PowerPoint Ground-Up: Interface, Navigation, Sections and File Craft

Presentation track, branch pr-foundations (Foundations) · Node PR0.02 · ~8 focused hours · Mastery gate ≥ 85%

A deck leaves a laptop at 48.70 MB and bounces off a mailbox that stops at 25. The sender's first instinct is to delete slides, which is the one repair that costs content. The repair that costs nothing takes about nine minutes: eighteen photographs carry 79.47% of the file's weight, and every one of them is a 4032-pixel camera image sitting in a frame 13.33 inches wide. That is 302.4 pixels per inch on a surface a 1080p projector will render at 144. Resample to 150 and the deck lands at 12.83 MB, a 73.64% reduction, with a picture nobody in the room can distinguish from the original.

Here is a second deck with a different disease. A 34-slide investor update changes topic 30 times across its 33 adjacent pairs: business, financials, product, business, financials again. Group the same 34 slides into a cover and four named sections, five contiguous groups in all, and the switch count falls to 4. Eighteen slides move, no slide is rewritten, and the total edit to the words on the slides is zero.

That deck was never disorganised in its argument. It was disorganised in its filing.

Both faults live in the same place, and it is not where people look. Most of what makes a deck painful to build, painful to receive and painful to hand on is filing. Does a piece of text live in a placeholder or a floating box. Does a slide belong to a named group. What is the file called, which copy is current, and how many pixels does a picture actually carry. The canvas is the part everybody looks at. The filing system underneath it decides whether eight hours of work survives a reordering, a co-author, a printer, a projector and a mail server.

Eight focused hours is a realistic budget for what follows, and the split is lopsided on purpose. Roughly two go to the keyboard, which is the only part that feels like drill and the only part that pays back on every deck you will ever build. Two go to sections and the Outline, which is where a deck stops being a pile. Two go to file craft: weight, fonts, naming, recovery. The last two go to the two artefacts most people never separate, the projected deck and the printed handout, plus the accessibility habits that cost nothing at the start and cost a rebuild at the end.

There is a boundary here worth stating before you cross it. Everything taught here stops at using a template someone handed you. Deciding what lives at master level, what lives at layout level and what lives on the slide, and building a template that enforces those decisions, is the first node of the mechanics branch and it is a ten-hour job of its own. The distinction matters because half the repairs people attempt on an inherited deck are attempts to fix a master from the slide, which is the expensive direction.

PowerPoint is the reference tool, because it is what most working analysts are handed and because a track that teaches mechanics has to teach them somewhere specific. Nothing in the arithmetic is vendor-specific: pixels per inch, paper, contrast ratios and sort order behave identically in Google Slides, Keynote and LibreOffice Impress. The last teaching section maps the vocabulary across those tools once, so the habits transfer and nothing here reads as an advertisement.

Learning objectives

You can:

1. Work the four views deliberately, saying which question each one answers, and diagnose a deck you did not build by reading its Slide Sorter and Outline before opening a single slide.
2. Run twenty keyboard operations that cover about 91% of the operations in a realistic build task, and compute the payback in runs and in hours a year from stated assumptions.
3. Section a deck so that it survives reordering, and measure the improvement as the fall in adjacent topic switches, stating how many slides had to move.
4. Audit a deck in the Outline view, computing what share of its titles and body text is invisible there, and say what each invisibility implies about how the deck was built.
5. Write speaker notes that carry the sentence, the source and the expected question, and state what must never be moved off a slide into them.
6. Compute a deck's file weight by component, choose a resampling target from the room the deck will be shown in, and bring a stated file under a stated mail ceiling without deleting content.
7. Open an inherited deck, rank the available repairs by slides fixed per minute, and stop at the point where further repair costs more than it returns.
8. Produce a handout as a separate artefact, computing the scale factor of each N-up layout and the smallest on-slide type size that stays legible on paper.
9. Set naming, versioning and save intervals so the expected cost of a crash is bounded, and compute both the expected loss and the number of ordering errors an ad-hoc naming scheme produces.
10. Build reading order and alt text as habits, checking foreground and background pairs against the 4.5:1 and 3:1 contrast thresholds and counting the pairs a screen reader would announce out of order.

Prerequisites & connections

Builds on. PR0.01, and nothing else. You should arrive able to write a one-sentence takeaway, fill in the three lines of an audience profile, and choose a medium on purpose. That sequence decides whether a deck should exist at all; the work here begins one step later, at the moment you have decided a deck is the right artefact and have to build it without the tool fighting you. The arithmetic assumes school-level percentages and the ability to read a table.

Feeds forward. PR1.01 takes the template you learn to use here and teaches you to build one, which is where masters, layouts and placeholders get their proper treatment. PR1.02 takes the grid and the type scale that a good template encodes and teaches the geometry. PR1.03 covers objects, animation and the export step. The chart-craft branch, PR2.01 through PR2.03, assumes you can already produce a clean, sectioned, correctly sized file and spends its hours on what goes inside the chart. PR5.01 to PR5.03 assume the file hygiene taught here is automatic, because a board pack that will not open is not a board pack.

Deliberate non-overlaps. PR1.01 owns slide masters, layouts, placeholders and template construction, and the repair sequence for a deck that fights its own master; the treatment here stops at recognising the symptom and naming the level where the fix belongs. PR1.02 owns alignment, distribution, grids, guides and the type scale, so type sizes appear here only as a legibility calculation on paper, not as a design system. PR2.02 owns colour as data-visualisation craft, including the highlight-against-context formula, so colour appears here only as a contrast ratio against an accessibility threshold. PR0.01 owns audience, message and medium, including the reading deck against the presented deck, and the handout treatment here is the production consequence of a choice made there. PR4.02 owns the action title and horizontal logic; the Outline audit here counts action titles as a diagnostic without teaching how to write one.

1. The window, and the four views that answer four different questions

The application window has four regions and each one has a job. The ribbon across the top is a set of task groups rather than a menu of features. Home holds the text and object operations you use constantly and Insert brings new objects onto the canvas. Design and Transitions act on the deck rather than on the object, View switches your vantage point, and Slide Show controls what happens when you present. The thumbnail rail down the left is the deck as a sequence. The canvas in the middle is one slide. The notes pane below it holds what you will say. The status bar at the bottom carries the slide number, the view buttons and the zoom control, and it is the fastest route to a view change.

Four views matter, and choosing the wrong one is why deck work sprawls.

Normal is where you build one slide. It is the only view that shows you the canvas at working size with the notes pane and the thumbnail rail visible at the same time, and it is the wrong place to make a decision about the deck as a whole, because you can only see four or five thumbnails at once.

Outline shows the text that lives in title and content placeholders, as a linear indented list, with no pictures, no charts and no floating text boxes. That restriction is the point rather than a limitation. What survives into the Outline is what a machine can read as the deck's argument, and what does not survive is invisible to search, to accessibility tools, to the handout that prints outline text, and to anyone trying to reorder your deck quickly.

Slide Sorter shows the deck as an array of thumbnails with section headers and slide numbers. It is the only view where you can see the shape of the whole thing: how many slides sit under each heading, where the density is, whether the argument has a middle. Reordering happens here because dragging a slide in a grid is fast and dragging in a vertical rail is not.

Reading view plays the deck full-screen inside the application window rather than taking over the display. It exists so you can check what a viewer sees without the disruption of a real slideshow, and it is the honest way to answer "does this slide actually read", because at presentation size you see the type at the size the room sees it.

There is a fifth surface you meet at show time. Presenter view puts the current slide, the next slide, the notes and a timer on the speaking machine while the audience sees only the slide. It is not a deck-building view, but knowing it exists changes what you put in the notes, which is why it belongs in the same sentence as the other four.

The working rule is short. Diagnose in Slide Sorter and Outline, build in Normal, check in Reading view.

Most people do all four jobs in Normal, and then a deck they have worked on for six hours still surprises them the moment it goes on a screen.

2. Keyboard-first, and the arithmetic of learning twenty operations

Deck work is a large number of small operations. A realistic build of a twenty-slide deck from an outline and a folder of exhibits runs to something like 174 discrete operations: creating slides, duplicating them, moving them, grouping objects, nudging, aligning, pasting, saving, switching views, undoing. Of those, 158 have a keyboard route. Sixteen do not have a useful one, because inserting a chart, choosing a picture off disk and typing the words are inherently pointing-and-typing tasks. Twenty distinct operations therefore cover 90.80% of everything you do.

Price the two routes. A ribbon-and-mouse operation costs about 3.4 seconds once you include the eye movement to find the control, the travel and the click. The keyboard equivalent costs about 0.9 seconds. On this task the mouse route runs 537.2 seconds, or 8.95 minutes; the keyboard route runs 142.2 seconds, or 2.37 minutes. The saving is 395.0 seconds a run, which is 73.53% of the slower route, and the keyboard is 3.78 times faster on the covered operations.

Learning costs something. Budget six minutes of deliberate practice per operation, which is 2.0 hours for the twenty. At 395.0 seconds saved per run, the investment repays itself after 18.23 runs of the task. At three such tasks a week over 46 working weeks the recovered time is 54,510 seconds, or 15.14 hours a year.

Those figures are a model with stated assumptions, not a measurement of you. The 3.4 and 0.9 second costs are the two numbers that drive everything, and if your ribbon habits are faster than 3.4 seconds the payback lengthens proportionally. What the model does establish is the shape of the answer: the payback is measured in weeks, not months, and it is dominated by the handful of operations you perform ten or more times per deck rather than by exotic chords.

The twenty, with the times each appears in the modelled task.

Key assignments are for the Windows desktop build of PowerPoint in Microsoft 365 (checked September 2026; verify against your own installation, because assignments move between releases and several differ on macOS, where Command generally replaces Ctrl). Where no single chord exists, the keyboard route is the Alt key sequence that walks the ribbon. The model prices all twenty at the same 0.9 seconds for that reason: an Alt sequence is three or four keystrokes with no eye travel, and it beats finding a control by looking for it.

Two habits are worth more than the list. Tab cycles through the objects on a slide in their stored order, which means you can select an object that is buried under another one without hunting, and it is also a free preview of the reading order that the accessibility section makes explicit. And Ctrl+Z is only useful if you save often enough that undo is not your only safety net, which is the bridge to the file craft below.

3. Sections: the deck's table of contents, and what they buy

A section is a named group of consecutive slides. You add one at the currently selected slide, and every slide from there to the next section boundary belongs to it. Collapse a section in the thumbnail rail or the Slide Sorter and its slides fold into a single header line with a count. Drag the header and the whole group moves together. Print a section and only its slides print.

The reason sections matter is measurable. Take the 34-slide investor update from the opening. Tag each slide with its topic and count how often the topic changes between adjacent slides: 30 changes across 33 adjacent pairs, a switch rate of 0.9091. The longest unbroken run on a single topic is 2 slides. A reader moving through that deck re-orients on almost every page, and re-orientation is the cost that makes a long deck feel longer than it is.

Group the same slides into Cover, Business, Financials, Product and Asks, keeping every slide's content untouched, and the switch count falls to 4, a reduction of 26 switches or 86.67%. The work required is 18 slides moved, which is 52.94% of the deck, because a longest non-decreasing run of 16 slides was already in an order consistent with the new grouping. Nothing was written, rewritten or deleted.

Where the boundaries go is a judgement, and there is one rule that decides most cases. A section is a claim group, not a chapter heading. If you cannot say what the section argues in a single clause, it is a container rather than a section, and the fix is usually to merge it with its neighbour. Five sections in a thirty-slide deck is comfortable; twelve sections in a thirty-slide deck means the sections are doing the work of slide titles.

Sections also survive things that slide order does not. Insert four slides in the middle of a sectioned deck and they land inside a named group, so the structure absorbs them. Insert four slides into an unsectioned deck and the only record of where they belong is your memory of what you were doing. When two people work on the same file, sections are the cheapest possible coordination device, because "take the Financials section" is an instruction that survives someone else adding three slides to Business.

One habit prevents the commonest failure. Name a section for its claim, not its department. "Unit economics turned positive in Q2" is a section name that tells a co-author what belongs in it. "Financials" is a name that will accept anything with a number on it, which is how a six-slide section becomes a thirteen-slide section over two weeks.

4. The Outline as a storyline check, not a typing surface

The Outline view shows exactly one thing: text that lives in a title placeholder or a content placeholder. Text you typed into a floating text box is not there. Text inside a chart, a table, a picture or a SmartArt graphic is not there. That single restriction turns the Outline into the most honest audit available of how a deck was actually built, because it separates the text a machine can read from the text a machine cannot.

Run the audit on the 34-slide update. Twenty-six slides have their title in the title placeholder, so 76.47% of the deck shows a title in the Outline and 8 slides show nothing where a title should be. Nineteen slides have their body text in a content placeholder, so 55.88% of the bodies reach the Outline and 15 do not. Every one of those gaps is a floating text box that somebody dragged onto a slide because it was faster at the time.

The cost of that shortcut is not aesthetic. Text outside a placeholder does not appear in the Outline and does not print on an outline handout. Assistive technology announces it in creation order rather than in the order a reader would take it, unless you fix that order by hand. It does not inherit the master's typography, so it drifts on the next theme change, and it does not move when a layout changes. One floating text box is a nuisance.

Fifteen of them is a deck nobody can restructure without opening every slide.

The second thing the Outline gives you is the read-only-the-titles test, and it is brutal in a useful way. Collapse the Outline to titles alone and read them in order. If the titles make an argument, the deck has a storyline. If they read as a list of topics, the deck is an index. In the audit deck, 4 of the 34 titles are full sentences with a verb, which is 11.76% of all slides and 15.38% of the titles the Outline can see. The other 22 visible titles are labels. That ratio is the single fastest measurement of whether a deck will land, and it takes ninety seconds.

What the Outline is not is a typing surface. It is possible to type an entire deck there, and the decks produced that way are recognisable at a glance: every slide is a title with three or four bullets, because that is the only structure the Outline can express. Use it to read and to reorder, then build the actual slide in Normal view where a chart, a diagram or a single sentence can be the whole content.

5. Notes: what belongs there, and what must never be moved there

The notes pane holds text attached to a slide that the audience never sees. Four things belong in it.

The sentence you will actually say when the slide appears, written out, because the first sentence on a new slide is the one people improvise worst. The source of every number on the slide, precise enough to answer a challenge: the filing, the table, the date, the query. The answer to the question you expect, which is usually one question per slide and is the difference between looking prepared and being prepared. And the thing you deliberately kept off the slide, with a note about why, so that when a reviewer asks "shouldn't we mention the contract renewal", you can say that you considered it and where it went.

One category must never move to the notes. The argument itself stays on the slide. A slide whose claim is only in the speaker notes is a slide that says nothing when it is forwarded, and decks are forwarded far more often than they are presented. If you find yourself writing the point in the notes because it will not fit on the slide, the slide has too much else on it, and the repair is on the canvas.

Two operational facts about notes are worth knowing before they bite. Notes travel with the file, so anything you would not want a recipient to read should not be there when you send it, and internal decks routinely carry lines like "do not mention the pricing review" straight into a client's inbox. Notes also print, on the notes-page layout, which puts one slide on a page with its notes underneath and exports that way to PDF as well, so a pack printed or exported from that layout carries the internal commentary onto paper. The three-per-page handout is the layout people confuse with it, and it rules blank lines beside each slide for the reader's own notes rather than printing yours. Strip or review notes as a step in the send checklist, not as a habit you rely on remembering.

6. File craft: names, weight, fonts, and the ceiling somebody else set

Four decisions determine whether a file is usable by anyone other than you on the day you made it.

Naming. A file name is a sort key, and the default habit produces a sort order that is wrong. Take eight versions of a board pack named the way they usually get named: boardpack.pptx, then boardpack_v2.pptx, then boardpack_final.pptx, boardpack_final_v2.pptx, boardpack_FINAL.pptx, boardpack_final_CFOedit.pptx, boardpack_use_this_one.pptx, and boardpack_final_FINAL_2.pptx. Those eight names produce 28 orderable pairs. Sorted alphabetically the way a file browser sorts them, 10 of those 28 pairs are in the wrong order, which is 35.71% wrong and 64.29% right. Rename the same eight as cedarridge-boardpack_2026-02-03_v01.pptx through cedarridge-boardpack_2026-02-20_v08.pptx and the wrong-order count is 0. The convention that produces that result is short: subject, then an ISO date as YYYY-MM-DD, then a zero-padded version number, all lowercase, no spaces. The date sorts correctly because ISO order is alphabetical order, and the zero padding is what keeps v10 from sorting before v02.

Weight. A deck's size is almost never in its slides. Decompose the 48.70 MB file from the opening: photographs 38.70 MB, screenshots 5.52 MB, an embedded video 2.80 MB, the slide XML with the theme and the media map 1.14 MB, and logos 0.54 MB. As shares, photographs are 79.47%, screenshots 11.33%, video 5.75%, XML and theme 2.34%, logos 1.11%. The top two categories carry 90.80% of the file and the bottom three carry 9.20%. Any repair that does not start with the photographs is decoration.

The number that decides how much you can remove is effective resolution: pixels divided by the inches the picture occupies on the canvas. A 4032-pixel-wide photograph placed full-bleed across a 13.333-inch widescreen slide is 302.4 pixels per inch. A 1080p projector renders that full-bleed slide at 1920 pixels, so it needs exactly 144 pixels per inch, and every pixel above that is invisible in the room. Resampling to 150 ppi requires 2000 pixels across, a linear ratio of 0.4960 and an area ratio of 0.2460, which removes 75.40% of the pixels. The photographs fall from 38.70 MB to 9.52 MB.

Fonts. If a recipient does not have the font a deck uses, the text is rendered in a substitute and the layout reflows: line breaks move, a two-line title becomes three, and the slide you spent an hour aligning arrives broken. Embedding solves it and costs bytes. Embedding only the characters actually used costs 0.42 MB on this deck; embedding the full character set costs 3.10 MB, which is 2.68 MB for the ability of a recipient to edit the file in a font they do not own. Embed the subset when you are sending a deck to be read, and the full set only when someone must edit it in your typeface. Font embedding is also the one feature in this list whose availability has historically differed between the Windows and macOS builds (checked September 2026; verify in your own installation before relying on it for an external send).

The ceiling. Somebody else sets the limit and you find out at the worst moment. Corporate mail systems commonly reject attachments somewhere between 20 and 25 MB, and the exact number is set per organisation (checked September 2026; verify with your own administrator rather than assuming). Design for the ceiling rather than discovering it: resample images once at the end, link large video rather than embedding it, and check the file size before the send rather than after the bounce.

7. Opening a deck somebody else built

The first ten minutes on an inherited deck are worth more than the next two hours, and the rule for them is to change nothing. Read it before you touch it.

Three global properties tell you most of what you need: the number of slide masters in the file, the number of distinct fonts in use, and the canvas ratio the content was built for. A file with three masters and seven fonts is a merged file, and merged files behave in ways that look like bugs until you know what they are. A file whose pictures were sized for a 4:3 canvas and pasted onto a 16:9 one carries a specific, visible distortion that no amount of nudging will fix.

Then read the Slide Sorter for shape and the Outline for argument, exactly as you would on your own deck. You now know where the deck is dense, whether it has sections, what share of its titles the Outline can see, and how many of those titles are claims.

Only then start fixing, and rank the repairs before you begin. The useful ranking is slides fixed per minute, because it separates a change that propagates from a change that does not. Re-pointing seven fonts to two theme fonts touches every slide in the file and costs about fourteen minutes, which is over three slides a minute. Retyping fifteen stray titles into their placeholders touches fifteen slides and costs about twenty minutes, which is three quarters of a slide a minute. Both are worth doing when you own the deck. Only the first is worth doing when you have forty minutes before a meeting.

Two repairs are almost always wrong under time pressure. Restyling the master is the first, because a change to type, placeholders or the background propagates to every slide including the ones you have not looked at, and you will discover the side effects during the meeting. Merging duplicate masters is the one exception worth knowing, and only where their layouts already match, because that merge changes which master a slide points at rather than what the master says. Even then you check the result in Slide Sorter before committing, and you leave the masters alone on a deck you are handing back to its owners. Retyping numbers is the second. If a figure looks wrong, note it and ask; a deck you inherited is a deck whose numbers you cannot yet defend, and silently correcting one is how a wrong number becomes your wrong number.

The 4:3 problem deserves its own arithmetic, because both available fixes are bad and you should choose the bad one deliberately. A legacy 4:3 canvas is 10.0 by 7.5 inches; the widescreen canvas is 13.333 by 7.5. The 4:3 area is exactly 0.75 of the 16:9 area. Fit the old content by height and it keeps its proportions, occupies 10.0 inches of a 13.333-inch canvas, and leaves 3.3333 inches of dead canvas, which is 25.00% of the width. Fill the width instead and everything stretches horizontally by a factor of 1.3333: a 2.00-inch logo becomes 2.6667 inches wide at unchanged height, and a circle becomes an ellipse 33.33% wider than it is tall.

Height-fitting is the right default. Dead canvas is a design problem you can solve; a stretched logo is a legal and credibility problem you cannot, because somebody's brand mark is now the wrong shape in a document with your name on it.

8. The Slide Sorter as a rehearsal instrument

Slide Sorter is usually taught as a place to drag slides around. Its more valuable use is as the cheapest rehearsal available, and it takes four minutes.

Zoom out until the whole deck fits on one screen. You are now looking at the deck the way a reader flicking through it will, which is at a size where words are unreadable and only structure and density register. Three faults show up immediately at that size and are nearly invisible at working size. A run of five slides that look identical, which means five slides that will feel like one. A slide with no visual anchor at all, which is a wall of text wearing a title. And a density break in the wrong place, where the deck goes from airy to crowded exactly where the argument was supposed to get simpler.

Then do the timing arithmetic, which is where sections earn their keep a second time. Assign each section a minute budget from the meeting length, count its slides, and divide. A section with nine slides and a four-minute budget is asking you to speak at 26.7 seconds a slide, and nobody makes an argument at that rate. The fix is upstream, in the section, not in your delivery.

Hidden slides belong here too. Hiding a slide keeps it in the file, out of the show, and available if a question demands it, which is exactly what an appendix is for. The honest use is a slide you expect one specific person to ask for. The dishonest use is hiding twelve slides you could not bear to delete, which leaves a file that is still 48 MB and still takes forty minutes to review.

One last rehearsal move. Collapse every section, then expand one at a time and say that section's claim out loud before looking at its slides. If you cannot state the claim from the section name, the audience will not be able to reconstruct it from the slides.

9. Printing, and why the handout is a second artefact

A printed deck behaves as a different object from the projected one, with different physics, and the physics is arithmetic you can do before you print anything.

Start with pages. Thirty slides print on 30 pages at one per page, 15 at two, 10 at three, 8 at four, 5 at six and 4 at nine per page. That last column is the whole appeal of an N-up handout, and the cost is hidden in the scale factor.

Compute the scale factor, which is the number that converts an inch on the slide to an inch on paper. A4 is 8.2677 by 11.6929 inches; at half-inch margins the printable area is 7.2677 by 10.6929. Allow a 0.2-inch gutter between cells, which is what the layouts leave, because the gutter changes the answer and an unstated one is why two people computing the same handout get different numbers. A two-up layout then gives each slide a cell of 7.2677 by 5.2465 inches, and fitting a 13.333 by 7.5 slide into that cell means a scale factor of 0.5451. A three-up layout, which puts the slides in a column 55% of the printable width and rules blank note lines beside them, gives 0.2998. A six-up grid of two columns by three rows gives cells of 3.5339 by 3.4310 inches and a scale factor of 0.2650.

Type shrinks by exactly that factor, and the numbers are unforgiving.

Set 9 pt as the floor for comfortable reading on paper and invert the arithmetic. To clear 9 pt on paper, on-slide type must be at least 16.511 pt at two-up, 30.021 pt at three-up-with-note-lines and 33.957 pt at six-up. A deck whose smallest type is 11 pt, which is an ordinary chart axis label, prints at 5.996 pt on a two-up and 2.915 pt on a six-up. The second of those has stopped being small text and become a grey smudge that happens once to have been text.

That is why the handout is a separate artefact rather than a print setting. If the deck must be read on paper at six-up, the charts need larger labels, fewer series and a bigger title, and those are changes to the slide rather than to the printer dialog. The alternative, which is often the right answer, is to accept two-up, spend the extra pages, and stop pretending a six-up handout is readable.

Two production details save a reprint. Print a grayscale proof before printing colour, because a chart that distinguishes its series by hue alone becomes four identical grey lines, and the fix is a marker or a direct label rather than a different palette. And check what the chosen layout carries before you print it, because the notes-page layout puts your speaker notes on the page while the three-up handout rules blank lines instead, and a notes-heavy deck printed from the wrong one hands internal commentary to a client.

10. Autosave, recovery, and the file that will not open

Two different features get called autosave and they behave differently, which is why people are surprised by both.

AutoRecover is the local one. It writes a recovery copy at an interval, ten minutes by default, and offers it back after a crash. It is not a save; the recovery copy disappears once you close the file normally, and it recovers to the last interval rather than to the last keystroke. AutoSave is the cloud one. It applies to files stored in a synchronised location, writes continuously, and keeps a version history you can roll back through. Both the default interval and the exact behaviour of the two features move between releases (checked September 2026; verify in your own installation, and check where your organisation actually stores files, because AutoSave does nothing for a file on a local drive).

Model the exposure. If a crash arrives uniformly at random between saves, the expected work lost is half the interval: 5.0 minutes at a ten-minute interval, 2.5 at five minutes, 1.0 at two. Now apply a crash rate. A six-hour build at one crash per nine editing hours has 0.6667 expected crashes, so the expected loss over the whole build is 3.3333 minutes at a ten-minute interval, 1.6667 at five and 0.6667 at two. Tightening from ten minutes to two saves 2.6667 expected minutes across an entire deck.

Read that number honestly, because it argues against the thing it looks like it argues for. Two and a half minutes is not why you save. The expected value is small and the variance is what hurts. The average is not the quantity to plan against. The tail is, where a crash lands twenty minutes into a restructure you cannot reconstruct from memory. That is a versioning problem rather than an interval problem, and the fix is a milestone copy before any structural change, named with the date and version so it sorts where it belongs.

When a file will not open, there are three moves in order. Use the application's own open-and-repair option, which recovers the majority of damaged files. Try opening the file in a different application, because a competing renderer is often more forgiving about the specific part that is corrupt. And if both fail, treat the file as what it physically is. A modern PowerPoint file is a zip container holding XML parts and a media folder. Rename a copy to a zip extension, open it, and you recover the images, the embedded video and often the raw slide text even when the deck itself will not load. Work on a copy, always, because the archive route is a salvage operation and not a repair.

11. Templates and themes, which are not the same thing

A theme is a set of formatting decisions expressed as roles: a heading font and a body font, a palette of named colours, and a set of effects. Changing the theme restyles everything in the deck that was built out of those roles, and changes nothing that was built out of hard-coded values. That sentence contains the entire practical difference between a deck that can be restyled in ten seconds and one that takes an afternoon.

A template is a file. It carries a theme, plus the slide master and layouts, plus in most cases a set of starter slides and a title slide with the organisation's marks already placed. You open it and it produces a new deck rather than opening itself for editing, which is what keeps a template from slowly becoming a deck.

Using one well requires three habits and no theory. Start from the template rather than from a blank deck and applying the theme afterwards, because the layouts come with the template and not with the theme. Choose the layout that matches your content, from the layout gallery, rather than taking the closest-looking one and dragging boxes around on it. And pick colours from the theme palette by role rather than by opening the colour picker, because a colour chosen from the picker is a hard-coded value that will survive every future restyle as a stain.

Where the treatment stops is the point where you would open the master and change it. What lives at master level, what belongs in a layout, how a placeholder differs from a text box you drew, and how to repair a deck that fights its own master, are all the first mechanics node's material and it needs ten hours of its own. Recognising the symptom is enough for now, and the symptom is simple: if changing one thing in one place should have changed twenty slides and did not, you are looking at a master problem, and reaching for it slide by slide is the expensive direction.

12. Accessibility as a build habit rather than a final pass

Three accessibility properties cost nothing while you are building and cost a rebuild afterwards. All three are checkable with arithmetic.

Reading order is the sequence in which assistive technology announces the objects on a slide, and it is the order in which those objects were added rather than the order in which a sighted reader would see them. The Selection pane shows that stored order and lets you change it, and pressing Tab walks it, which makes it free to check. Take a slide with nine objects: a title, three KPI figures, a chart, a callout, a footnote, a logo and a page number. A sighted reader reads them in that order. If the stored order is chart, logo, title, third KPI, callout, first KPI, page number, second KPI, footnote, then of the 36 orderable pairs, 16 are announced out of order, which is 44.44% of them, and two objects are announced before the title. A listener meets the chart and the logo before learning what the slide is about.

Alt text is a sentence attached to a non-text object. On the same nine-object slide, five objects are content and need alt text; two are decorative and should be marked as such so they are skipped rather than described. The rule that makes alt text useful rather than a compliance artefact is that a chart's alt text states the chart's message, not its type. "Bar chart of revenue by state" describes the picture. "Maharashtra is the largest state at 32.2% of revenue, ahead of Karnataka at 22.5%" describes what the picture is for, and it is also the sentence that should already be in the title.

Contrast is a computable ratio between the relative luminance of foreground and background. The thresholds in force are 4.5 to 1 for normal-size text and 3 to 1 for large text, where large means at least 18 pt or 14 pt bold (WCAG 2.2, checked September 2026; verify, because guideline versions advance and the thresholds can move). Eight pairs, computed:

Two lines in that table are the ones that change behaviour. The habit of setting captions and footnotes in a light grey lands at A6A6A6, which is 2.434 and fails both thresholds; moving to 595959 gives 7.005 and passes both, and nobody looking at the slide will describe it as less subtle. And a brand amber on white is 2.009, which means a headline number set in the brand colour is illegible to a meaningful share of the room, while the same amber on the brand navy reaches 4.311 and is fine at title size and not at body size.

One more habit costs five seconds a slide. Give every slide a unique title, even where the title is hidden from view, because slide titles are how assistive technology navigates a deck and a deck of untitled slides is a deck that cannot be skimmed by anyone using one.

13. The same ideas in the other tools

Nothing above is specific to one product. The vocabulary differs, two of the concepts have no direct equivalent, and knowing which is which stops you hunting for a control that does not exist.

Feature names and availability in that table move with releases (checked September 2026; verify in the build in front of you). The two rows worth memorising are the first two, because sections and the Outline are the two habits taught here that a learner working in a browser tool has to reproduce by hand: a naming convention on the slides themselves in place of sections, and a written title list in place of the Outline.

``ai-augment-json { "skill": "Auditing an inherited deck's structure and file weight before touching a single slide", "use": "After you have written your own diagnosis by hand from the Slide Sorter and Outline, hand an assistant the extracted title list, the section boundaries and a table of the file's components with their sizes, and ask it to rank the repairs by slides fixed per minute and to name the three it would skip. Use the ranking as a challenge to yours; the repair order you execute is the one you can defend.", "tools": ["Chat assistants (Claude, ChatGPT, Gemini)", "A spreadsheet for the file-weight table", "The presentation tool's own Outline view, for the title list"], "prompt": "Here is the title list of a 34-slide deck, in order, with a topic tag on each: <LIST>. Here are its file components in MB: photographs <A>, screenshots <B>, video <C>, logos <D>, slide XML and theme <E>. Do four things. 1) Count the adjacent topic switches and say what the count would be if the slides were grouped into contiguous sections by topic. 2) Tell me how many slides would have to move to reach that grouping. 3) Rank the file-weight repairs by megabytes recovered, and say which single repair gets me under 25 MB. 4) Name the three repairs on this deck you would NOT do under a forty-minute deadline, and why. Do not rewrite any title.", "verify": "Recount the topic switches yourself from the title list before accepting any of it: an assistant will confidently report a switch count it did not compute, and the count is a two-minute manual check. Recompute every megabyte figure from pixels and placed inches rather than from the assistant's arithmetic, because the compression saving is an area ratio and a linear ratio quoted in its place understates the saving by roughly the square. Then open the file and confirm the section boundaries the assistant proposed actually fall between slides rather than inside a run that has to stay together. Any repair ranking that does not state the minutes it assumes is not a ranking.", "diy": "You must be able to open a stranger's deck cold, read its Slide Sorter and Outline, and say within ten minutes how many masters and fonts it contains, what share of its titles the Outline can see, where its sections should fall and where its megabytes are. The gate rewards that because it is the skill that transfers: the assistant sees the list you extracted, and extracting the list correctly is the whole diagnosis. An analyst who cannot do the audit by hand cannot tell a good repair ranking from a plausible one.", "market": "IN" } ``

Common mistakes & how experts think differently

1. Deleting slides to fix a file size. The instinct is to treat size as a content problem. In the deck audited here, the slide XML with the theme and the media map is 1.14 MB of a 48.70 MB file, which is 2.34%. Deleting half the slides removes almost nothing and costs the argument. An expert opens the size problem as an inventory question, ranks the components, and finds that two of the five carry 90.80% of the weight.

1. Treating a resampling target as a quality setting. Picking "high quality" or "email quality" from a dropdown is picking a number without knowing what it has to survive. The number that matters is pixels divided by placed inches, compared against the pixels the display can actually show. A full-bleed image on a widescreen canvas needs 144 ppi to match a 1080p projector exactly and 288 ppi to match a 4K one, so 150 ppi is a deliberate choice with 4.17% of headroom over 1080p and 52.08% of what a 4K projector could use. An expert states the room first and the setting second.

1. Fixing the alignment before fixing the filing. A deck with fifteen floating text boxes and no sections can be made to look immaculate, slide by slide, in about three hours. It will still be unreorderable, unsearchable and invisible to the Outline. The order that pays is structure, then content, then geometry, because geometry work on a slide that later merges or moves is work thrown away.

1. Believing the Outline is a place to write. Typing a deck in the Outline produces title-plus-bullets on every slide, because the Outline can express nothing else. The tell is a deck where every slide has the same shape. An expert uses the Outline in the read direction, as a check that the titles make an argument, and builds the actual slide where a chart or one sentence can be the entire content.

1. Naming files so that the current one is findable only by memory. Eight ad-hoc names put 10 of their 28 orderable pairs in the wrong order, and the person who has to find the current version is usually not the person who named the files. Dates in an ISO format sort correctly because alphabetical order and chronological order coincide, which is the only property a naming convention needs.

1. Printing the deck instead of building the handout. Six-up looks efficient until you compute the scale factor of 0.2650 and discover that a standard 11 pt chart label prints at 2.915 pt. The expert decision is made before the design: either the deck will be read on paper, in which case the charts are built for a scale factor, or it will not, in which case print two-up and spend the pages.

1. Treating accessibility as a compliance pass at the end. Reading order is the order objects were created, so it is free to get right while you build and expensive to repair afterwards on forty slides. A deck built by dragging boxes onto slides can easily announce 44.44% of its object pairs in the wrong order, and no visual inspection will reveal it.

1. Reaching for the master under time pressure. A master change propagates everywhere, including to the slides you have not opened, and you find out during the meeting. The disciplined move on an inherited deck is to rank repairs by slides fixed per minute, take the propagating fixes that are safe, and leave the structural rebuild for a day when the deck is not due.

1. Confusing the theme with the template. A theme is a set of roles; a template is a file that carries a theme plus masters, layouts and starter slides. Applying a theme to a deck built from a blank file gives you the colours and not the layouts, which is why the result usually looks like a themed blank deck.

1. Trusting autosave to be versioning. The recovery copy that a crash returns is not a version history, it disappears on a clean close, and on a locally stored file the continuous cloud save is not running at all. The expected loss from a ten-minute interval over a six-hour build is 3.3333 minutes; the loss that actually hurts is a structural change you cannot reconstruct, which no interval prevents and a milestone copy does.

Worked examples

Every figure below was computed before it was written down, and the Verify block at the end of this section restates each one at full precision. All companies are synthetic composites; no slide content is edited in any example that says no slide content was edited.

Worked example 1: Thirty-four slides, four sections, nothing rewritten (India, ₹)

The setting. Kaveri Kitchens (synthetic) runs a cloud-kitchen network across Bengaluru and Hyderabad and is preparing a Series A conversation. Its founder has been adding to the same investor update for six months and sends you the file on a Monday, asking whether it is ready for Thursday. Two constraints come with it. The numbers have been reviewed and the wording has been through counsel, so you may not change the content of any slide. And you have one working session, not a week.

The deck is 34 slides. Tag each one with its topic and the order it arrived in looks like this.

The wrong version. The founder's instinct is to fix the deck slide by slide, starting with the ones that look worst. Three measurements say that is the wrong order of work.

The topic changes between 30 of the 33 adjacent pairs, a switch rate of 0.9091, and the longest unbroken run on a single topic is 2 slides. A reader re-orients on almost every page. The four slides carrying the ask sit at positions 13, 23, 29 and 33, which means the request the whole document exists to make is spread across two thirds of the deck.

The Outline view says how the deck was built. Twenty-six of the 34 titles are in the title placeholder, so 76.47% of the deck shows a title there and 8 slides show nothing. Nineteen of 34 bodies are in a content placeholder, so 55.88% of the body text reaches the Outline and 15 slides worth of text does not. Of the 26 titles the Outline can see, 4 are full sentences with a verb, which is 11.76% of the deck and 15.38% of the visible titles; the other 22 are labels.

The founder also has a summary exhibit for the restructure conversation, and it is drawn as a five-slice pie with a legend down the right side. Five slices is two more than a pie can carry, the legend forces the reader to bounce between colour and label, and the one thing the exhibit is supposed to show, which is that two topics dominate, is exactly what angle comparison is worst at.

The corrected version. Do the structural work first, in the Slide Sorter, and touch no slide's content.

Group the 34 slides into a cover plus four sections in the order Business, Financials, Product, Asks. The switch count falls from 30 to 4, a reduction of 26 switches or 86.67%, and the switch rate falls from 0.9091 to 0.1212. Sixteen slides are already in an order consistent with that grouping, which is the longest non-decreasing run, so 18 slides have to move, or 52.94% of the deck.

The session costs 15.5 minutes on a stopwatch: 360 seconds reading the Slide Sorter without editing, 272 seconds tagging 34 slides at 8 seconds each, 100 seconds creating 4 sections at 25 seconds each, and 198 seconds dragging 18 slides at 11 seconds each. Words of slide content edited: zero.

Redraw the summary exhibit as a sorted bar: five categories on the vertical axis, slide count on the horizontal, sorted descending, with the two largest bars in the accent colour and the other three in grey. The action title states the finding rather than the subject: "Financials and Business carry 23 of the 34 slides, and the four ask slides are scattered from 13 to 33." A sorted bar is the right form here. The message is a comparison across categories with no intrinsic order, and sorting by value is part of the claim rather than a decoration on it.

What the restructure did not do is worth saying out loud. The deck is still 34 slides long, still has 22 label titles, and still has 15 slides whose body text is invisible to the Outline. Sectioning made the shape of the argument visible and made the deck reorderable in one drag per section. It did not make the argument good, and the Thursday answer is that the deck is now honest about its own problem: the ask needs to move to a single section near the front, and 22 titles need rewriting, which is the next session's work.

The checklist to apply.

1. Tag every slide with one topic before moving anything, and count the adjacent switches.
2. Read the Outline and record two shares: titles in placeholders, and bodies in placeholders.
3. Count the titles that are sentences with a verb. Report it as a share of visible titles, not of slides.
4. Name each section for the claim it carries, and merge any section whose claim you cannot state in a clause.
5. Record the number of slides that had to move; it is the honest measure of how far the deck was from its own structure.
6. Change no content in the structural session, so that the two kinds of work can be reviewed separately.

Worked example 2: 48.70 MB into a 25 MB mailbox (US, $)

The setting. Lakeshore Analytics (synthetic) delivers an operations readout to a US hospital group. The deck carries 18 site photographs at 2.15 MB each, 6 interface screenshots at 0.92 MB each, 3 logo files at 0.18 MB each, one embedded 2.80 MB video clip, and 1.14 MB of slide XML, theme and media map. The file is 48.70 MB. The client's mail system rejects attachments above 25.0 MB, so the deck is 23.70 MB over, at 1.948 times the ceiling. It has to arrive as an attachment because the client's security policy blocks external file-sharing links, and it will be projected on a 1080p room projector.

The wrong version. The team deletes eight slides and re-exports. The file barely moves, because slide XML with the theme and the media map is 1.14 MB of 48.70, which is 2.34%, and none of the deleted slides carried a photograph. Somebody then zips the file, which also does almost nothing: a modern PowerPoint file is already a zip container, and JPEG photographs are already compressed, so a second pass over compressed data recovers close to nothing.

The team's own diagnostic exhibit is part of the problem. It shows the five components as a 3D pie, tilted, with the smallest two slices behind the tilt. The 3D projection makes the front slices look larger than they are, which is a distortion introduced by the drawing rather than by the data, and it is being used to decide where to spend an hour.

The corrected version. Inventory first, sorted, with cumulative shares.

Drawn as a sorted horizontal bar with the top two bars accented, and titled "Photographs and screenshots carry 90.8% of the 48.70 MB file", the exhibit answers the only question anyone has. A bar chart is the form for a comparison across five categories, sorting is what makes the top-two claim visible, and the cumulative column belongs in the table rather than as a second line on a second axis.

Now the arithmetic that sets the compression target. Each photograph is 4032 by 3024 pixels, which is 12.19 megapixels, placed full-bleed across the 13.3333-inch canvas, so it sits at 302.4 pixels per inch. The room's projector renders that full-bleed slide at 1920 pixels, which is exactly 144 ppi across the canvas. Resampling to 150 ppi needs 2000 pixels across, a linear ratio of 0.496032 against the original width and an area ratio of 0.246047, which removes 75.40% of the pixel count. The photographs fall from 38.70 MB to 9.522 MB.

The screenshots are 2560 by 1440 placed 8.00 inches wide, so 320.0 ppi. At 150 ppi they need 1200 pixels, a linear ratio of 0.46875 and an area ratio of 0.219727, taking 5.52 MB to 1.2129 MB.

The video is replaced by a link with a still frame on the slide, removing 2.80 MB, and the deck is one that will be read as often as it is shown. Fonts are embedded as a used-characters-only subset at 0.42 MB rather than the full character set at 3.10 MB, saving 2.68 MB for a file nobody outside the team will edit.

The file lands at 12.8349 MB, a reduction of 35.8651 MB or 73.64%, with 12.1651 MB of headroom under the ceiling and a size that is 51.34% of what the mailbox allows.

Verify it by an independent route rather than by looking at the pictures. A full-bleed image on that projector receives 1920 pixels. The compressed file supplies 2000, which is 80 pixels more than the room can render and a supplied-over-needed ratio of 1.041667. The compression therefore cannot be visible in that room, and saying so is stronger than saying the images still look fine.

State the limits in the same breath, because both bite later. On a 4K projector the same full-bleed slide needs 3840 pixels; the 150 ppi file supplies 52.08% of that, a shortfall of 1840 pixels, so a deck built for a 4K room needs 288 ppi and a different size budget. And if the client later prints the pack at 220 ppi, the photographs alone come to 20.483 MB and the deck to 25.192 MB, which is over the 25.0 MB ceiling. One resampling pass cannot serve a projector and a printer, and choosing which one it serves is a decision, not a default.

The checklist to apply.

1. Inventory the file by component before changing anything, and sort the inventory.
2. Compute effective resolution as pixels divided by placed inches, for each class of image.
3. Compute the pixels the display can actually render, and set the target from that rather than from a quality label.
4. Convert a linear ratio to an area ratio before quoting a saving; the saving goes as the square.
5. Link video rather than embedding it whenever the file has to travel.
6. Embed fonts as a subset unless somebody must edit the file in your typeface.
7. State the two limits explicitly: the higher-resolution room, and the print run.

Worked example 3: The handout that could not be printed six-up (India, ₹)

The setting. Nagarjuna Cement (synthetic) is holding an analyst day. The investor-relations team has a 30-slide pack and wants it printed for the room on A4, six slides to a page, double-sided, to keep the paper down. On 14 of the 30 slides the smallest type is 11 pt, which is where the chart axis labels and the footnote sources sit. That is 46.67% of the pack.

The wrong version. Print it six-up and hand it out. The page count is attractive: 5 pages, or 3 double-sided sheets, against 30 pages at one per slide.

Then compute the scale factor. A4 is 8.2677 by 11.6929 inches, and at half-inch margins the printable area is 7.2677 by 10.6929. A two-column, three-row grid with a 0.2-inch gutter between cells gives each slide a cell of 3.5339 by 3.4310 inches, and fitting the 13.3333 by 7.5 inch slide into that cell is a scale factor of 0.265039.

Everything on the slide shrinks by that factor. A 40 pt title prints at 10.602 pt. An 18 pt body line prints at 4.771 pt. The 11 pt chart labels on 14 of the 30 slides print at 2.915 pt. Inverting the arithmetic against a 9 pt floor for comfortable reading, on-slide type must be at least 33.957 pt to survive a six-up handout, which no chart label has ever been.

The handout has gone past small and become unreadable, and it will be unreadable in a room of people who came to read the numbers.

The corrected version. Treat the handout as a second artefact with its own type floor, and choose the layout from the arithmetic rather than from the paper budget.

Two-up is the only layout on which ordinary slide type survives, and even there the 11 pt labels print at 5.996 pt and fail. So the handout gets built rather than printed: on the 14 offending slides, the smallest type is raised to 18 pt, which prints at 9.811 pt two-up and clears the floor. The pack prints on 15 pages, or 8 double-sided sheets, against 3 sheets for the unreadable version. Five extra sheets a head is the price of a handout anyone can read.

Raising a chart label from 11 pt to 18 pt has a consequence that is the real lesson. A six-series line chart with a legend does not fit at 18 pt, so the chart has to change form. The repair is six small panels sharing one vertical scale, each with its series named directly at 18 pt inside the panel, which removes the legend, removes the colour lookup and makes the six comparisons available at a glance. Small multiples are what a legible handout forces on you, which is one reason handout-first decks tend to have better charts.

Two production checks close the job. Print one grayscale proof, because a chart that separates its series by hue alone collapses into identical grey lines on a black-and-white office printer, and the fix is direct labels rather than a different palette. And check which layout prints the speaker notes, because the notes-page layout does and the three-up handout does not, ruling blank lines for the reader instead. Print this pack from the wrong one and a notes-heavy deck puts internal commentary onto a page you are about to hand to an analyst.

The checklist to apply.

1. Decide before designing whether the deck will be read on paper, because the answer changes the chart.
2. Compute the scale factor of the layout from the printable area, not from the number of slides per page.
3. Set a legibility floor on paper, then divide to get the minimum on-slide type size.
4. Check the smallest type on every slide, not the body text, because the floor is set by axis labels and footnotes.
5. When the floor forces a chart to change form, change the form rather than the floor.
6. Proof in grayscale, and check what the chosen layout does with the notes.

Worked example 4: Twenty operations, on a stopwatch (US, $)

The setting. Harborline Capital (synthetic) is a US credit fund whose associate rebuilds a standard twenty-slide investment-committee deck from an outline and a folder of exhibits about three times a week, across roughly 46 working weeks. Decomposed into discrete operations, one rebuild is 174 actions. Of those, 158 have a keyboard route and 16 do not, because inserting a chart, choosing a picture off disk and typing the words are pointing-and-typing work. Twenty distinct operations therefore cover 90.80% of the task.

The wrong version. The associate does everything from the ribbon and the mouse, at about 3.4 seconds an operation once the eye movement, the travel and the click are counted. The task takes 537.2 seconds of pure operation time, which is 8.95 minutes on top of every minute spent thinking, and the conclusion drawn from the experience is that the software is slow.

The productivity exhibit the team builds to argue for training is drawn as a 3D column chart of all twenty operations in alphabetical order. Alphabetical order hides the only pattern in the data, the 3D depth makes the front columns look taller, and a reader cannot tell from it which operations matter.

The corrected version. Sort by frequency and look at the shape.

Six operations account for 74 of the 158 keyboard operations, which is 46.84% of the total. They are undo at 15, nudging an object at 14, creating a new slide at 12, opening or closing the notes pane at 12, saving at 11, and bold at 10. Those six are exactly the operations that appear ten or more times in one rebuild. Learn them first, and half the benefit arrives in the first half hour.

Drawn as a sorted horizontal bar of the twenty operations by frequency, with the top six accented and the remaining fourteen in grey, the exhibit carries its own title: "Six of the twenty operations carry 46.8% of the keystrokes." Bars rather than columns because the labels are long, sorted because the ranking is the message, one colour of emphasis because the message names one group.

Now the payback. At about 0.9 seconds a keyboard operation, the task runs 142.2 seconds, or 2.37 minutes, against 537.2 seconds by ribbon. The saving is 395.0 seconds a run, which is 73.53% of the slower route, and the keyboard route is 3.777778 times faster on the covered operations. Six minutes of deliberate practice per operation is 2.0 hours for all twenty, which is repaid after 18.23 runs. At three rebuilds a week over 46 weeks the annual recovery is 54,510 seconds, or 15.14 hours.

Read the limits, because a productivity number quoted without them is a sales figure. The 3.4 and 0.9 second costs are assumptions, not measurements of this associate, and the whole result scales with their ratio. The 16 uncovered operations are 9.20% of the task and include every step that requires a decision, so no shortcut touches them. And the 15.14 hours a year is recovered in three-minute pieces spread across 138 rebuilds, which is real but is not a free afternoon. The honest claim is narrower and still worth making: the mechanical part of deck-building can be made roughly four times faster for two hours of practice, and the part that was actually slow was never the mechanics.

The checklist to apply.

1. Count the operations in the task before optimising any of them.
2. Sort the operation list by frequency and learn the top six first.
3. State the seconds-per-operation assumption on both routes, since the whole result is their ratio.
4. Compute payback in runs of the task, not in weeks, because runs are what you control.
5. Name the share of the task that no shortcut can reach, and stop claiming it.
6. Never sort a ranking chart alphabetically when the ranking is the message.

Worked example 5: Forty minutes with somebody else's forty-six slides (India, ₹)

The setting. Meridian Logistics (synthetic) moves freight between Chennai and the northern corridor. The analyst who built its quarterly operations review has left, and the deck has landed with you 40 minutes before the review starts. It is 46 slides. A cold read of the file's properties gives you three numbers before you open a single slide: 3 distinct slide masters, 7 distinct fonts, and 12 slides whose content was built for a 4:3 canvas and pasted onto a 16:9 one. Reading the Slide Sorter adds two more: 9 slides have an object hanging outside the printable area, and 31 of the 46 slides have their title in the title placeholder, which is 67.39%.

The wrong version. Start with the slides that look worst. The 12 stretched slides are the most visibly wrong, so they get the attention: re-framing them properly takes about 36 minutes and touches 12 slides. The 40 minutes are gone, the fonts are still seven, the masters are still three, and 34 of the 46 slides have had nothing done to them at all.

The corrected version. Rank the available repairs by slides fixed per minute before starting any of them, because that ratio separates a change that propagates from a change that does not.

Drawn as a sorted bar of the last column, with the top two accented, the exhibit titles itself: "Two repairs reach all 46 slides in 36 minutes; the other three reach 36 in 74." All five repairs together cost 110 minutes and touch 128 slide-instances. The first two cost 36 minutes, which is 32.73% of the total repair time, and capture 71.88% of the slide-instances. Inside a 40-minute window, those two are the whole plan, and the remaining four minutes go to writing down what you did not fix so you can say it out loud if asked.

The 4:3 slides are the interesting refusal, because both available fixes are bad and the choice has to be deliberate. The old canvas is 10.0 by 7.5 inches against 13.3333 by 7.5, so its area is exactly 0.75 of the widescreen canvas. Fit the old content by height and it keeps every proportion, uses 10.0 inches of width, and leaves 3.3333 inches of dead canvas, which is 25.00% of the slide. Fill the width instead and everything stretches horizontally by 1.333333: a 2.00-inch logo becomes 2.6667 inches at unchanged height, and a circle becomes an ellipse whose width is 1.333333 times its height.

Height-fit, always, and take the dead band. A margin of empty canvas is a design problem with several solutions. A client's logo rendered 33.33% too wide is a different kind of problem, and it is on a document with your firm's name at the bottom.

One repair stays off the list under time pressure even though it is tempting, and consolidating three masters into one is on the list only because it is the second-highest-yield change available and the file is going to be presented rather than edited. On a deck you will hand back to its owners, leave the masters alone: a master change propagates to slides you have not opened, and you find out which ones during the meeting.

The checklist to apply.

1. Read three file-level properties before opening any slide: masters, fonts, and the canvas ratio the content was built for.
2. List the repairs with their minutes and the slides each one touches, then divide.
3. Do the propagating repairs first, and only those, when the window is short.
4. Height-fit legacy content; never width-fill anything containing a logo, a circle or a face.
5. Write down what you did not fix, before the meeting, so it is a stated limit rather than a discovered one.
6. Do not retype a number you cannot defend.

Worked example 6: The board pack that existed eight times (US, $)

The setting. Cedar Ridge Partners (synthetic) builds a board pack over three weeks with three contributors. Eight successive versions exist. They were named the way files usually get named, and the versions arrived in this order: boardpack.pptx, boardpack_v2.pptx, boardpack_final.pptx, boardpack_final_v2.pptx, boardpack_FINAL.pptx, boardpack_final_CFOedit.pptx, boardpack_use_this_one.pptx, boardpack_final_FINAL_2.pptx. The pack is built across roughly 6 hours of editing, and the firm's laptops crash about once every 9 editing hours.

The wrong version. Trust the file browser to show the newest first, and trust autosave to protect the work.

Eight names produce 28 orderable pairs. Sorted the way a file browser sorts, which is case-insensitive alphabetical, the eight land like this.

Ten of the 28 pairs are in the wrong order, which is 35.71% wrong and 64.29% right. A person looking for the current version has a better than one in three chance of picking wrongly from any pair they compare, and the two names that shout loudest, boardpack_FINAL.pptx and boardpack_use_this_one.pptx, are the fifth and seventh versions rather than the eighth.

Two numbers do not need a chart, and the comparison here is exactly two numbers. Saying "64.29% of pairs ordered correctly against 100%" in a sentence is faster to read than any picture of it, which is a rule worth carrying into every deck: a chart with two bars is a sentence wearing a costume.

The corrected version. Rename to a convention that sorts, and separate versioning from recovery.

Rename the same eight to cedarridge-boardpack_2026-02-03_v01.pptx through cedarridge-boardpack_2026-02-20_v08.pptx: subject first, then an ISO date as year, month, day, then a zero-padded version, all lowercase with no spaces. Sorted alphabetically, 0 of the 28 pairs are out of order, which is 100.00% correct. The ISO date does the work because alphabetical order and chronological order coincide in that format, and the zero padding keeps a tenth version from sorting ahead of a second.

Then size the recovery exposure, which is a different problem. If a crash arrives uniformly at random between saves, the expected work lost is half the interval: 5.0 minutes at a ten-minute interval, 2.5 minutes at five, 1.0 minute at two. Over a 6-hour build at one crash per 9 editing hours there are 0.6667 expected crashes, so the expected loss over the whole build is 3.3333 minutes at ten, 1.6667 at five and 0.6667 at two. Tightening the interval from ten minutes to two saves 2.6667 expected minutes across the entire pack.

That figure is the argument against treating the interval as the safeguard. Two and two thirds minutes is not what anyone is afraid of. The loss that matters sits in the tail, at a crash twenty minutes into a restructure, where the minutes are recoverable and the sequence of decisions is not. The control for that is a milestone copy saved under the naming convention before any structural change, which costs four seconds and bounds the worst case at one milestone rather than one interval.

If a file will not open at all, work through three moves on a copy. Use the application's own open-and-repair option first. Then try a different application, because a competing renderer is often more tolerant of the specific damaged part. If both fail, use the fact that a modern PowerPoint file is a zip container of XML parts and a media folder: renaming a copy to a zip extension and opening it recovers the images, the embedded video and usually the raw slide text, which is enough to rebuild rather than to restart.

The checklist to apply.

1. Name files so that alphabetical order is chronological order: subject, ISO date, zero-padded version.
2. Count the inversions if you doubt it; an ad-hoc scheme puts about a third of its pairs in the wrong order.
3. Keep the recovery interval and the version history as separate controls, because they solve different failures.
4. Save a milestone copy before any structural change, not on a timer.
5. Know the salvage sequence before you need it, and always run it on a copy.
6. Do not draw a chart of two numbers.

Worked example 7: The slide a screen reader reads backwards (India, ₹)

The setting. The Kaveri Kitchens pack from the first example now has to go to an investor who uses a screen reader, and the same file will be read on a phone. One slide carries nine objects: a title, three key-figure callouts, a chart, a highlight callout, a footnote, the company logo and a page number. The deck's palette is a navy 1F4E79, a brand amber F2A900, an alert red C00000 and three greys at 595959, 7F7F7F and A6A6A6, all on a white FFFFFF ground.

The wrong version. Build the slide by dragging boxes onto it in whatever order they occurred to you, then run the accessibility checker the night before sending.

The stored object order on that slide is chart, logo, title, third key figure, highlight callout, first key figure, page number, second key figure, footnote. A sighted reader takes them in the order title, three key figures, chart, callout, footnote, logo, page number. Compare the two orders across all 36 orderable pairs and 16 of them are announced out of sequence, which is 44.44%. Two objects are announced before the title, so a listener meets a chart and a logo before learning what the slide is about.

The colours are a second failure and a quieter one. The footnote and the source line are set in the light grey A6A6A6, which is a contrast ratio of 2.434 against white and fails both thresholds. The headline figure is set in the brand amber on white, at 2.009, which also fails both. Neither failure is visible to the person who chose them, because both look fine to a designer sitting eighteen inches from a bright screen.

The corrected version. Fix the order while building, describe the objects by their message, and check the palette once against the two thresholds.

Reorder in the Selection pane so the stored order matches the visual order, which takes about twenty seconds and drops the out-of-order pairs from 16 to 0. Pressing Tab then walks the slide in the order a listener will hear it, which makes the check free on every subsequent slide.

Of the nine objects, 5 are content and need alt text, and 2 are decorative and should be marked as such so they are skipped rather than described. The chart's alt text states the finding rather than the picture: a description that begins "bar chart showing" has told a listener nothing that the title should not already have told them.

Then the palette, computed once against 4.5 to 1 for normal text and 3 to 1 for large text, where large means at least 18 pt or 14 pt bold.

Four of the eight pairs fail the 4.5 to 1 threshold for body text and two fail even the 3 to 1 threshold for large text, so exactly half the palette is usable for ordinary text. Drawn as a sorted bar of the eight ratios with reference lines at 4.5 and 3.0, the exhibit needs no legend and titles itself: "Four of the eight brand pairs fail the 4.5:1 threshold for body text."

Two substitutions fix the slide. Footnotes and sources move from A6A6A6 at 2.434 to 595959 at 7.005, and nobody will describe the result as less subtle. The headline figure moves off amber-on-white at 2.009 to navy on white at 8.663, and the amber is kept for a small accent on the navy band, where it reaches 4.311 and is legitimate at title size and not at body size. The mid grey at 4.004 stays available for large text only, which is a rule you can state in a template note rather than remember.

The checklist to apply.

1. Fix the stored object order while building; press Tab to hear it.
2. Give every slide a unique title, hidden if it must be, because titles are how a deck is navigated by ear.
3. Write alt text as the message, not as the chart type, and mark decorative objects decorative.
4. Compute every foreground and background pair in the palette once, against both thresholds.
5. Treat a colour that passes only the large-text threshold as a title colour, and write that down.
6. Never rely on hue alone to distinguish series, because the grayscale proof and the colour-blind reader fail together.

Verify

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

Canvas: widescreen 13.3333 x 7.5 inches, ratio 1.777778; legacy 10.0 x 7.5, ratio 1.333333; area share 0.750000; dead band 3.3333 inches, 25.0000% of width; width-fill stretch 1.333333; a 2.00 inch logo becomes 2.6667; 1080p across the canvas is 144.0 ppi; 4K is 288.0 ppi.

Kaveri Kitchens: 34 slides; switches 30 of 33 pairs, rate 0.9091; longest single-topic run 2; tags Cover 1, Business 10, Financials 13, Product 6, Asks 4, summing to 34; shares 2.94%, 29.41%, 38.24%, 17.65%, 11.76%; Financials plus Business 23 slides, 67.6471%; switches after sectioning 4, rate 0.1212, reduction 26 and 86.6667%; longest in-order run 16; slides that must move 18, 52.9412%; titles in placeholders 26, 76.4706%, invisible 8; bodies in placeholders 19, 55.8824%, invisible 15; action titles 4, 11.7647% of slides and 15.3846% of visible titles, label titles 22; restructure 360 + 272 + 100 + 198 = 930 seconds, 15.5 minutes; words edited 0.

Lakeshore Analytics: photographs 18 x 2.15 = 38.70 MB, screenshots 6 x 0.92 = 5.52, logos 3 x 0.18 = 0.54, video 2.80, XML and theme 1.14; total 48.70; shares 79.4661%, 11.3347%, 5.7495%, 2.3409%, 1.1088%; cumulative 79.47%, 90.80%, 96.55%, 98.89%, 100.00%; top two 90.8008%, bottom three 9.1992%; ceiling 25.0, overshoot 23.70 MB and 1.948 times. Photographs 4032 x 3024 px, 12.1928 MP, placed 13.3333 in, 302.4 ppi; at 150 ppi need 2000.0 px, linear 0.496032, area 0.246047, pixels removed 75.3953%, size 9.5220 MB. Screenshots 2560 x 1440 placed 8.00 in, 320.0 ppi; need 1200.0 px, linear 0.46875, area 0.219727, size 1.2129 MB. Fonts subset 0.42, full 3.10, difference 2.68. After 9.5220 + 1.2129 + 0.5400 + 1.1400 + 0.4200 = 12.8349 MB; component shares of the after file 74.19%, 9.45%, 8.88%, 4.21%, 3.27%; reduction 35.8651 MB and 73.6449%; headroom 12.1651; 51.3397% of the ceiling. Projection check: available 1920 px, supplied 2000.0, surplus 80.0, ratio 1.041667. 4K: need 3840, shortfall 1840.0, supplied share 52.0833%. Print at 220 ppi: need 2933.33 px, photographs 20.4830 MB, deck 25.1920 MB, under 25.0? False.

Nagarjuna Cement: 30 slides; pages 30, 15, 10, 8, 5, 4 at 1, 2, 3, 4, 6 and 9 per page; A4 8.2677 x 11.6929 in, printable at 0.5 in margins 7.2677 x 10.6929; two-up cell 7.2677 x 5.2465, scale 0.545079; three-up-with-note-lines column 3.9972 in (0.55 of the printable width; 0.2 in gutter between cells), scale 0.299793; six-up cell 3.5339 x 3.4310, scale 0.265039. Printed sizes: 40 pt title 21.803 / 11.992 / 10.602; 24 pt subhead 13.082 / 7.195 / 6.361; 18 pt body 9.811 / 5.396 / 4.771; 12 pt label 6.541 / 3.598 / 3.180. Minimum on-slide pt for 9 pt on paper 16.511 / 30.021 / 33.957. Slides whose smallest type is 11 pt: 14 of 30, 46.6667%; 11 pt prints at 5.996 two-up and 2.915 six-up. Sheets: two-up 15 pages, 8 duplex; six-up 5 pages, 3 duplex; saving 10 pages.

Harborline Capital: distinct shortcut operations 20; keyboard operations 158; operations with no shortcut 16; all operations 174; coverage 90.8046%; six most frequent total 74, 46.8354%, and 6 operations appear ten or more times. Mouse 3.4 s and keyboard 0.9 s per operation; mouse 537.20 s (8.953 min), keyboard 142.20 s (2.370 min); saving 395.00 s, 73.5294%, ratio 3.777778. Learning 20 x 6 = 120 minutes, 2.0 hours; payback 18.2278 runs; at 3 runs a week over 46 weeks, 54,510.00 s or 15.1417 hours a year.

Meridian Logistics: 46 slides, 3 masters, 7 fonts, 12 slides of 4:3 content, 9 slides with an object outside the frame, 31 titles in placeholders, 67.3913%. Repair yields 3.286, 2.091, 0.750, 0.500, 0.333 slides per minute; total 110 minutes and 128 slide-instances; first two 36 minutes and 92 instances, 71.8750% of instances in 32.7273% of the time; remaining three 74 minutes and 36 instances.

Cedar Ridge: 8 versions, 28 pairs; ad-hoc inversions 10, 35.7143% wrong and 64.2857% right; ISO inversions 0, 100.0000% right. Expected minutes lost 5.0 / 2.5 / 1.0 at 10, 5 and 2 minute intervals; 6.0 editing hours at 0.111111 crashes an hour gives 0.666667 expected crashes and 3.3333 / 1.6667 / 0.6667 expected minutes; moving from 10 to 2 saves 2.6667.

Accessibility: 9 objects, 36 pairs, 16 announced out of order, 44.4444%; 2 objects before the title; 5 objects need alt text and 2 are decorative. Contrast ratios 8.663 navy on white, 8.663 white on navy, 7.0047 dark grey on white, 6.4789 red on white, 4.3111 amber on navy, 4.0041 mid grey on white, 2.4344 light grey on white, 2.0094 amber on white; 4 of 8 fail 4.5:1 and 2 of 8 fail 3.0:1.

Practice set

Work each problem fully before reading its solution. A calculator is assumed. Numeric answers within 2% of the stated figure count as correct unless the problem says otherwise, and where a problem asks for a judgement, the test is whether you named the number the judgement rests on. Entities are synthetic.

P1 (guided). Audit an Outline (US). A 52-slide deck has its title in the title placeholder on 39 slides. State the share of the deck the Outline can see as a title, the number of slides that show nothing there, and two consequences of that gap that are not about appearance.

Solution. Visible share is 39/52 = 75.00%, and 52 - 39 = 13 slides show no title in the Outline. The consequences that matter are operational rather than visual. Those 13 slides cannot be read in the read-only-the-titles test, so the deck's argument cannot be checked without opening every slide. They are also invisible to the outline handout, to deck search, and to assistive technology's slide navigation, and their text does not inherit the master's typography, so it will drift the next time the theme changes. A common wrong answer is that the slides "look inconsistent", which is the least important effect on the list.

P2 (guided). Section a short deck (India). A 12-slide update covers three topics in the order A, B, A, C, B, A, C, C, B, A, B, C. Count the adjacent topic switches as received, state the switch count after the slides are grouped into three contiguous sections, and give the reduction as a count and a share.

Solution. Compare each adjacent pair: A to B, B to A, A to C, C to B, B to A, A to C, C to C (no switch), C to B, B to A, A to B, B to C. That is 10 switches across 11 adjacent pairs. Grouped into three contiguous sections there are exactly 2 switches, one at each section boundary, because a deck of n contiguous sections has n - 1 boundaries. The reduction is 10 - 2 = 8 switches, or 8/10 = 80.00%. The general result is worth carrying: sectioning cannot get below one switch per boundary, so the achievable floor is set by how many sections you choose, not by how well you sort.

P3 (guided). Size a compression (US). A photograph 5184 pixels wide is placed 9.00 inches wide on a slide. Give its effective resolution, the pixels needed at a 150 ppi target, the area ratio, the new size of a 3.60 MB file, and the total saving across 22 such images.

Solution. Effective resolution is 5184/9.00 = 576.0 ppi. At 150 ppi the image needs 150 x 9.00 = 1350.0 pixels across, a linear ratio of 1350/5184 = 0.260417 and an area ratio of 0.260417² = 0.067817. The file becomes 3.60 x 0.067817 = 0.2441 MB. Across 22 images the saving is 22 x 3.60 x (1 - 0.067817) = 73.8289 MB. The trap is quoting the linear ratio as the saving, which would predict a 74% reduction when the true reduction is 93.2%: file size follows pixel count, and pixel count goes as the square of the linear ratio.

P4. Set a handout floor (India). A pack will print six-up on A4 with half-inch margins, where the scale factor is 0.265039. A chart label is 14 pt on the slide. State what it prints at, and state the minimum on-slide type size that would clear a 10 pt floor on paper.

Solution. 14 x 0.265039 = 3.7106 pt on paper, which is not readable at any sensible distance. To clear 10 pt on paper the on-slide size must be at least 10/0.265039 = 37.7302 pt. No chart label is 37.73 pt, so the honest conclusion is that this chart cannot go on a six-up handout in its present form: either the layout changes to two-up, or the chart is rebuilt as small multiples with direct labels large enough to survive.

P5. Choose between two bad fixes (US). A slide built for a 10.0 by 7.5 inch canvas is pasted onto a 13.3333 by 7.5 inch one. Give the share of the canvas left empty if the content is fitted by height, the width a 3.00-inch box becomes if the content is stretched to fill, and say which you would choose and why.

Solution. Height-fitting preserves proportions and uses 10.0 inches of width, leaving 13.3333 - 10.0 = 3.3333 inches empty, which is 3.3333/13.3333 = 25.00% of the canvas. Width-filling stretches everything by 13.3333/10.0 = 1.333333, so a 3.00-inch box becomes 4.0000 inches at unchanged height. Choose the height fit. Empty canvas is a layout problem with several fixes, while a 33.33% horizontal stretch distorts every logo, circle and photographed face on the slide, and distortion of somebody else's brand mark is a problem you cannot design your way out of.

P6. Get a deck under a ceiling (India). A deck is 31.5 MB of photographs, 9.8 MB of video, 4.4 MB of screenshots, 1.2 MB of slide XML and 0.6 MB of logos, and must come under 20.0 MB. Give the total, the component shares, the megabytes that must go, and whether removing the video alone is enough. Then state the size after removing the video and resampling the photographs at an area ratio of 0.246.

Solution. Total is 47.5 MB. Shares are photographs 66.3158%, video 20.6316%, screenshots 9.2632%, XML 2.5263% and logos 1.2632%. To reach 20.0 MB you must remove 27.5 MB. Removing the video alone leaves 37.7 MB, which is not close, and it is also the intuitive first move because the video feels like the big object. Removing the video and resampling the photographs at an area ratio of 0.246 leaves 47.5 - 9.8 - 31.5 x (1 - 0.246) = 13.949 MB, comfortably under. The photographs were always the problem; the video was merely the most visible item on the list.

P7. Price a save interval (US). A team saves every 15 minutes and crashes about once every 9 editing hours. Give the expected minutes lost per crash, and the expected minutes lost across a 4-hour editing session. Then say what the number argues for.

Solution. With crashes arriving uniformly at random between saves, the expected loss per crash is half the interval: 15/2 = 7.5 minutes. Over 4 editing hours the expected number of crashes is 4 x (1/9) = 0.4444, so the expected loss is 4 x (1/9) x 7.5 = 3.3333 minutes. The number argues against treating the interval as the safeguard. Three and a third expected minutes is not worth a policy; the tail is, and the tail is a crash in the middle of a restructure whose decisions cannot be reconstructed. That is a versioning control, which is a milestone copy before a structural change, not an interval control.

P8. Judge two greys (India). A footnote is set in A6A6A6 on white, contrast 2.4344. The proposal is to change it to 595959, contrast 7.0047. State whether each passes the 4.5 to 1 threshold for normal text, and state what is actually being traded.

Solution. A6A6A6 at 2.4344 fails 4.5 to 1 and also fails the 3 to 1 large-text threshold, so it is not usable for text of any size. 595959 at 7.0047 passes both. What is being traded is nothing measurable: the darker grey remains visibly subordinate to black body text and to a navy heading, so the hierarchy the designer wanted survives while the legibility problem disappears. The habit of setting secondary text in a very light grey is a screen-brightness artefact, and it is the single commonest contrast failure in corporate decks.

P9. Test a naming scheme (US). Six versions arrive in this order: deck.pptx, deck2.pptx, deck_final.pptx, deck_final2.pptx, deck_new.pptx, deck_newest.pptx. Count the orderable pairs, count how many a case-insensitive alphabetical sort puts in the wrong order, and say whether the scheme is safe.

Solution. Six files give 6 x 5 / 2 = 15 orderable pairs. Sorted alphabetically the six land in exactly their true order, so 0 pairs are wrong, or 0.00%. The scheme is not safe. It sorted correctly by accident, because each successive name happened to extend or follow its predecessor alphabetically; add one file called deck_FINAL_v2.pptx or deck_approved.pptx and the order breaks immediately, with no warning and no visible change to the files that already worked. A convention is a rule that holds for names you have not written yet, and an ISO date plus a zero-padded version is such a rule while "add a word that sounds later" is not.

P10. Compute a keyboard payback (India). A task contains 64 keyboard-eligible operations. Ribbon operations cost 3.4 seconds and keyboard operations 0.9 seconds. Give the saving per run, and the number of runs needed to repay 90 minutes of learning.

Solution. The saving is 64 x (3.4 - 0.9) = 160.0 seconds a run. Ninety minutes is 5,400 seconds, so the payback is 5400/160 = 33.75 runs. Whether that is worth doing depends entirely on how often the task recurs, which is the question to ask before any productivity investment: at three runs a week the answer is eleven weeks, and at one run a month it is nearly three years, from the same arithmetic.

P11. Count a reading order (US). A slide has six objects. A sighted reader takes them in the order 1, 2, 3, 4, 5, 6; the stored order announces them as 3, 1, 5, 2, 6, 4. Count the orderable pairs and the pairs announced out of order, and give the share.

Solution. Six objects give 15 orderable pairs. Comparing every pair of positions in the announced sequence against the visual order, 5 pairs are inverted: (3,1), (3,2), (5,2), (5,4) and (6,4). That is 5/15 = 33.33%. The share matters more than the count, because it is comparable across slides with different object counts, and a deck-level average of it is the cheapest single accessibility metric you can compute.

P12. Plan a print run (India). A 44-slide pack will be printed. Give the pages at three-up and at six-up, the sheets saved by choosing six-up, and the double-sided sheet count at six-up.

Solution. Three-up needs ⌈44/3⌉ = 15 pages and six-up needs ⌈44/6⌉ = 8 pages, so six-up saves 7 pages. Printed double-sided, six-up needs ⌈8/2⌉ = 4 sheets. The saving is real and it is also not the decision: at a six-up scale factor of 0.265039, everything under 33.957 pt on the slide falls below a 9 pt floor on paper, so the seven sheets are saved by making the pack unreadable unless the slides were designed for it.

P13. Match a file to a room (US). A full-bleed image spans the 13.3333-inch canvas. Give the resolution needed to match a 1080p projector exactly and a 2560-pixel-wide display exactly, the pixel width of a 96 ppi file across that canvas, and its shortfall against 1080p.

Solution. A 1080p projector renders 1920 pixels across the canvas, so it needs 1920/13.3333 = 144.0 ppi. A 2560-pixel display needs 2560/13.3333 = 192.0 ppi. A 96 ppi file supplies 96 x 13.3333 = 1280.0 pixels, a shortfall of 1920 - 1280 = 640.0 pixels against 1080p, which is exactly a third of what the projector can show. The lesson runs both ways: 96 ppi is visibly soft on a modern projector, and 300 ppi is invisible waste unless the deck will be printed.

P14. Audit titles (India). A 28-slide deck has 6 titles that are full sentences with a verb. Give the share, the count of label titles, and state the one test you would run before rewriting any of them.

Solution. The share is 6/28 = 21.43%, and 22 titles are labels. Before rewriting anything, run the read-only-the-titles test: collapse the Outline to titles and read them in order, writing down the argument they make. That test tells you whether the 22 labels are hiding an argument that already exists, in which case rewriting is a two-hour job, or whether there is no argument in the deck at all, in which case rewriting titles is the wrong repair and the deck needs its storyline decided first.

Applied mini-project

Restructure an inherited deck without changing a word of it.

Take a 30-slide deck you did not build. Use a real one if you have it, from work or from a company's published investor material; if you have none, the fallback brief after the deliverables supplies one. Three artefacts come out, and each has a test it must pass before it counts.

Artefact 1 is the structural rebuild, and three tests decide whether it counts. Open the deck cold and, before editing anything, record six file-level numbers. Slide count, distinct masters, distinct fonts, and the canvas ratio the content was built for. Then the count of slides whose title is in the title placeholder, and the count whose body text is in a content placeholder. Then tag every slide with one topic and count the adjacent topic switches. Then section the deck, naming each section for the claim it carries rather than for its department. Test 1: the switch count after sectioning equals the number of sections minus one, which proves every section is contiguous. Test 2: the number of slides you moved is reported, and equals the slide count minus the longest run that was already in section order. Test 3: the word count of slide content is identical before and after, which proves you restructured rather than rewrote.

Artefact 2 is the file hygiene pass, with three tests of its own. Inventory the file by component in megabytes, sorted descending, with cumulative shares. Compute the effective resolution of the largest image class as pixels divided by placed inches. Choose a target resolution from the display the deck will actually meet, and state that display. Resample, link any embedded video, and embed fonts as a subset. Rename the file to the convention. Test 4: the component inventory sums to the original file size to within 0.1 MB. Test 5: the predicted size after resampling, computed from area ratios, matches the actual saved size to within 15%, with any gap explained. Test 6: the supplied pixel width across the canvas is at least what the stated display renders, shown as a ratio.

Artefact 3 is the handout, and two tests apply to it. Produce a printable handout as a separate file. Compute the scale factor of your chosen layout from the printable area of your paper size. Find the smallest type size anywhere in the deck, including axis labels and footnotes, and compute what it prints at. Raise anything that falls below a 9 pt floor on paper, changing chart form where a larger label will not fit. Test 7: the smallest printed type size is at or above 9 pt, shown as the on-slide size times the scale factor. Test 8: a grayscale proof of one chart-bearing slide is included, and every series in it stays distinguishable without colour.

Deliverable. Two files, the presented deck and the handout, plus a five-line note. The five lines are fixed, one each. What the sectioning made obvious that was not obvious before. Which repair had the highest slides-per-minute yield, and what you skipped. Where the file's megabytes actually were. The display or paper you optimised for, and what that choice costs elsewhere. And the one thing in the deck you did not fix, with the reason.

The fallback brief. A 30-slide operations pack from a synthetic mid-size manufacturer arrives with 3 masters, 7 fonts, 12 slides of 4:3 content pasted onto a 16:9 canvas, 14 slides whose smallest type is 11 pt, and 22 label titles. Its 47.5 MB is 31.5 MB of photographs, 9.8 MB of video, 4.4 MB of screenshots, 1.2 MB of slide XML and 0.6 MB of logos. It will be shown on a 1080p projector and handed out on A4. Mail attachments are capped at 20.0 MB. Run all three artefacts on that.

The rubric below runs to twenty points, passing at 16, with no zero on any row.

Any artefact whose test fails scores zero on the rows that test supports, however well the deck reads. The point of the exercise is that structure, weight and legibility are all checkable before anyone gives an opinion, and an opinion offered on a deck that fails Test 1 is an opinion about the wrong problem.

Reading & resources

Core (do these).

• The vendor's own keyboard-shortcut reference for PowerPoint, published by Microsoft and free on its support site. Print the page, mark the twenty operations from the table above, and keep it beside the keyboard for a fortnight. Assignments differ between the Windows and macOS builds and move between releases (checked September 2026; verify against the build you actually use). [Free] [Beginner]
• Microsoft's guidance on making presentations accessible, also free, which is the practical companion to the reading-order and alt-text work here and names the controls by their current interface labels. Read the sections on reading order, alt text and slide titles; skip the parts that duplicate the contrast standard below. [Free] [Beginner]
• W3C Web Content Accessibility Guidelines, free from the W3C. The contrast minimum for normal-size text is 4.5 to 1 and for large text 3 to 1, and those two numbers are what the palette table above is computed against (WCAG 2.2, checked September 2026; verify, because guideline versions advance and thresholds can move). [Free] [Intermediate]
• **Matthew Butterick, *Practical Typography***, readable free online. The chapters on point size, line length and line spacing are the reason the handout arithmetic has a 9 pt floor rather than an opinion, and the treatment of why type behaves differently on paper and on screen is the clearest short version available. [Free] [Beginner]
• **Edward Tufte, *The Cognitive Style of PowerPoint***. A short, hostile essay about what the default deck structure does to an argument, and worth reading precisely because it is unfair in places. Read it as a description of the failure mode a sectioned, action-titled deck is built to avoid. [Paid] [Beginner]

Going deeper.

• **Stephen Kosslyn, *Clear and to the Point***. Slide design derived from perception and memory research rather than from taste, and the best available account of why a slide that is technically complete can still be unreadable at presentation distance. [Paid] [Intermediate]
• **Michael Alley, *The Craft of Scientific Presentations***. The assertion-and-evidence structure, argued from engineering and science practice. The title work belongs to the storytelling branch, but the book's account of why a topic-labelled slide fails is the sharpest one in print. [Paid] [Intermediate]
• **Robin Williams, *The Non-Designer's Design Book***. Four principles applied to ordinary documents, and the fastest route from "something is wrong with this slide" to a name for what is wrong. The precision branch takes the same ideas much further. [Paid] [Beginner]
• The Office Open XML specification, ISO/IEC 29500 and ECMA-376, free from ECMA International. Long, and you will not read it end to end. Read enough of the packaging section to see that a presentation file is a zip container of XML parts plus a media folder, because that fact is what makes the salvage route above work. [Free] [Advanced]

Standards and tools (free).

• ISO 216, the paper-size standard, which fixes A4 at 210 by 297 millimetres, or 8.2677 by 11.6929 inches. Every scale factor in the handout section descends from those two numbers and the margins you choose.
• A contrast checker, of which several are free in the browser, including the one published by the WebAIM project. Paste two hex values, read the ratio, and check both thresholds. Doing the palette once takes ten minutes and never needs repeating.
• A screen reader you already own. VoiceOver ships with macOS and iOS, Narrator ships with Windows, and NVDA is free for Windows. Turn one on and let it read one of your own slides; the experience is more persuasive than any argument about reading order (checked September 2026; verify current availability, since bundled tools change with operating-system releases).
• The help centres for Google Slides, Keynote and LibreOffice Impress, free, and the place to confirm the mapping table above against the build in front of you rather than trusting a table written on one date.

Do this, not just read. Take the last deck you received from somebody else. Without opening a single slide, record six numbers: slide count, masters, fonts, canvas ratio, titles in placeholders, bodies in placeholders. Then count the adjacent topic switches. Seven numbers, about eight minutes, and you will know more about that deck than its author does.

Flashcards

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

Mastery check

Two parallel forms. Closed book, calculator allowed, about 40 minutes per form. Numeric answers within ±2% score as correct unless the item states otherwise. Pass threshold: ≥ 85%. With twelve one-point items that is 11 of 12. If you score 9 or 10, redo the worked example that covers the cluster you missed and sit the other form after a break.

Form A

A1 (MCQ). Which view answers the question "do the titles of this deck make an argument"? (a) Normal (b) Outline (c) Slide Sorter (d) Reading view

A2 (numeric). A 40-slide deck is grouped into five contiguous sections by topic. How many adjacent slide pairs still change topic?

A3 (numeric). A photograph 3600 pixels wide is placed 12.00 inches wide on a slide. Give its effective resolution in pixels per inch.

A4 (numeric). That photograph's file is 4.00 MB. Resampled to a 150 ppi target across the same 12.00 inches, what does it become, in MB?

A5 (MCQ). A 48.70 MB deck must reach a 25.0 MB mailbox. Component shares are photographs 79.47%, screenshots 11.33%, video 5.75%, slide XML and theme 2.34%, logos 1.11%. The first move is: (a) delete eight slides (b) zip the file (c) resample the photographs against the pixel width the display can render (d) remove the three logos

A6 (numeric). A handout prints two-up on A4 at a scale factor of 0.545079. What does 20 pt type on the slide measure on paper, in points, to four decimal places?

A7 (MCQ). Content built for a 10.0 by 7.5 inch canvas is pasted onto a 13.3333 by 7.5 inch one. The right treatment is: (a) stretch to fill the width, because empty canvas looks unfinished (b) fit by height and accept 25.00% of the canvas empty (c) stretch vertically to match instead (d) crop a quarter off each image

A8 (numeric). A team saves every 6 minutes. If a crash arrives uniformly at random between saves, what is the expected work lost per crash, in minutes?

A9 (MCQ). A mid grey whose contrast against white is 4.004 is: (a) usable for text of any size (b) usable for large text only, meaning at least 18 pt or 14 pt bold (c) unusable at any size (d) usable for body text but not for titles

A10 (MCQ). Eight versions of a file are named ad hoc, and 10 of their 28 orderable pairs sort into the wrong order. The repair is: (a) keep the newest in a folder called Final (b) name files as subject, ISO date, zero-padded version, lowercase, no spaces (c) append "v2", "v3" and so on (d) rely on the file system's modified date

A11 (short). A 34-slide deck has its title in the title placeholder on 26 slides. Give the share, and state two consequences of the gap that have nothing to do with appearance.

A12 (short). You have 40 minutes with an inherited 46-slide deck before a meeting. Five repairs are available, with these yields in slides per minute: re-point 7 fonts to 2 theme fonts 3.286, consolidate 3 masters to 1 2.091, retype 15 stray titles 0.750, pull 9 objects inside the frame 0.500, re-frame 12 pasted 4:3 slides 0.333. State what you do, what you skip, and the one thing you write down before the meeting.

Form A key. A1: b. The Outline shows only text in title and content placeholders, which is what the read-only-the-titles test needs. Normal shows one slide, Slide Sorter shows shape without readable words, and Reading view shows the artefact rather than its argument. A2: a deck of n contiguous sections has exactly n − 1 boundaries, so 5 − 1 = 4. A3: 3600 / 12.00 = 300.0 ppi. A4: at 150 ppi the image needs 150 × 12.00 = 1800 pixels, a linear ratio of 1800/3600 = 0.5 and an area ratio of 0.25, so 4.00 × 0.25 = 1.00 MB; answering 2.00 MB is the standard error of applying the linear ratio to a file size. A5: c. Photographs carry 79.47% of the weight, so nothing else is worth an hour; deleting slides removes part of the 2.34% that is XML and theme, and zipping recovers almost nothing because the file is already a zip container of already-compressed images. A6: 20 × 0.545079 = 10.9016 pt. A7: b. Height-fitting preserves proportions and leaves (13.3333 − 10.0)/13.3333 = 25.00% of the canvas empty, which is a layout problem with several solutions; width-filling stretches everything by 1.333333, so every logo, circle and face is 33.33% too wide. A8: expected loss is half the interval, 6/2 = 3.0 minutes. A9: b. The thresholds are 4.5 to 1 for normal text and 3 to 1 for large text, so 4.004 clears the second and misses the first; the practical rule is that it is a title colour, and that belongs written down in the template note. A10: b. ISO dates sort chronologically because alphabetical and chronological order coincide in that format, and zero padding stops a tenth version sorting ahead of a second; the modified date is lost the moment a file is copied or mailed. A11: 26/34 = 76.4706%, so 8 slides show no title in the Outline. Two consequences that are not visual. Those slides cannot be read in the read-only-the-titles test, so the deck's argument cannot be checked without opening every slide. And their text is invisible to the outline handout, to deck search and to assistive technology's slide navigation, and it does not inherit the master's typography, so it drifts on the next theme change. A12: Do the two propagating repairs, which are re-pointing the fonts (14 minutes) and consolidating the masters (22 minutes); together they cost 36 minutes, touch all 46 slides, and capture 71.88% of the total slide-instances in 32.73% of the total repair time. Skip the three low-yield repairs, which cost the other 74 minutes and reach only 36 slide-instances between them, and in particular skip the 4:3 re-framing at 0.333 slides a minute even though it is the most visible fault. Write down what you did not fix, so the limits are stated by you rather than discovered by the room.

Form B

B1 (MCQ). Which view answers "where is this deck dense, and does its shape match its argument"? (a) Normal (b) Outline (c) Slide Sorter (d) Reading view

B2 (numeric). A 30-slide deck is grouped into four contiguous sections by topic. How many adjacent slide pairs still change topic?

B3 (numeric). A photograph 4800 pixels wide is placed 10.00 inches wide on a slide. Give its effective resolution in pixels per inch.

B4 (numeric). That photograph's file is 5.00 MB. Resampled to a 150 ppi target across the same 10.00 inches, what does it become, in MB, to four decimal places?

B5 (MCQ). Zipping a 48.70 MB presentation file recovers almost nothing because: (a) zip is a weak algorithm for large files (b) the file is already a zip container and its photographs are already in a compressed format (c) presentation files cannot be compressed at all (d) the slide XML dominates the file

B6 (numeric). A handout prints six-up on A4 at a scale factor of 0.265039. What does 32 pt type on the slide measure on paper, in points, to four decimal places?

B7 (MCQ). Content built for a 4:3 canvas and stretched to fill a 16:9 one is scaled horizontally by 1.333333. The consequence that decides the case is: (a) text becomes marginally harder to read (b) the file size increases (c) every logo, circle and photographed face is 33.33% too wide (d) the slide numbers move out of position

B8 (numeric). A team saves every 20 minutes. If a crash arrives uniformly at random between saves, what is the expected work lost per crash, in minutes?

B9 (MCQ). A brand amber whose contrast against white is 2.009 is: (a) acceptable for headline figures, because headlines are large (b) acceptable for body text (c) below both the 4.5 to 1 and the 3 to 1 thresholds, so it is not a text colour on white at any size (d) acceptable once it is set in bold

B10 (MCQ). A theme and a template differ in that: (a) they do not differ, the words are interchangeable (b) a theme is a file and a template is a colour set (c) a theme is a set of formatting roles, while a template is a file carrying a theme plus masters, layouts and starter slides (d) a template governs printing only

B11 (short). A 46-slide inherited deck has its title in the title placeholder on 31 slides. Give the share, then name the three file-level properties you would read before opening any slide and say what each one tells you.

B12 (short). A 30-slide pack will be printed six-up on A4, where the scale factor is 0.265039, and the smallest type on 14 of the slides is 11 pt. State what those labels measure on paper, the minimum on-slide size that clears a 9 pt floor at six-up, and the decision you would take.

Form B key. B1: c. Slide Sorter is the only view that shows the whole deck at once, at a size where words are unreadable and structure, density and section shape are what register. B2: 4 − 1 = 3. B3: 4800 / 10.00 = 480 ppi. B4: at 150 ppi the image needs 1500 pixels, a linear ratio of 1500/4800 = 0.3125 and an area ratio of 0.097656, so 5.00 × 0.097656 = 0.4883 MB. B5: b. A modern presentation file is a zip container of XML parts and a media folder, and JPEG photographs are already compressed, so a second pass recovers close to nothing. The XML with the theme was only 2.34% of the file. B6: 32 × 0.265039 = 8.4813 pt, which is below a 9 pt floor, so even a 32 pt slide heading fails on a six-up handout. B7: c. The distortion is the deciding fact, because a client's brand mark rendered a third too wide sits on a document carrying your firm's name; the alternative leaves 25.00% of the canvas empty, which is a solvable layout problem. B8: 20/2 = 10.0 minutes. B9: c. 2.009 is below 3 to 1, so it fails even the large-text threshold; the same amber on the brand navy reaches 4.311 and is legitimate at title size, which is the substitution to make. B10: c. Applying a theme to a deck built from a blank file gives you the colours and fonts without the layouts, which is why the result usually looks like a themed blank deck. B11: 31/46 = 67.3913%, so 15 slides show no title in the Outline. Three properties. The number of slide masters tells you whether the file is a merge and how far a master change would propagate. The number of distinct fonts tells you how much formatting is local rather than inherited, and it is usually the highest-yield repair available. The canvas ratio the content was built for matters because 4:3 content on a 16:9 canvas carries a distortion that no nudging will fix. B12: 11 × 0.265039 = 2.915 pt on paper, and clearing a 9 pt floor at six-up needs at least 9/0.265039 = 33.957 pt on the slide, which no chart label ever is. The decision is to stop treating the handout as a print setting: either print two-up on 15 pages, where 18 pt type prints at 9.811 pt and clears the floor, or build a separate handout file whose charts are re-formed as small multiples with direct labels large enough to survive. Six-up saves 10 pages by making the pack unreadable.

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

Teach it back & journal

Teach it back. Find somebody who builds decks and has never thought about any of this, and take ten minutes with a deck of theirs on the screen. Do three things in order, out loud.

First, tag their slides by topic in front of them and count the adjacent switches. Say the number, then say what the number would be if the deck were grouped into sections, using the rule that n contiguous sections give n minus 1 switches. Watch what happens when they see that the fix costs fifteen minutes and changes no words.

Second, open their Outline and count two things: how many titles it can see, and how many of those are sentences with a verb. Do not offer to rewrite anything. The point of the exercise is that both numbers were available in ninety seconds and neither of them is a matter of taste.

Third, ask what their mail system's attachment limit is, then take their largest photograph, ask how wide it sits on the slide, and compute the effective resolution in front of them. If it comes out above 200 ppi for a deck that will be projected, you have just found megabytes they did not know they were carrying, and you can say exactly how many because the saving goes as the square of the linear ratio.

The test of whether you understand this is whether you can do all three without opening a settings dialog. Every number involved is a division you can do on paper.

Journal. Write four hundred words on the last deck you sent to somebody else, answering four questions with numbers rather than adjectives. How many adjacent topic switches did it have, and how many sections would have fixed that? Record next what share of its titles the Outline could see, and what share of those were claims. Then where its megabytes were, and what resolution its largest image actually needed for the room it was shown in. Then what you named the file, honestly, including every earlier version still sitting in the same folder.

Then write one paragraph on the habit you are going to change first, and why that one. The candidates are not equal. Sectioning changes how every future deck is built and costs nothing to adopt. The twenty keyboard operations cost two hours and repay in weeks. The naming convention costs nothing and saves somebody else's afternoon rather than yours, which is why it is the one most people skip. Pick one, write down the date, and check yourself against it in a month.