertical rhythm 0.25 in gives 27 lines from 0.5 to 7.0. Distinct vertical grid lines 24.

Worked example 1: before, 12 elements, 48 edges, 22 horizontal breaks, 23 vertical breaks, 45 total, score 1 − 45/48 = 0.0625; distinct left edges 11, distinct top edges 11; bounding-box ink 55.156400000000005 sq in = 0.551564. After: 0 breaks, score 1.0; distinct left edges 6, distinct top edges 5; ink 61.05000000000001 sq in = 0.6105000000000002. Score improvement 0.9375; left-edge count fell 0.454545. Composite before 0.3925, after 0.9257142857142857, improvement 0.5332142857142856 = 1.358508 relative; machine mean before 0.3875, after 0.9428571428571428, improvement 0.5553571428571429.

India data: FY25 total 6300.0, FY24 total 5688.0, growth 0.10759493670886067. West 2417.0/2071.0 − 1 = +0.167069, share 0.38365079365079363; South 1905.5/1702.0 − 1 = +0.119565, share 0.30246031746031744; North 1284.5/1210.0 − 1 = +0.061570, share 0.20388888888888887; East 693.0/705.0 − 1 = −0.017021276595744705, share 0.11. Shares sum 1.0; rounded to two decimals they sum to 100.01%. Growth spread 0.18409032536445546 = 18.41 points. West contribution (2417.0 − 2071.0)/612.0 = 0.565359477124183. East change −12.0 cr.

Worked example 3: before gaps [0.13000000000000034, 0.25, 0.22000000000000064], mean 0.20000000000000032, sd 0.0624499799839839, range 0.11999999999999966 in = 8.639999999999976 pt. Before widths [2.3, 2.28, 2.35, 2.3], sd 0.02986078811194833, range 0.07000000000000028. Before tops range 0.040000000000000036 in = 2.8800000000000026 pt. After gaps [0.2, 0.2, 0.2], sd 4.19e-16; after widths all 2.9333333333333336. Card span before 9.830000000000002, after 12.333333333333334.

Type scale: ratios from 18 pt at 4/3 give 18.0, 24.0, 32.0 and, downward, 13.5. Shipped 14.0, 18.0, 24.0, 32.0; caption deviation 0.03703703703703698. Adjacent ratios 1.2857142857142858, 1.3333333333333333, 1.3333333333333333; minimum 1.2857142857142858, clears the 1.15 floor. Title to caption 2.2857142857142856. A fifth level between body and lead-in must lie from 20.7 to 20.869565217391305 pt; the midpoint 20.784609690826528 pt has ratio 1.154700538379251 to each neighbour, just above the floor; 21 pt gives 24/21 = 1.1428571428571428 and 20.5 pt gives 20.5/18 = 1.1388888888888888, both below it. Hierarchy before, sizes [11, 14, 16, 18, 20, 24, 28, 40], min ratio 1.1111111111111112, score 0.0; after, sizes [14, 18, 24, 32], score 1.0. Autofit ladder from 24 pt [24.0, 22.0, 20.0, 18.0, 16.0]; 16/24 = 0.6666666666666666; apparent area 0.4444444444444444; cap-height loss at 9× projection 0.7 in.

Worked example 2: before words 247, after words 46, reduction 201 = 0.8137651821862348; after share 0.1862348178137652. At 200 wpm, before 74.10000000000001 s, after 13.8 s, saving 60.3 s. Five-second glance 16.666666666666664 words; before 14.820000000000002 glances, after 2.7600000000000002. Before sentences 8, longest 63 words. NRR miss −0.08000000000000007 = −8.00 points, relative −0.07142857142857151. Coverage gap −0.20000000000000018× = −0.06666666666666676; pipeline needed 120.0 against 112.0, shortfall 8.000000000000007. Glance test 5/24 = 0.20833333333333334 and 21/24 = 0.875; gain 0.6666666666666666; ratio 4.2.

Legibility: 10 ft screen, magnification 9.0, cap fraction 0.7. Max distance at 15 arcmin: 11 pt 18.382279268673827 ft; 14 pt 23.395628160130318; 18 pt 30.080093348738984; 24 pt 40.106791131651974; 28 pt 46.791256320260636; 32 pt 53.4757215088693. Rows [15, 20, 25, 30, 35, 40, 45] ft, 24 seats each, 168 total; seats served 11 pt 24, 14 pt 48, 18 pt 96, 24 pt 144, 28 pt 168. Cannot read 14 pt 120/168 = 0.7142857142857143; cannot read 18 pt 72/168 = 0.4285714285714286. Minimum size at 45 ft 26.928107922043957 pt; at 30 ft 17.952071948029307; at 15 ft 8.976035974014653; at 22.5 ft 13.464053961021978. Shortfall of 14 pt in a 45 ft room 21.604371839869682 ft. 18 pt reach by screen width: 10 ft 30.080093348738984; 12 ft 36.09611201848678; 14 ft 42.11213068823458; 16 ft 48.12814935798237. Laptop: 13 in wide at 24 in, magnification 0.975, 18 pt cap 0.170625 in, 24.439819198437167 arcmin, 1.6249829908675257 times the 30 ft figure.

Measure and leading: mean advance 0.5 em. Characters per line at 18 pt: 4 columns 31.82; 6 columns 48.53333333333333; 8 columns 65.24444444444445; 9 columns 73.6; 12 columns 98.66666666666667. 66 characters at 18 pt needs 8.25 in; 8 columns is 8.155555555555557, short by 0.094444. Line boxes at 18 pt: 1.00 18.00 pt, 1.15 20.70, 1.20 21.60, 1.35 24.30, 1.45 26.10, 1.60 28.80; five lines 1.25, 1.4375, 1.5, 1.6875, 1.8125, 2.0 in; the 1.20 against 1.45 difference over five lines 0.3125 in. Lines of 18 pt at 1.35 in 6.5 in 19.25925925925926.

Worked example 6: 120 words at 6.1 characters = 732.0 characters. 8 columns 18 pt 1.35: 65.24 chars/line, 11.2193 lines rounded to 12, height 4.050000000000001 in. 12 columns: 98.67 chars/line, 7.4189 rounded to 8, height 2.7. 6 columns: 48.53, 15.0824 rounded to 16, height 5.4. 8 columns 14 pt: 83.89, 8.7262 rounded to 9, height 2.3625000000000003. 8 columns 18 pt 1.20: height 3.5999999999999996. Available height 5.5 in; 8-column block leaves 1.4499999999999993 in. Cutting to 46 words: 5 lines, 1.6875000000000002 in, reduction 0.6166666666666667.

Optical corrections: square side 1.0, equal-area circle diameter 1.1283791670955126, 12.8379% wider; equal-diameter circle area 0.7853981633974483, −21.4602%; overshoot 2% −18.2872%, 3% −16.6771%, 5% −13.4099%. Font box: ascender 0.75 em, descender 0.21, box 0.96, centre 0.27 em above baseline, caps centre 0.35; offset −0.07999999999999996 em = −1.4399999999999993 pt at 18, −1.919999999999999 at 24, −2.5599999999999987 at 32 = −0.035555555555555535 in. Pill 0.40 in tall at 24 pt: offset −0.026666666666666655 in = 0.06666666666666664 of the pill. Triangle 0.6 in: box centre 0.3, centroid 0.19999999999999998, correction 0.09999999999999999 in = 7.199999999999999 pt = 0.16666666666666666 of the width and 0.08333333333333334 of a 1.2 in circle. KPI marker at 0.36 in: square area 0.1296, circle area 0.10178760197630929, equal-area diameter 0.4062165001543845; 2% overshoot area 0.10589982109615217, 3% 0.10798646693666655. Quotation mark at 24 pt overhang 0.09333333333333334 in. Total movement of the three Worked example 5 corrections 0.12666666666666665 in, plus a 0.0072 in resize on the circle.

Margins: live areas 87.86, 83.9733, 80.1667, 76.44, 71.0, 62.3333 sq in at margins 0.30, 0.40, 0.50, 0.60, 0.75, 1.00; shares 0.8786, 0.8397, 0.8017, 0.7644, 0.7100, 0.6233. Moving 0.50 to 0.30 buys 7.693333333333342 sq in = 0.095967. Overscan at 2.5% per edge removes 0.33333333333333337 in horizontally and 0.1875 vertically; 0.5 in margin clears by 0.16666666666666663 and 0.3125; 0.25 in margin is short by 0.08333333333333337 horizontally.

Hierarchy devices: size title to body 1.7777777777777777, body to caption 1.2857142857142858; weight 1.6 working figure across 1.5 to 1.8; proximity 0.25 in above against 0.0833333333 below = 3.0000000011999997.

Practice set

Work each one before you read the solution. You'll need a calculator. Unless a problem says otherwise, the canvas is 13.3333 by 7.5 inches, the grid is twelve columns with 0.5 inch margins and 0.2 inch gutters, the alignment tolerance is 0.02 inches, capitals take up 0.7 of the em box, the legibility threshold is 15 arcminutes, and numeric answers within 2% count as correct.

P1 (guided). Derive a grid on a different canvas (India). A client's house template is still on the 4:3 canvas, 10 by 7.5 inches. Keep the 0.5 inch margins, the 0.2 inch gutter and twelve columns. Give the column width in inches and in English Metric Units, and the width of a three-column card.

Solution. The content band is 10 − 2 × 0.5 = 9.0 inches. Eleven gutters take 11 × 0.2 = 2.2 of it, leaving 6.8 inches for twelve columns. So the column width is 6.8/12 = 0.5666666667 inches, which is 0.5666666667 × 914,400 = 518,160.0 EMU. That's a whole number, because 0.56667 inches is 40.8 points and the EMU was chosen to divide cleanly. A three-column card spans three columns and two gutters: 3 × 0.5666666667 + 2 × 0.2 = 2.1 inches. That card is 28.4% narrower than the 2.9333 inch card on the widescreen grid. It's the real cost of the older canvas, and the reason you can't widen a 4:3 deck by stretching it.

P2 (guided). Which card counts fit (US). On the widescreen band of 12.3333 inches with 0.2 inch gutters, compute the width of n equal cards for n = 2, 3, 4, 5 and 6, and say which land on a column line.

Solution. The band holds twelve columns, so n cards land on column lines only when n divides 12. The widths are two cards 6.0666666667 (6 columns each, lands), three cards 3.9777777778 (4 columns, lands), four cards 2.9333333333 (3 columns, lands), five cards 2.3066666667 (lands on nothing, since 12/5 = 2.4 columns) and six cards 1.8888888889 (2 columns, lands). Five across is the one to avoid, and the reason is arithmetic. No column boundary falls at 2.4 columns, so the internal edges of a five-card row can't align with anything else on the slide.

P3 (guided). Fix a three-object row (India). Three boxes sit at x = 1.10 (width 1.80), x = 5.42 (width 1.90) and x = 10.55 (width 2.10). Compute the two gaps and their difference, then give the equal-gap positions.

Solution. Box 1 ends at 1.10 + 1.80 = 2.90, so gap 1 is 5.42 − 2.90 = 2.52 inches. Box 2 ends at 5.42 + 1.90 = 7.32, so gap 2 is 10.55 − 7.32 = 3.23. The difference is 0.71 inches, which is 51.12 points. The row spans 1.10 to 10.55 + 2.10 = 12.65, a total of 11.55 inches. The three boxes take 1.80 + 1.90 + 2.10 = 5.80 of that, leaving 5.75 inches of free space and an equal gap of 5.75/2 = 2.875. Box 2 moves to 1.10 + 1.80 + 2.875 = 5.775, a move of 0.355 inches, and box 3 stays at 10.55. Only one object moves, and that's the general result. Distributing n objects between fixed endpoints moves the n − 2 objects in the middle.

P4. Build and check a scale (US). Base 16 point, ratio 1.25, four levels. Give the exact sizes, the rounded sizes, the largest rounding error as a fraction, and say whether the rounded scale clears the 1.15 floor.

Solution. Exact: 16.0000, 20.0000, 25.0000, 31.2500. Rounded to whole points: 16, 20, 25, 31. The only rounding is at the top level, 0.25/31.25 = 0.008, so the largest error is 0.8%. Adjacent ratios on the rounded scale are 20/16 = 1.25, 25/20 = 1.25 and 31/25 = 1.24. All of them are above 1.15, so the scale clears. Note that a 1.25 ratio gives a smaller top-to-bottom range than 4/3 does, 31/16 = 1.94 against 32/14 = 2.29. So a 1.25 scale has less room to express rank, and it suits a reading document better than a projected slide.

P5. Two rooms (India and US). (a) A 16 point caption on a 12 foot screen viewed from 40 feet. (b) A 28 point line on a 16 foot screen viewed from 60 feet. Compute the magnification, the projected cap height and the angular size, and say which passes 15 arcminutes.

Solution. (a) Magnification is 12 × 12 / 13.3333 = 10.8. The cap on the slide is 16 × 0.7/72 = 0.155556 inches, projected to 1.68 inches. The angle is atan(1.68/480) in arcminutes = 12.03, which fails. (b) Magnification is 16 × 12 / 13.3333 = 14.4. The cap is 28 × 0.7/72 = 0.272222 inches, projected to 3.92, and the angle is atan(3.92/720) = 18.72 arcminutes, which passes. Compare the two. The second case is a much bigger room and it works, because the screen grew faster than the distance did. Screen width and viewing distance both scale the angle linearly, so their ratio is what decides it.

P6. Set the measure (US). A text block spans columns 1 to 7 at 20 point. Compute the width, the characters per line, and say whether it sits inside the readable range.

Solution. Columns 1 to 7 span seven columns and six gutters: 7 × 0.844444 + 6 × 0.2 = 7.111111 inches, which is 512 points. At 20 point with a 0.5 em mean advance, each character takes 10 points, so the line runs 51.2 characters. That's inside the 45 to 75 range with room on both sides, so seven columns is a usable body span at 20 point. Now try the same seven columns at 14 point. That gives 512/7 = 73.1 characters, still inside the range but right at its top edge, so a longer line at 14 point would need narrowing. Keeping an eye on it wouldn't be enough.

P7. Correct the capitals (India). A section divider sets a line of capitals at 40 point inside a coloured band. Compute the optical correction in points and in inches.

Solution. The offset between the font box's centre at 0.27 em and the visual centre of capitals at 0.35 em is −0.08 em. At 40 point that's −0.08 × 40 = −3.2 points, or −3.2/72 = −0.0444 inches. The text needs to move up by that amount. The correction scales linearly with size. At 18 point the same construction needs only 1.44 points and is easy to miss, while at 40 point you can see it from across a room. That's the general rule for optical corrections. They matter most where the type is largest, which means the title and the section divider.

P8. Predict a score (US). A nine-element slide has 11 edges off the grid. Give the alignment score. Then say how many edges would have to be fixed to reach 0.90.

Solution. Nine elements give 4 × 9 = 36 edges, so the score is 1 − 11/36 = 0.6944. To reach 0.90 the breaks must fall to 0.10 × 36 = 3.6, so at most 3 breaks, which means fixing 8 of the 11. This arithmetic explains why partial alignment work rarely moves a score much. Each fixed edge is worth 1/36 = 0.0278, so fixing three edges takes the score from 0.6944 to 0.7778, and it still fails.

P9. Measure the ink (India). A slide's elements have bounding boxes totalling 62.4 square inches. Give the coverage as a share of the canvas, then the fall in coverage if the total is cut to 41.2 square inches, and say what the second number does and does not tell you.

Solution. The canvas is 100 square inches, so 62.4 square inches is 0.624 of the slide and 41.2 is 0.412. The fall is (62.4 − 41.2)/62.4 = 0.3397, a 33.97% reduction in covered area. It tells you that a third of the slide's occupied area is gone. It doesn't tell you whether the slide reads better, and Worked example 1 is the counter-example. There the rebuilt slide covers 0.6105 of the canvas against the original's 0.5516, and it's far clearer. Coverage measures density. Clarity is a question of whether the gaps are organised.

P10. Price a glance (US). A slide carries 137 words. At 200 words a minute, compute how many five-second glances it takes to read and the total reading time. Then compute the reduction needed to reach 22 words.

Solution. A five-second glance covers 5/60 × 200 = 16.6667 words, so 137 words is 137/16.6667 = 8.22 glances, or 8.22 × 5 = 41.1 seconds of reading. Cutting to 22 words is a reduction of 1 − 22/137 = 0.8394, or 83.94%. That leaves a slide 22/16.6667 = 1.32 glances deep, inside the budget an audience actually gives. Carry this number away with you. A slide more than about two glances deep won't be read during a presentation, whatever else is true about it.

P11. Diagnose an autofit (India). A body placeholder set at 28 point has been autofitted down to 17 point on one slide, while the caption level stays at 14 point. Compute the ratio of the shrunk size to the intended size and the resulting body-to-caption ratio, and say what happens at the next autofit step.

Solution. The shrunk body is 17/28 = 0.6071 of the intended size. Against the 14 point caption it's 17/14 = 1.2143. That still clears the 1.15 floor, but it's well below the intended 18/14 = 1.2857, so on this slide the two levels have moved closer together, and nowhere else. One more autofit step to 16 point gives 16/14 = 1.1429, below the floor. At that point body and caption are the same rank as far as a reader is concerned. Worse, the deck's type scale now depends on the slide. Paragraph length is setting rank where role should, so a reader can't infer importance from size anywhere in the file.

P12. Work backwards from a gate (US). A slide scores 0.9020, 0.8600, 0.9100, 0.9500, 0.8800, 0.8200 and 0.8947 on the seven Linter rules and the author's committed rubric self-score is 0.75. Give the machine mean, the composite, and the rubric score that would be needed for a bare pass at 0.80.

Solution. The machine mean is (0.9020 + 0.8600 + 0.9100 + 0.9500 + 0.8800 + 0.8200 + 0.8947)/7 = 0.8881. The composite is 0.6 × 0.8881 + 0.4 × 0.75 = 0.8329, which passes. Working backwards, a bare pass needs (0.80 − 0.6 × 0.8881)/0.4 = 0.6679, so the rubric could fall to 0.67 and the slide would still clear the gate. That headroom is the point of the blend. A slide with clean geometry buys some tolerance on judgement. A slide with weak geometry can't buy its way through on judgement alone, because 0.6 × 0.50 + 0.4 × 1.00 = 0.70 fails even with a perfect rubric.

Applied mini-project

Three flawed slides, rebuilt. Budget about 150 minutes, roughly 50 minutes a slide. The three come as written specifications, so build each one as described first, then repair it, because reproducing the defect is what teaches you to recognise it in a deck somebody hands you.

The first is the twelve-element regional slide from Worked example 1 (India, ₹). Build the before state from its position table, using the Meghdoot regional figures, then rebuild it on the twelve-column grid.

The second is the wall of text from Worked example 2 (US, $). Build a slide carrying the 247 word Northwind paragraph in a single autofitted text box, with eight type sizes scattered across the slide's furniture, then rebuild it.

The third is your own, and it should be the worst slide you have. Take something real from your own work, or from a deck you received, that you already suspect is weak. Measure it before you touch it.

You deliver five things for each of the three.

1. The before-state measurement. Count the elements, count the edges that miss the grid at a 0.02 inch tolerance, and give the alignment score. Count the distinct left-edge values and distinct top-edge values. Count the distinct type sizes and give the smallest adjacent ratio. Count the words.
2. The rebuild, on a stated grid, with every element's span written down. Nothing may be positioned by dragging. Every x and every width comes from a column line.
3. The after-state measurement: the same six counts, plus the seven Linter rule scores and the machine mean.
4. The committed rubric self-score. Write your own 0 to 1 score for the slide before you look at the rubric below, then compute the composite as 0.6 × machine + 0.4 × rubric. Commit first. A self-score written after you've read the rubric measures nothing.
5. One sentence naming the rule. Don't describe what you changed. Name the single rule the before slide broke and the number that proves it. "The four cards had a 0.12 inch gap range and a 0.07 inch width range, so distribute had been run before the widths were equalised" is a sentence that qualifies. "It looked messy" isn't.

The pass condition. Each rebuilt slide must reach a composite of 0.80. The third slide is your own, so it also has to reach a word count you can defend against the glance test. Show it to somebody for five seconds and score whether they can state your message.

The rubric is twenty points, passing at 16, with no zero on any row.

Any slide whose after-state measurement is missing scores zero on rows 1 and 4, however good the rebuild looks, because an unmeasured rebuild is an opinion. The rubric grades the process. Appearance earns nothing on its own. The argument is that a slide built by a repeatable method will be good the next forty times, and a slide that happened to look right won't.

Reading & resources

Core (do these).

• Robert Bringhurst, The Elements of Typographic Style: the standard reference, and the source of the 45 to 75 character measure used throughout. Read the chapters on rhythm and proportion and on the shaping of the page. The historical material repays a later pass. [Paid] [Advanced]
• Ellen Lupton, Thinking with Type: the shortest useful route into type anatomy, scale and grids, written for people who need to use type without becoming type designers. The companion site carries much of the reference material free. [Paid] [Beginner]
• Matthew Butterick, Practical Typography: free to read at practicaltypography.com, opinionated, and the fastest correction available for the habits most analysts arrive with. The sections on point size, line spacing and line length map straight onto slide work. [Free] [Beginner]
• Josef Müller-Brockmann, Grid Systems in Graphic Design: where the column-grid method taught above comes from, with the arithmetic worked through for print. It's slower than it needs to be, and worth it anyway, because it trains you to see every layout as a set of derived numbers. [Paid] [Intermediate]

Going deeper.

• Kimberly Elam, Grid Systems: Principles of Organizing Type: a compact treatment of why a grid's proportions matter, with worked examples that transfer directly to a slide canvas. [Paid] [Intermediate]
• Stephen Few, Show Me the Numbers: read the table-design chapters here and leave the chart chapters to the chart-craft branch. Its treatment of numeric alignment, column ordering and white space in tables is the best short version available. [Paid] [Intermediate]
• Nancy Duarte, slide:ology: the design-side case for restraint on a slide, including the argument for a small set of type roles. Read it alongside the grid material above, as a companion and no replacement. [Paid] [Beginner]
• Jean-luc Doumont, Trees, Maps, and Theorems: dense, short and unusually precise about typographic decisions in professional documents and slides, including line length and the relationship between a slide and its speaker. [Paid] [Advanced]

Standards and references (free).

• W3C Web Content Accessibility Guidelines: the contrast minimums are 4.5 to 1 for normal text and 3 to 1 for large text. Large means 18 point and above or 14 point bold and above (WCAG 2.2, checked September 2026; confirm, since the guideline version advances and the thresholds can move with it).
• ECMA-376, Office Open XML File Formats: free from ecma-international.org, and the primary source for the English Metric Unit, which is 914,400 to the inch and is what your file actually stores. Look it up when two objects that report the same position aren't in the same place.
• Your presentation software's own accessibility and layout documentation. The current widescreen default is 13.333 by 7.5 inches at 16:9 in desktop PowerPoint, autofit behaviour is configurable per placeholder, and both the default and the autofit ladder follow the release (Microsoft 365 desktop, checked September 2026; confirm against your own installation).
• A modular-scale calculator. Several free web tools generate a type scale from a base size and a ratio. Use one to explore ratios quickly, then write the four chosen sizes into your master by hand. A scale that only exists in a browser tab isn't a decision anybody can inherit.

Do this, not just read. Open any deck you didn't build and count two things: the number of distinct left-edge values on the busiest slide, and the number of distinct (font, size, weight) triples across the whole file. With the selection pane and the format panel, both counts take four minutes. Both will be higher than the owner would guess, and the second one is usually the bigger shock.

Teach it back & journal

Feynman prompt. Explain to a colleague who thinks slide design is a matter of taste why two slides carrying identical content can score 0.0625 and 1.0000 on the same rule. Don't use the words clean, professional or polish. Use the arithmetic. Twelve elements have 48 edges, and a slide built by dragging gets three of them onto a guide, which is roughly what chance produces. Twelve columns with 0.5 inch margins and a 0.2 inch gutter give a column width of 0.844444 inches. So a three-column card is 2.9333 inches and four of them fill the band to the edge, while five equal cards are 2.3067 inches and miss every column line. Four type sizes at a 4/3 ratio give 14, 18, 24 and 32 point, with a smallest adjacent ratio of 1.2857. Eight sizes give a smallest ratio of 1.1111 and no rank a reader can use. Then make the harder point. None of those numbers is about appearance. Each one is about whether a decision was made once and inherited, or made forty times by hand. If your colleague can tell you, unprompted, why five cards across a twelve-column grid is a problem, you've taught the idea and not just the arithmetic.

Journal prompt. Open the last deck you built and measure three things on its busiest slide before you change anything. Count the distinct left-edge values, using the position dialog and not your eye. Count the distinct (family, size, weight) triples across the whole file. And time yourself reading the busiest slide aloud, then divide the word count by 16.67 to get its depth in glances. Write the three numbers down. Then write a paragraph on which of the two failure families you belong to, and go by the evidence instead of your self-image. People whose left-edge count is high and whose type-role count is low are usually fast builders who never set up a grid. Their fix is one hour spent on a master, and it pays back on every deck afterwards. People whose type-role count is high and whose edges are fine are usually careful builders who've been solving each slide's problem locally. Their fix is harder. It means accepting that a slide which needed a fifth type size was a slide with too much on it. Finish with one habit small enough to survive a bad week: guides on the master before the first slide, and a rule that no object is ever positioned by dragging.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.