The Consulting Deck and the Project Readout

Presentation track, branch pr-archetypes (Archetypes) · Node PR5.01 · ~12 focused hours · Mastery gate ≥ 85%

Twenty-five pages. A forty-five minute slot, of which three go to introductions and twelve are held back for discussion, leaving thirty minutes of presenting time. That is 72 seconds a page, and nobody has ever presented a page of analysis in 72 seconds.

The arithmetic is not a warning about being concise. It is the reason the consulting deck has the shape it has. Somebody worked out, decades ago, that a deck which must survive both a reader alone at 11pm and a room that will interrupt on page two cannot be one continuous document. So it was cut into three parts with different jobs. A front section that carries the argument and is sized to the meeting. An appendix that carries the evidence and is sized to the questions. And a single summary page at the top that works even if the reader stops there, which many of them do.

That three-part shape is an archetype: a deck skeleton that encodes what a particular audience expects, so that the expectations do not have to be renegotiated every time. A partner opening a client deck knows the executive summary is page 2 and that the method sits in the appendix, indexed. A hospital administrator opening a project readout knows the next steps have owners and dates on them, because a readout without owners is a status update wearing a recommendation's clothes. Neither person reads the deck from the beginning. They both go to the page their archetype tells them to go to, and if it is not there they lose trust in the pages that are.

Two archetypes are built here, and they are close cousins with one important difference. The consulting deck argues: it exists to make somebody believe a claim they currently do not hold, and its spine is a chain of action titles that a reader can follow with the exhibits hidden. The project readout reports: it exists to close out work that a sponsor commissioned, and its spine is a fixed five-part sequence that the sponsor expects to find in order, because the readout is as much a governance artefact as a persuasive one. Confusing the two produces the commonest defect in corporate decks, which is an argument deck that never says what happens next, or a readout that spends nine pages persuading a sponsor who already agreed.

Both archetypes are demonstrated on real arithmetic. On the India side, a quick-commerce network called Tez Dabba (synthetic) runs 440 dark stores across four city cohorts and made ₹4.6759 crore of contribution last month on 5,550,750 orders, which is ₹8.42 an order. That number is positive and it hides something: one cohort of four produces ₹12.1094 crore and the other three lose ₹7.4335 crore between them. On the US side, Meridian Valley Medical Center (synthetic) runs an eight-room operating suite at 73.63% block utilisation, which sounds like a turnover problem and is not.

Both of those decks are built page by page below, including the pages that were wrong first. What carries over from the earlier branches is assumed and isn't taught again. The pyramid that produces a governing thought, the action title that carries a claim, the horizontal and vertical logic that make a title stack readable, the seven-rule floor a slide has to clear: all of that is upstream, and it is only named here. What is new is the container, and the container matters more than people expect, because a good argument in the wrong archetype gets read in the wrong order and arrives as a different argument.

One limit before the work starts. An archetype is a default that saves you from redesigning a deck's shape under time pressure, and defaults are worth a great deal when you have nine hours and a Thursday deadline. They are worth nothing when the situation is really unusual, and the last section here is about recognising that case, because the most expensive thing an archetype does is make a two-page answer look like it needs twenty-five pages of scaffolding.

Learning objectives

You can:

1. Name the three parts of the action-title consulting deck (executive summary, horizontal-logic body, indexed appendix), state the job each part does for a specific reader, and size each part against the meeting it serves.
2. Write an executive summary page that carries a governing thought and three or four key lines, each key line resting on one headline figure, and defend the decision to write that page last.
3. Build a body page to the four-element anatomy (action title, one exhibit, a source-and-date line, one supporting sentence), and reject a page that carries a second claim.
4. Apply the appendix contract: decide what leaves the body, index it so a partner can reach a page in one instruction, and answer a hostile question with a page number rather than a promise.
5. Structure a project readout in its five-part sequence (objective, findings, recommendation, next steps, owners and dates), and say why that order is fixed by governance rather than by rhetoric.
6. Write a findings page that puts evidence before conclusion, and keep one disconfirming finding in the deck with the reason it survives review.
7. Write next steps that are actionable: one owner by role, one date, one measurable first move, and the value the step carries with any double-count between steps disclosed.
8. Apply the two shipped templates to a live brief, filling the storyline placeholders before any exhibit is built, and check the result with the read-only-the-titles test.
9. Adapt either archetype for a hostile audience, stating precisely what changes (sourcing density, appendix depth, method placement) and what must not change.
10. Separate the pre-read from the presented cut, price both in reader-minutes, and diagnose the archetype failure where a decision memo would have served better than a deck.

Prerequisites & connections

Builds on. Nothing outside the archetypes branch is a gate. The pyramid, the action title and the read-only-the-titles test are assumed as working knowledge. If any of the three is unfamiliar, the storytelling branch (PR4.01, PR4.02) supplies them in a form you can use here directly. Chart-form choice comes up constantly below and is never taught here. PR2.01 owns it, and PR2.03 owns the one-message rule and the action title as chart craft. If you're arriving from the consulting galaxy with CN1.03 already done, you'll recognise the governing thought under a different name and can move faster through the executive-summary material.

Feeds forward. PR5.02 takes the archetype idea to the board deck and the investor pitch, where the reader's job changes from being persuaded to being accountable, and the structure changes with it. PR5.03 takes it to the research deck and the model walkthrough. PR6.01 delivers what is built here, including the Q&A habits that the appendix index exists to serve. PR7.01 runs the whole sequence end to end on one brief, and expects you to be able to use both templates from memory by then.

Five neighbours own material that touches this one, and none of it is rebuilt here. CN7.03 owns the consulting engagement capstone, which is the whole client engagement from scoping through to the final presentation, with the deck as one of its deliverables. What's owned here is the deck craft that capstone consumes. PR4.01 owns the executive storyline, the pyramid, SCQA and SCR, so the governing thought arrives here already formed. PR4.02 owns action titles, horizontal logic and vertical logic. The body section below assumes a title stack that already passes those tests, and shows what to hang under it. M10.03 owns the written analyst's deliverables, the memo and the one-pager, and the failure case at the end points there instead of competing with it. PR2.02 owns decluttering, colour restraint and the data-ink argument. Where an exhibit below is called clean, that's the standard being invoked, with nothing new added.

Core teaching content

1. What an archetype is, and why a deck shape is worth memorising

Anyone opening a deck spends the first fifteen seconds working out what kind of document it is. That's a different question from what it says. Until it's settled, they can't decide where to start, what to skip, or whether something missing is an omission or a convention.

An archetype settles it straight away. Here the word means a fixed skeleton with a fixed section order, a fixed page anatomy and fixed conventions about what goes where. It's shared widely enough in a professional community that a reader recognises it without being told. Three consequences follow, and they're the whole reason to learn one.

The first is navigation. A partner who knows the archetype opens to page 2 for the answer and to the appendix index for the evidence. They never read pages 3 to 12 unless the answer surprises them. Put your answer on page 19 and nobody reads as far as page 19. Somebody interrupts on page 4 to ask what the point is, and that's a reasonable response to a broken convention.

The second is that absence becomes information. When every body page carries a source line, a page without one has a number you can't check, and the reader knows to ask. When every next step carries an owner, a step without one is a step nobody has agreed to. In a document with no conventions a missing element is invisible, because there was no pattern for it to break.

The third is speed under pressure. The archetype takes about a dozen structural decisions out of a deck build. That matters most on the Thursday when you've got nine hours and a client call at four. You aren't designing a document. You're filling a known skeleton with the argument you already have, and the hours go into the argument.

That's also the archetype's cost, and it belongs in the same breath. A skeleton you didn't design is a set of decisions somebody else made, for a situation that may not be yours. Twenty-five pages of scaffolding around a single claim is an expensive way to say one thing. That case comes back at the very end.

10. Hostile audiences, sponsors, and what actually changes

The same content goes to two rooms. In the first, the sponsor commissioned the work, wants it to be right, and will help you defend it. In the second, the audience believes the work is a threat, has a prior conclusion, and will look for a reason to discard the whole thing.

Three things change, and all three are about the evidence path. The argument stays where it is.

Sourcing density goes up. In the sponsor version, 6 of the 14 exhibits carry a source line, which is 42.86%, and the rest inherit their sourcing from the appendix method page. In the hostile version every exhibit carries one, 14 of 14. A single unsourced number is the opening a hostile reader is looking for, and they'll spend the rest of the meeting there.

Appendix depth goes up. The sponsor version's appendix holds 8 exhibits, and the hostile version's holds 15, an increase of 7 and 87.50%. None of the additions is new analysis. They're the intermediate steps: the raw data extract, the reconciliation to the client's own reported numbers, the two alternative definitions you tested and rejected, and the sensitivity that shows the conclusion surviving each of them.

Method moves forward. In the sponsor version the method page is appendix position 3. In the hostile version it becomes a body page, position 3, and the body grows from 9 pages to 10 while the total goes from 25 to 30. It's a matter of sequencing. A hostile reader won't accept a finding from a method they haven't seen, so you get the method agreed before you present the finding, instead of defending it after the finding is challenged.

The conclusion, the figures and the disconfirming finding don't change. Softening a recommendation for a hostile room is the failure the whole apparatus exists to prevent. It's worse than useless, because a hostile audience can smell a hedge and will treat it as an admission. Tier-2 needs 750 orders a store a day against 185 today, in every room.

One more difference, and this one is about the meeting itself. With a sponsor you can present the front section and let the appendix answer questions. With a hostile audience you present the method, then one finding, then stop and take challenges before moving on. It's the same deck. You just walk a different path through it.

11. The pre-read and the presented cut

Two versions of one deck, with different word densities, and the arithmetic decides how different.

The pre-read is the full 25 pages at about 74 words a page, which is 1,850 words and 7.7083 minutes of reading at 240 words a minute. It has to stand alone, because the reader has no speaker. So every supporting sentence is written out, every exhibit is annotated, and the appendix index goes on the divider page where a reader can use it.

The presented cut is 9 pages at about 31 words a page, which is 279 words. That's an 84.92% reduction in words against the pre-read, and you get it by cutting pages. People get this part wrong by shrinking sentences instead. The cut drops the context page, the method, three of the body pages that support claims nobody will dispute in the room, and the whole appendix, which stays available on the machine.

Circulating a pre-read costs the audience real time, so put a price on it. Seven attendees reading a 25-page pre-read spend 53.96 reader-minutes between them, against 315 attendee-minutes for the 45-minute meeting itself. The pre-read is 17.13% of the meeting's cost, and it's what lets the meeting be a discussion. Without it, you get a reading.

Three failure modes show up here reliably. Send the presented cut as the pre-read and nobody can follow it, because 31 words a page assumes a speaker. Present the pre-read and you get the dense-slide collision measured in the foundations material. Or build only one artefact, at some compromise density around 74 words a page, and read it aloud. That produces the commonest corporate deck in existence, which serves the reader badly and the room worse.

The version that gets forwarded afterwards is always the pre-read. So if you build only one, build that one, and make the presented cut by deleting pages on the morning of the meeting. Deletion takes fifteen minutes. Expansion takes three hours.

12. Version marks, confidentiality and the things a master carries

Every page of a client deck carries three pieces of furniture. All three belong on the master, and none of them should be typed onto individual pages.

The confidentiality mark states the classification and the recipient in the terms the engagement letter uses. The engagement decides its form, and taste doesn't come into it. There are two errors, and they're equally bad. One is a deck with no mark, which can't be controlled once it's forwarded. The other is a mark so heavy that it competes with the content on every page.

The version string is a date and a sequence. It's there so that two people in a meeting can establish in four seconds whether they're looking at the same document. Any scheme works as long as it sorts and is unambiguous about the date. What doesn't work is "final", and its well-known descendants.

The draft mark is the one people forget to remove. A deck that goes out marked draft when it's final invites a reader to wait for the real one. A deck that goes out final when a number is still moving is a bigger problem. Decide which it is before you send it, and say so in the covering note as well.

All three go on the slide master, and that's the reason masters were covered before archetypes. Put them on the master and the version string on all 25 pages changes with one edit. Type them on individual pages and you'll ship a deck where page 17 says something different from page 16. That's the kind of small incoherence that makes a careful reader wonder what else wasn't checked.

13. Where the archetype is the wrong answer

The consulting deck is a machine for changing a mind across a long argument. Point it at a short one and the scaffolding costs more than the content.

The diagnostic is a count of claims. If the whole answer is one claim with one supporting exhibit, the archetype is wrong. Twenty-five pages carrying a single claim is 25 pages a claim. The same answer as a two-page decision memo is 2 pages a claim, a reduction of 92.00%. Send the deck version cold to nine readers and they spend 69.38 reader-minutes between them, with no discussion to show for it.

There's nothing to discuss. The reader agreed on page 2 and read the remaining 23 pages out of politeness.

Four situations call for something else, and recognising them early is worth more than any amount of deck craft.

A single decision with a single supporting analysis wants a decision memo: one page of recommendation, one page of evidence, a named decision date. The written-deliverables material in the analyst track owns that form, and it's the right one far more often than the deck-building habit suggests.

An update with no decision attached wants an email or a one-page dashboard. If the four recovery questions return "nothing is being decided", the deck is a ceremony.

An open, exploratory conversation, where the client's question isn't formed yet, wants a working session with a whiteboard and a handful of printed exhibits. Bring a polished deck into that room and it signals you've already decided. It'll get argued with for that reason alone.

And a deliverable that somebody who wasn't in the room will read six months from now wants a document with prose in it. The speaker pays for a deck's compression. Without a speaker, the reader is left doing the reconstruction, and whatever they reconstruct wrongly is on you.

Put plainly, the archetype is a default, and a default is a bet about the common case. It pays off most of the time. The skill worth having is noticing the day it doesn't, and that's usually the day the answer turned out to be simpler than the engagement assumed.

Common mistakes & how experts think differently

1. Writing the executive summary first and then not revising it. It's tempting, because the summary feels like the important page and starting there feels decisive. What actually happens is that the body evolves and the summary doesn't, so the deck argues with its own first page. The expert writes a throwaway summary in a text file to set direction, builds the body, reads the titles aloud, and then writes the real page from what the titles said.

1. Treating the appendix as storage. Everything that didn't fit gets pushed back, unordered and unindexed, and the deck is declared thorough. Then a partner asks a question and gets a promise to follow up, when what they wanted was a page number. The expert's appendix is a rehearsed instrument: 8 exhibits, an index phrased as questions, and a four-minute drill where somebody asks the six hardest questions and each answer is a page number said out loud.

1. Putting a chart on every body page. The instinct says a page without a chart looks thin. Follow it and you get the decorative bar chart of two numbers, which costs the reader attention and proves nothing. A third of the India deck's body pages carry no chart. Their claims are about decisions, which a short table carries better than any chart of quantities.

1. Leaving the source line off because the client already knows the data. They know the data. They don't know which extract, which definition, or which date. Nobody notices the gap until a number is questioned, and then the whole deck slows down while somebody checks. Experts write the source line at the moment they build the exhibit, while the definition is still in their head. Leave it for a final sweep and it won't be.

1. Presenting a readout in the consulting deck's order. Recommendation first feels stronger, and it is, for a reader who hasn't agreed to anything. A sponsor who commissioned the work is checking scope before merit, and will flip back looking for the objective. The expert knows which artefact they're building and doesn't mix the two spines, because the reader's first question differs.

1. Dropping the disconfirming finding. Almost nobody does this dishonestly. It gets cut because the deck was long, the finding was awkward, and the argument survived without it. The cost arrives later, when the sponsor finds it independently and discounts everything else by an unknown amount. Experts give it its own page and state what it costs the argument, which is usually less than it feels like.

1. Next steps with no owner or no date. "Implement a block release policy" is a wish. The four-part form is an action, an owner by role, a date, and a first move that can start without new approval. Experts name the owner by role. The deck outlives the individual, and a named person who wasn't consulted will decline in the room.

1. Adding up next-step values that overlap. The total is the most-quoted number on the page, and it's the one nobody checks. At Meridian Valley two rows share the same ENT block, worth $904,676, and the unadjusted total overstates by 9.38%. The expert prints both figures and the one-line reason, which costs nothing and buys the credibility of every other number in the deck.

1. Softening the recommendation for a hostile room. It feels like judgement and reads like a retreat. For a hostile audience you change the evidence path: every exhibit sourced, the appendix nearly doubled from 8 exhibits to 15, and the method moved from the appendix into the body at position 3. The conclusion and the figures don't move.

1. Using one artefact for the pre-read and the presentation. The compromise density, around 74 words a page, is too thin to read alone and far too dense to present. Experts build the pre-read, because that's the version that gets forwarded. Then they cut the presented version down to nine pages on the morning, which takes about fifteen minutes.

1. Reaching for the archetype when the answer is two pages. Twenty-five pages of scaffolding around one claim is 25 pages a claim against a memo's 2. The reader agrees on page 2 and reads on out of politeness. The expert counts the claims before choosing the container, and is comfortable handing a memo to a client who expected a deck, with one sentence saying why.

1. Typing the version string on individual pages. It works until page 17 disagrees with page 16, which happens on the third revision under time pressure. Furniture belongs on the master, where one edit changes all 25 pages, and where a page that is missing the mark is visibly missing it.

Worked examples

Every figure below comes from one computation, and the Verify block after the last example restates each one at full precision. Both companies are synthetic. Tez Dabba is a quick-commerce network with 440 dark stores in four city cohorts. Meridian Valley Medical Center is an eight-room operating suite. Where a chart is specified, the numbers in it are the numbers in the table above it.

Worked example 1: The executive summary page, written last (India, ₹)

The setting. Tez Dabba's board has asked for a view on whether the network is working. The month is 30 days. Four cohorts, with stores, orders per store per day, average order value, gross margin as a share of that value, last-mile cost an order, and dark-store fixed cost a store a month:

Solution. Step 1, get the per-order economics. Gross profit an order is AOV times gross margin. Fixed cost an order is the store's monthly fixed cost divided by its orders in the month, which is density times 30. Contribution an order is gross profit less last-mile less fixed.

Step 2, check it two ways. The cohort contributions sum to ₹4.6759 crore, matching the network figure computed independently as gross profit ₹60.74 crore less last-mile ₹22.56 crore less store fixed ₹33.51 crore. The residual is 6 hundred-millionths of a rupee on ₹4.68 crore. That's floating-point noise, nothing more.

Step 3, the first draft of the summary page, written before the body. It said: "Network contribution is positive at ₹4.68 crore a month and improving; the plan should continue with tighter cost control in the weaker cohorts." Every word of that is defensible from the total, and every word of it is wrong. The total is an average over a business that's really two businesses.

Step 4, build the body, then read the titles. Read alone, the body's nine titles make the argument in order. Contribution has risen every quarter, and one cohort earns all of it. Between them, the other three lose ₹7.4335 crore. Density drives it, and city tier doesn't. Each cohort has its own break-even density. Tier-2 is 4.05x short of its break-even, while metro expansion is 39.78 orders a day short and closeable. Closing Tier-2 recovers ₹3.7207 crore a month, and the recovered money is worth more in metro density than anywhere else.

Step 5, write the real summary from what the titles said.

The network is profitable in aggregate at ₹4.6759 crore a month, and the aggregate is hiding a business that has to be closed. Metro mature contributes ₹12.1094 crore a month, which is 2.59 times what the whole network keeps. The other three cohorts lose ₹7.4335 crore a month between them. Tier-2 needs 750 orders a store a day to break even against 185 today, a gap of 4.05 times that no operating plan closes. Ask: approve closure of the 95 Tier-2 stores at the March review.

Read it. Four claims, four figures, one ask, and the network total appears once. Notice what the page doesn't say. It doesn't say the business is failing, because it isn't. Closing Tier-2 lifts network contribution from ₹4.6759 crore to ₹8.3966 crore a month on 90.50% of the orders. That's ₹16.7147 an order against ₹8.4239 today. A Tier-2 density assumption is what could break the page. At 750 orders a store a day Tier-2 breaks even, so anyone who believes that density is reachable in Tier-2 towns has a different deck to write. The appendix has to carry the evidence that it isn't.

The checklist for this page. One governing thought that is a claim. Three or four key lines, each resting on one headline figure. No figure repeated. An ask stated as an approval with a date. Written after the body titles were final, and reconciled against them.

Worked example 2: A body page that could not carry its claim (India, ₹)

The setting. Page 2 of the body has to make one claim, that metro mature earns everything the network makes. The analyst's first version was a pie chart of order share across the four cohorts, titled "Order mix by cohort".

The wrong version, described. A four-slice pie. Metro mature 46.70% of orders, Tier-1 24.97%, metro expansion 18.84%, Tier-2 9.50%, with a legend to the right and percentages in a data-label callout. Title: "Order mix by cohort".

Solution. Step 1, name the mismatch between the claim and the exhibit. The claim is about contribution and the exhibit shows orders. Those are different quantities, and they rank differently. By orders the ranking is metro mature, Tier-1, metro expansion, Tier-2. By contribution an order it's metro mature, metro expansion, Tier-1, Tier-2. Take the ranking from the pie and you take the wrong ranking.

Step 2, name the form problem, which exists even if the data were right. Four slices means the reader compares angles. Tier-1 at 24.97% is 89.89 degrees of arc and metro expansion at 18.84% is 67.81 degrees. The gap of 6.13 percentage points is 22.08 degrees, right at the edge of what an eye reads reliably without labels. Above three parts, the pie is doing a job a sorted bar does better, and the chart grammar used across the track rules it out for that reason.

Step 3, name the third problem, which is the title. "Order mix by cohort" is a filing label. It says what's on the page without saying what the page claims. So the reader has to derive the claim themselves, and half of them will derive a different one.

Step 4, rebuild. One sorted bar, contribution an order, four bars, sorted descending, zero line drawn heavily because three of the four bars are negative, one colour for the positive bar and a single neutral grey for the rest.

Title: "Only metro mature makes money on an order: +₹46.72 against −₹70.57 at Tier-2." Source line: Tez Dabba store ledger, 30-day month to 31 August 2026; contribution = gross profit less last-mile less allocated store fixed; excludes central overhead and marketing. Supporting sentence: the spread across cohorts is ₹117.29 an order, and all of it is fixed cost per order, which falls only with density.

Read it. The rebuilt page makes one claim. The reader's eye now does the comparison along a common baseline instead of across angles, and the definitional caveat makes the number checkable. What it still doesn't show is why density differs by cohort. That's page 4's job, and putting it on this page is how a one-claim page becomes a two-claim page.

The checklist for this page. The exhibit measures the quantity the title claims. The form matches the message type, and comparison across categories takes a sorted bar. Sorted by value, since cohort order is not intrinsic. Colour carries the one distinction that matters. A source line with a date and the definition. One supporting sentence.

Worked example 3: The hinge page, and the title stack that proves it (India, ₹)

The setting. The argument has to get from "three cohorts lose money" to "close Tier-2 but fix metro expansion". Something has to explain why one is fixable and the other isn't. That explanation is the mechanism. Contribution an order is set by density, because density is what divides the fixed cost.

The wrong version, described. Page 4 was originally a table of the four cohorts against six columns, titled "Cohort unit economics detail", with every input and every output on one page. Everything true, nothing claimed.

Solution. Step 1, state the claim the page has to make. Contribution an order is set by density rather than by city tier, so a cohort's problem is arithmetic and not geography.

Step 2, build the exhibit that shows a relationship. A relationship between two continuous quantities takes a scatter. Twenty-four stores, six per cohort, at density offsets of −34%, −19%, −7%, +6%, +21% and +33% around each cohort's mean. Density on the horizontal axis, contribution an order on the vertical, cohort carried by marker shape so no second axis is needed, and the zero line drawn.

Step 3, read what the scatter says and what it does not. Density explains 87.90% of the variation in contribution an order across these 24 stores. The crossover is tight. No store below 438.90 orders a day is positive and no store above 475.20 is negative, so the break-even falls inside a band about 36 orders a day wide. That band differs by cohort because the fixed cost and the gross profit an order differ.

Step 4, title it. "Contribution per order is set by density, not by city tier: r = 0.94 across 24 stores." Supporting sentence: the band between the highest loss-making store at 438.90 orders a day and the lowest profitable store at 475.20 is where every cohort's own break-even sits.

Step 5, test the page by removing it. Read the nine titles with page 4 deleted. The stack now jumps from "the other three lose ₹7.4335 crore between them" to "each cohort has its own break-even density". That's a non-sequitur, since nothing has established that density is the variable that matters. So the page is holding the argument up, and finding pages like that is what the read-only-the-titles test is for.

Read it. A correlation of 0.9375 across 24 synthetic stores describes a constructed dataset. It isn't evidence about the real world, and the page says so in its source line. It does let you talk about break-even density on the next page. A cohort where gross profit an order is so low that no density clears the fixed cost would break it, and that's close to Tier-2's actual position at 750 orders a store a day.

The checklist for this page. The claim is a relationship, so the form is a scatter. The correlation is stated with its sample size. The zero line is drawn because the sign matters. The page is tested by deletion against the title stack. The source line says the data is synthetic.

Worked example 4: A findings page that led with the wrong lever (US, $)

The setting. Meridian Valley Medical Center runs eight operating rooms, five days a week, 480 minutes of prime time each, which is 320.0 hours a week. Service lines hold 270.0 hours of block, 84.38% of prime time, leaving 50.0 hours of open time. Twelve weeks of block data:

The peer turnover benchmark is 25 minutes. Of the 40 first cases a week, 54% start on time, and the late ones run a mean 23 minutes behind.

The wrong version, described. The first draft's finding page was titled "Turnover times exceed benchmark across all services", with a bar chart of mean turnover by service against a 25-minute benchmark line. It's true, it's vivid, and it points at the smallest of the three levers.

Solution. Step 1, size all three levers in the same unit, which is hours of room time a week.

Turnover excess is cases times minutes over benchmark, summed: 78(9) + 96(4) + 22(16) + 14(13) + 40(1) + 35(6) = 702 + 384 + 352 + 182 + 40 + 210 = 1,870 minutes, which is 31.1667 hours.

Released block is allocated hours never used: 270.0 − 198.8 = 71.2 hours.

Late first cases are 40 × 0.46 = 18.4 late starts at 23 minutes, which is 423.2 minutes, or 7.0533 hours.

Step 2, state the ratio that decides the recommendation. Released block is 2.28 times the turnover excess. A turnover programme that succeeded completely, taking every service to the 25-minute benchmark, would free 31.17 hours a week. A block-release rule that recovered 40% of unused block frees 28.48 hours. On effort, the two aren't close. One means changing the behaviour of six services in the room, and the other needs a scheduling rule.

Step 3, rebuild the page. One sorted bar of the three levers in hours a week, sorted descending, with the released-block bar highlighted and the other two neutral. Title: "Released block is 71.2 hours a week, 2.28 times the turnover excess." Source line: Meridian Valley block schedule and case log, twelve weeks to 29 August 2026; released block = allocated minus in-room hours; benchmark turnover 25 minutes. Supporting sentence: released block is hours a service holds and does not use, which nobody can book because the schedule does not release them until the day before.

Step 4, put the conclusion at the foot and keep it out of the title. The page's bottom line reads: the largest recoverable pool is scheduling, not behaviour. A readout's findings page states what was measured in the title and lets the reader reach the conclusion with the exhibit in front of them.

Read it. The 73.63% utilisation figure is what the sponsor asked about, and it's the least useful number on the page. It says the suite is short of a target without saying which of three things to do. A definition change would break the finding. If released block were measured after the day-before release and not at allocation, most of the 71.2 hours would move into open time. The recommendation would then be about booking instead of release. That's the reason the source line names the definition.

The checklist for this page. All levers converted to one unit before any of them is ranked. Exhibit sorted by size. Title states the measurement and leaves the conclusion out, and the conclusion goes at the foot. A definition in the source line, because the finding depends on it.

Worked example 5: The disconfirming finding, kept in (US, $)

The setting. The readout's recommendation is a block-release rule. The analysis contains one service that argues against the turnover story so strongly that leaving it out would be convenient, and one that argues against the utilisation story.

Solution. Step 1, find the finding that is uncomfortable. ENT has the best turnover in the suite at 26 minutes, one minute above the peer benchmark, and the lowest utilisation at 52.00%. Its turnover prize is 40 minutes a week, which is 0.6667 hours, and its released block is 12.0 hours a week. That's a ratio of 18.00 to 1.

Step 2, decide whether it damages the recommendation. It doesn't. ENT is the strongest single case for the block-release rule. The service that's best at the thing everybody talks about is the service with the most unused room time. The finding is uncomfortable only because it contradicts the framing the analyst brought into the engagement.

Step 3, find the one that does cut against the recommendation. Cardiac runs the highest utilisation in the suite at 89.00% and the worst turnover at 41 minutes. For cardiac, released block is 4.4 hours a week and turnover excess is 352 minutes, which is 5.8667 hours. It's the only service where the turnover prize exceeds the block prize, so any rule applied uniformly across all six services is wrong for cardiac.

Step 4, write the page. Title: "ENT has the suite's best turnover and its lowest utilisation, at 26 minutes and 52%." Exhibit: a two-column table of the six services, turnover minutes against utilisation, sorted by utilisation. Supporting sentence: cardiac is the mirror image at 89% utilisation and 41-minute turnover, and is the one service where a turnover programme beats a release rule.

Step 5, state what it costs the argument, on the page. The recommendation's scope narrows. The block-release rule applies to the five services other than cardiac, and cardiac gets a turnover pilot instead. That's a better recommendation than the one the analyst started with, and it exists only because the inconvenient service was kept in.

Read it. A disconfirming finding that changes nothing is a finding you haven't thought about hard enough. This one changed the scope of the recommendation from six services to five, and added a different intervention for the sixth. It doesn't weaken the headline, though. The block-release rule still addresses 65.07% of the pool.

The checklist for this page. The disconfirming finding has its own page and its own action title. The page states what it costs the argument. The recommendation is amended rather than defended. Nothing moves to the appendix to make the body tidier.

Worked example 6: Next steps, and the total that was 9.38% too big (US, $)

The setting. The recommendation is approved in principle and the readout needs a next-steps page. The suite runs 285 cases a week over 50 operating weeks, which is 14,250 cases a year at a contribution margin of $4,200 a case, or $59,850,000 a year. Mean in-room time is 198.8 hours over 285 cases, which is 41.8526 minutes a case. A slot is in-room time plus one benchmark turnover, which makes 66.8526 minutes, or 1.1142 hours.

Solution. Step 1, value the main step at three different capture rates.

The readout uses 40% as the base case, which is $5,367,747 a year and 8.97% more cases on the same rooms. For comparison, a turnover programme capturing 100% of the excess is worth 1,398.60 cases a year, or $5,874,114. So the block rule at 40% capture is 0.9138 times the full turnover prize, and the turnover prize needs every one of six services to change behaviour.

Step 2, value the other steps on the same slot arithmetic. ENT right-sized to an 80% target needs 13.0 / 0.80 = 16.250 hours of block against 25.0 today. That releases 8.750 hours a week, which is 12.29% of all released block and 392.65 cases a year, or $1,649,150. A parallel-induction pilot in two orthopaedic rooms attacks 78 cases at 9 minutes over benchmark, which is 702 minutes a week, or 525.03 cases a year and $2,205,149. A 07:00 first-case huddle attacks the 423.2 late minutes a week, which is 316.52 cases a year and $1,329,372.

Step 3, find the double count before somebody else does. ENT holds 12.0 of the 71.2 released hours, which is 16.85% of the pool. At the 40% capture rate in row 1, ENT's share inside row 1 is 4.800 hours a week, worth 215.40 cases a year and $904,676. Rows 1 and 2 overlap by that amount.

Step 4, write the page with the disclosure on it. Five rows, each with an action, an owner by role, a date, and a value. Under the table, one line: rows 1 and 2 overlap on ENT block, worth $904,676; the combined figure net of double-count is $9.65m against a printed total of $10.55m. The net figure is 16.12% of the suite's current annual contribution.

Read it. The disclosure line costs twenty-eight words and buys the credibility of the other four numbers. The capture-rate assumption is the one that could break the whole table. At 30% instead of 40%, the main row falls from $5,367,747 to $4,025,810. So the sensitivity goes in the appendix and gets named on the page. And no row here claims a margin improvement, because every value is incremental contribution on existing fixed cost, as the source line states.

The checklist for this page. Every row has a verb, an owner by role, a date and a value. Values computed on one shared slot definition. Any overlap between rows quantified and disclosed on the page. The key assumption named, with its sensitivity in the appendix.

Worked example 7: The pre-read and the presented cut, counted (India, ₹)

The setting. The Tez Dabba deck is 25 pages, 13 in the front section and 12 in the appendix, which is 48.00% of the document. The meeting is 45 minutes with seven attendees, of which 3 minutes go to introductions and 12 are reserved for discussion.

Solution. Step 1, compute the presenting pace for three candidate cuts. Presenting time is 45 − 3 − 12 = 30 minutes.

At a three-minute floor a page, 30 minutes carries 10 pages. The nine-page cut is the only one of the three you can present at a human pace.

Step 2, size the two artefacts in words. The pre-read runs about 74 words a page over 25 pages, which is 1,850 words and 7.7083 minutes of reading at 240 words a minute. The presented cut runs about 31 words a page over 9 pages, which is 279 words. The reduction is 84.92%, and it comes from deleting pages. Shortening sentences isn't how you get there.

Step 3, price the pre-read against the meeting. Seven readers at 7.7083 minutes each spend 53.96 reader-minutes. The meeting itself costs 45 × 7 = 315 attendee-minutes. So the pre-read is 17.13% of the meeting's cost, and it turns the meeting from a reading into a discussion.

Step 4, decide what leaves the presented cut. Four things come off. The context page goes, since the room commissioned the work. So does the method page, which is in the appendix. Three of the nine body pages go too, the ones whose claims nobody in this room disputes. And the whole appendix comes off too, though it stays on the machine. What's left is the title, the executive summary, six body pages including the hinge and the disconfirming one, and the next-steps page. That comes to the nine pages the pace in step 1 assumes.

Read it. The presented cut is a page subset of the pre-read. It isn't a summary of it, and that's what keeps the two artefacts consistent. You'd break this by editing a page in the presented cut without editing it in the pre-read. That's how the forwarded version comes to disagree with the version that was shown. Build the pre-read, cut on the morning, and never edit the cut.

The checklist for this pair. The presented cut is a page subset and never a rewrite. The pre-read stands alone with no speaker. Both priced in minutes before either is built. It's the pre-read that gets forwarded, so it's the one that must be right.

Worked example 8: Two versions for two rooms, and the day the archetype is wrong (India and US)

The setting. The Tez Dabba deck has to go to two audiences. The first is the chief operating officer who commissioned it. The second is a regional head whose Tier-2 expansion is the thing being closed and who has seen a version of these numbers before.

Solution. Step 1, change the evidence path and nothing else.

The seven added appendix exhibits are the intermediate steps: the raw store extract, the reconciliation to the company's own reported contribution, the two alternative allocation bases tested and rejected, the density-lift sensitivity, and the closure cost estimate at two assumptions.

Step 2, state what does not change. Tier-2 needs 750 orders a store a day against 185 today. Metro mature contributes ₹12.1094 crore. The ask is closure of 95 stores. A hostile room gets more evidence and the same conclusion.

Step 3, change the meeting rather than the argument. With the sponsor, present the front section and answer from the appendix. With the regional head, present the method page, stop, take challenges on the allocation basis, and only then present the first finding. Trust gets built in a different order, even though the pages are the same.

Step 4, the case where the archetype should not have been used at all. Suppose the whole question had been narrower. Say the board asked only whether Tier-2 should close, and the answer is one claim with one exhibit. The full archetype puts 25 pages of scaffolding around one claim, which is 25 pages a claim. A two-page decision memo is 2 pages a claim, a reduction of 92.00%. Circulated cold to nine readers, the deck version spends 69.38 reader-minutes and produces no discussion, because the reader agreed on page 2.

Read it. The hostile variant costs about five extra pages and a day of work, and you spend that day reconciling numbers. You don't spend it on design. What it can't do is survive a softened conclusion. A hostile reader who detects a hedge stops evaluating the evidence and starts evaluating the author. And the archetype-failure case says nothing against the archetype itself. It says you choose the container after you count the claims, and not before.

The checklist for both versions. Sourcing density, appendix depth and method placement are the three dials. The conclusion and the figures are not dials. The meeting path changes with the room. Count the claims before choosing the container.

Verify

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

For metro mature, gross profit an order is 585.00 × 0.220 = 128.70000000 and fixed cost an order is 950000 / (720 × 30) = 43.98148148. Contribution is 128.70000000 − 38.00 − 43.98148148 = 46.71851852. Orders are 120 × 720 × 30 = 2,592,000, and contribution a month is 121,094,400.00, or ₹12.10944000 crore. Break-even density is 950000 / ((128.70 − 38.00) × 30) = 349.13634693.

Metro expansion runs the same steps, 540.00 × 0.205 = 110.70000000, then 900000 / (410 × 30) = 73.17073171, then 110.70000000 − 44.00 − 73.17073171 = −6.47073171. Orders are 1,045,500 and the month's contribution is −6,765,150.00, or −₹0.67651500 crore. Break-even is 900000 / ((110.70 − 44.00) × 30) = 449.77511244, which leaves a gap of 410 − 449.77511244 = −39.77511244.

For Tier-1 the steps are 462.00 × 0.190 = 87.78000000, then 680000 / (330 × 30) = 68.68686869, then 87.78000000 − 41.00 − 68.68686869 = −21.90686869. Orders are 1,386,000 and the month's contribution is −30,362,920.00, or −₹3.03629200 crore. Break-even is 680000 / ((87.78 − 41.00) × 30) = 484.53755166. As a waterfall, it's 462.0000 − 374.2200 − 41.0000 − 68.6869 = −21.9069, with cost of goods at 81.0% of AOV.

Tier-2 works out as 395.00 × 0.175 = 69.12500000, then 520000 / (185 × 30) = 93.69369369, then 69.12500000 − 46.00 − 93.69369369 = −70.56869369. It has 527,250 orders and a monthly contribution of −37,207,343.75, or −₹3.72073438 crore. Break-even is 520000 / ((69.125 − 46.00) × 30) = 749.54954955, so the ratio to current density is 749.54954955 / 185 = 4.05161919 and the shortfall is 564.54954955 orders a store a day.

Across the network there are 5,550,750 orders. Gross profit is 607,436,486.25 (₹60.74364863 crore), last-mile 225,577,500.00 (₹22.55775000 crore) and store fixed 335,100,000.00 (₹33.51000000 crore). Contribution is 46,758,986.25 (₹4.67589863 crore), and 46,758,986.25 / 5,550,750 = 8.42390420 an order. Positive cohorts make 121,094,400.00 and the negative ones −74,335,413.75, which sum to 46,758,986.25 with a residual of −6.0e−8. Metro mature's share is 46.69639238% of orders and 54.91774162% of gross profit, and it makes 121,094,400 / 46,758,986.25 = 2.58975674 times the network total.

Close Tier-2 and 5,023,500 orders remain, which is 90.50128361% of today. Remaining contribution is 83,966,330.00, or ₹8.39663300 crore, +79.57% against today, and 16.71470688 an order against 8.42390420.

There are 24 stores in the scatter, at offsets −0.34, −0.19, −0.07, +0.06, +0.21, +0.33. Mean density is 411.25000000 and mean contribution −17.09514686. Pearson's correlation is 0.93754641 and r squared 0.87899327. Positive stores are 8 of 24 = 33.33333333%. Lowest positive density is 475.20 and highest negative density is 438.90. Across cohort contributions, the spread is 46.71851852 − (−70.56869369) = 117.28721221.

By quarter, Q1 had 4,100,000 orders and 8,800,000 of contribution, which is 2.14634146 an order. Q2 had 4,620,000 and 18,600,000 at 4.02597403 an order. Q3 had 5,110,000 and 32,100,000 at 6.28180039, and Q4 5,550,750 and 46,758,986.25 at 8.42390420. From Q1 to Q4, orders rose +35.38414634% and contribution +431.35211648%, while contribution per order rose +6.27756274, or +292.48%. Quarter-on-quarter steps are +1.87963257, +2.25582636, +2.14210381.

The pie arcs are 46.70% → 168.11°, 24.97% → 89.89°, 18.84% → 67.81° and 9.50% → 34.20°. The shares sum to 100.00%, and the closest pair differ by 6.13 points and 22.08°.

At Meridian Valley, prime time is 8 × 5 × 480 / 60 = 320.0 hours. Block is 270.0 (84.375%), open time 50.0 and used time 198.8. Utilisation is 198.8 / 270.0 = 73.62962963%, which is 6.37037037 points under 80%. Unused block is 71.2. Turnover excess is 78(9) + 96(4) + 22(16) + 14(13) + 40(1) + 35(6) = 1,870 minutes = 31.16666667 hours. Late starts are 40 × 0.46 = 18.4 at 23 minutes = 423.2 minutes = 7.05333333 hours. That makes a pool of 109.42000000, with shares of 65.06853409%, 28.48079509%, 6.45067082%. Released block to turnover excess is 71.2 / 31.16666667 = 2.28449198.

ENT's turnover prize is 40 minutes = 0.66666667 hours against released block of 12.0, a ratio of 18.00, at a utilisation of 52.00%. Cardiac runs at 89.00% utilisation, with turnover excess of 352 minutes = 5.86666667 hours against released block of 4.4.

A slot starts from mean in-room time, 198.8 × 60 / 285 = 41.85263158 minutes, so the slot is 41.85263158 + 25 = 66.85263158 minutes = 1.11421053 hours. Capture 30%: 21.360 hours, 19.17052 cases a week, 958.52622 a year, $4,025,810.11, 6.72650%. Capture 40%: 28.480, 25.56070, 1,278.03496, $5,367,746.81, 8.96867%. Capture 50%: 35.600, 31.95087, 1,597.54369, $6,709,683.51, 11.21083%. Turnover at 100%: 1,398.59865 cases and $5,874,114.31, for a ratio of 1,278.03496 / 1,398.59865 = 0.91379679. The annual base is 285 × 50 = 14,250 cases and $59,850,000.

ENT right-sizing needs 13.0 / 0.80 = 16.25 hours and releases 8.75 (12.28932584% of 71.2), which is 392.65470 cases a year and $1,649,149.74. Parallel induction recovers 702 minutes a week = 525.03543 cases a year = $2,205,148.80. Next, the first-case huddle recovers 423.2 minutes = 316.51708 cases = $1,329,371.75. Added up, the table total is $10,551,417.10. Then there's the overlap, 12.0 × 0.40 = 4.80 hours = 215.39915 cases = $904,676.43. Rows 1 and 2 come to $7,016,896.55 gross and $6,112,220.12 net. Net of that, the table comes to $9,646,740.67, so the overstatement is 9.37805%. Net as a share of current contribution is 16.11820%.

For the deck geometry, the front is 1 + 1 + 1 + 9 + 1 = 13 and the appendix 1 + 11 = 12, for a total of 25 and an appendix share of 48.00%. Of 14 exhibits built, 6 are in the body, which is 42.85714286%, and 8 in the appendix, at 0.72727273 a page over 11 pages. Presenting time is 45 − 3 − 12 = 30 minutes. That gives 30 / 25 = 1.20000000 minutes (72.00 seconds), 30 / 13 = 2.30769231 (138.46) and 30 / 9 = 3.33333333 (200.00), and a three-minute floor allows 10 pages. In words, the pre-read is 25 × 74 = 1,850 and 1,850 / 240 = 7.70833333 minutes. For the presented cut it's 9 × 31 = 279 and 1.16250000 minutes, a reduction of 84.91891892%. Seven readers spend 53.95833333 reader-minutes against 45 × 7 = 315 attendee-minutes, which is 17.12963%.

In the hostile variant, source lines go from 6/14 = 42.85714286% to 14/14 = 100%. The appendix goes 8 → 15, which is +7 or +87.50%, the body 9 → 10 and the total 25 → 30. For the archetype failure, it's 25 pages a claim against 2, a 92.00% reduction, and nine readers spend 9 × 7.70833333 = 69.375 reader-minutes.

Linter blends come out at 0.6(0.88) + 0.4(0.74) = 0.8240, 0.6(0.94) + 0.4(0.62) = 0.8120, 0.6(0.80) + 0.4(0.85) = 0.8200 and 0.6(0.72) + 0.4(0.95) = 0.8120. At rubric 0.62 the machine score needed is 0.92000000, and at machine 0.72 the rubric score needed is 0.92000000.

Practice set

Work each problem before reading its solution. A calculator is assumed. Numeric answers within 2% of the stated figure count as correct unless the problem says otherwise. Where a problem asks for a title or a page, the test is whether a reader could dispute what it says. The first three are guided and carry more scaffolding. The rest are bare.

P1 (guided). The body page that has to carry one claim (India). Tier-1 has 140 stores at 330 orders a store a day, an average order value of ₹462.00, a gross margin of 19.0%, last-mile cost of ₹41.00 an order and dark-store fixed cost of ₹6.80 lakh a store a month, over a 30-day month. Compute contribution an order. Then decide the exhibit form for a page whose claim is "the loss is fixed cost an order, not gross margin", and write the action title.

Solution. Gross profit an order is 462.00 × 0.190 = ₹87.7800. Fixed cost an order is 680,000 / (330 × 30) = ₹68.6869. Contribution is 87.7800 − 41.00 − 68.6869 = −₹21.9069. The claim is about the composition of a single number, and that takes a waterfall: five bars running AOV ₹462.0000, less cost of goods ₹374.2200 (81.0% of AOV), giving gross profit ₹87.7800, less last-mile ₹41.0000, less dark-store fixed ₹68.6869, closing at −₹21.9069. A serviceable title is "Tier-1 loses ₹21.91 an order, and ₹68.69 of it is dark-store fixed cost." A sorted bar of the four cohorts would be the wrong exhibit here. It answers a comparison question the page isn't asking.

P2 (guided). Rank three levers in one unit (US). Meridian Valley holds 270.0 hours of block and uses 198.8. Turnover excess over a 25-minute benchmark is 1,870 minutes a week. First cases number 40 a week, 54% start on time, and the mean delay when late is 23 minutes. Rank the three levers by hours a week and give each one's share of the total pool.

Solution. Released block is 270.0 − 198.8 = 71.20 hours. Turnover excess is 1,870 / 60 = 31.1667 hours. Late starts are 40 × 0.46 = 18.4 cases at 23 minutes, which is 423.2 minutes, or 7.0533 hours. The pool is 109.42 hours, with shares of 65.07% for released block, 28.48% for turnover and 6.45% for late starts. The ranking matters more than the arithmetic, because the exhibit that carries it is a sorted bar and the sort order is the finding. Converting to one unit first is what makes any ranking possible. Compare "71.2 hours" against "1,870 minutes" against "54% on time" and you're comparing three things that can't be ranked.

P3 (guided). Repair an executive summary page (India). A draft summary page reads: "Network contribution is ₹4.6759 crore a month, up from ₹0.88 crore in Q1. Contribution per order has improved every quarter, from ₹2.15 to ₹8.42. Order volume is up 35.4%. Metro mature is the strongest cohort. Tier-2 remains challenging. Recommendation: continue the plan with tighter cost control." Name four defects and rewrite the governing thought.

Solution. Four defects. First, there's no governing thought, just six sentences of equal weight and no claim any of them serves. Second, five figures appear and none of them is attached to a decision, so the reader can't tell which one matters. Third, "Tier-2 remains challenging" is a hedge that survives every outcome, when the fact is that Tier-2 needs 750 orders a store a day against 185 today, a gap of 4.05 times. Fourth, the ask isn't an ask, because "continue the plan with tighter cost control" names no approval, no owner and no date. A rewritten governing thought reads "The network is profitable in aggregate, and the aggregate is hiding a business that has to be closed." Under it go three key lines carrying one figure each, and an ask that reads "approve closure of the 95 Tier-2 stores at the March review".

P4. Density lift at the closeable cohort (India). Metro expansion runs 85 stores at 410 orders a store a day, AOV ₹540.00, gross margin 20.5%, last-mile ₹44.00, fixed ₹9.00 lakh a store a month. Compute its break-even density, the gap, the percentage density lift needed, and the contribution an order at a 20% lift.

Solution. Gross profit an order is 540.00 × 0.205 = ₹110.7000. Break-even density is 900,000 / ((110.70 − 44.00) × 30) = 449.7751 orders a store a day. The gap is 410 − 449.7751 = −39.7751, so the cohort is 39.78 orders a day short, which is a lift of 39.7751 / 410 = 9.70%. At a 20% lift, density is 492.0 a day, fixed cost an order is 900,000 / (492 × 30) = ₹60.9756, and contribution is 110.70 − 44.00 − 60.9756 = +₹5.7244 an order, worth ₹0.7182 crore a month. A 9.70% density lift is inside what a real operating plan can promise. So this cohort survives the recommendation, and Tier-2 doesn't.

P5. What one service can and cannot fix (US). Suite block utilisation is 73.63% on 270.0 hours allocated and 198.8 used. ENT holds 25.0 hours and uses 13.0. Compute suite utilisation if ENT alone reached 80%, and if every service reached 80%. Then say which of the two numbers belongs on a findings page.

Solution. ENT at 80% uses 0.80 × 25.0 = 20.0 hours, so the suite uses 198.8 − 13.0 + 20.0 = 205.80 hours and utilisation becomes 205.80 / 270.0 = 76.22%, a gain of 2.59 points. Every service at 80% gives 0.80 × 270.0 = 216.0 hours used, 54.0 unused, and (216.0 − 198.8) × 60 / 66.8526 = 15.4369 extra cases a week, or 771.8 a year. The second number belongs on a findings page, because it sizes the prize. The first goes in the appendix. It answers "what if only ENT moves", a question somebody will ask and nobody will ask first.

P6. Why the pie was rejected (India). Order shares are metro mature 46.70%, Tier-1 24.97%, metro expansion 18.84%, Tier-2 9.50%. Give each slice's arc in degrees, identify the closest pair, and state the rule the exhibit breaks.

Solution. Arcs are share times 3.6: 168.11°, 89.89°, 67.81° and 34.20°, summing to 360.00°. The closest pair is Tier-1 and metro expansion, 6.13 percentage points apart, which is 22.08 degrees of arc. The rule broken is the part-to-whole ceiling. A pie is acceptable for at most three parts, because above that the reader has to rank angles instead of lengths, and the ranking becomes unreliable at the very gap sizes that matter. The repair is a sorted bar, which puts every value on a common baseline. Note too that the shares summing to 100.00% is a necessary check and not a sufficient one. The exhibit is still wrong.

P7. Disclose the double count (US). Row 1 of a next-steps table is a block-release rule worth $5,367,747 a year at 40% capture of 71.2 released hours. Row 2 right-sizes ENT, releasing 8.75 of ENT's 12.0 unused hours, worth $1,649,150. Compute the overlap and the true combined figure, and write the disclosure line.

Solution. ENT holds 12.0 / 71.2 = 16.85% of released block. At 40% capture, ENT's contribution already inside row 1 is 12.0 × 0.40 = 4.80 hours a week, which is 4.80 × 60 / 66.8526 × 50 = 215.40 cases a year and $904,676. Rows 1 and 2 gross to $7,016,897. Net of the overlap, they're $6,112,220. Across all five rows the printed total is $10,551,417 and the net figure is $9,646,741, an overstatement of 9.38%. The disclosure line: rows 1 and 2 overlap on ENT block, worth $904,676; the combined figure net of double-count is $9.65m. Twenty-eight words, and they're the difference between a table a sponsor can quote and one a sponsor gets caught quoting.

P8. Turn a contents list into an index (India). An appendix divider currently reads: "A1 Methodology. A2 Store data. A3 Break-even. A4-A5 Scatter. A6 Quarterly. A7-A8 Closure. A9 Sensitivity. A10 Risks. A11 Sources." Rewrite three of those entries as index entries and say what makes the difference.

Solution. Written as questions, A1 asks how contribution is defined and what is excluded. A3 asks for the break-even density formula worked for each cohort. A7 to A8 ask what closing Tier-2 would cost, one time and ongoing. The first difference is that a partner under challenge is holding a question, and an exhibit name is the last thing on their mind. So an index phrased as questions can be searched by the thing in their head. The second difference is auditability. A question with no exhibit behind it is a visible gap, while "A10 Risks" conceals whether anything was actually tested. Eight exhibits over eleven pages is 0.73 exhibits a page, so several entries cover more than one page and the ranges must be explicit.

P9. Pick the archetype (US). A sponsor commissioned a twelve-week operating-room study, agreed the scope in writing, and now wants the results. An analyst proposes to open with the recommendation on page 2, in the consulting deck's order. State whether that is right and why.

Solution. It's wrong for this audience. A consulting deck opens with its answer because the reader has agreed to nothing and attention must be bought. A sponsor who commissioned the work is checking scope before merit. They'll flip back looking for the objective, and that costs more attention than the answer-first ordering saves. The readout's fixed sequence is objective, findings, recommendation, next steps, then owners and dates, and the first page states what was asked, what data was used, and what was excluded. The exclusion line here is that nothing about surgeon-level performance was analysed, which protects the work from a question it can't answer and sets the boundary before anybody tests it.

P10. Size the two artefacts (India). A 25-page deck runs about 74 words a page. The presented cut is 9 pages at about 31 words a page. The meeting is 45 minutes with seven attendees, of which 3 minutes are introductions and 12 are reserved for discussion. Compute the presenting pace for the full deck and for the cut, the word reduction, and the pre-read cost in reader-minutes at 240 words a minute.

Solution. Presenting time is 45 − 3 − 12 = 30 minutes. The full deck is 30 / 25 = 1.2000 minutes a page, which is 72.00 seconds. The cut is 30 / 9 = 3.3333 minutes, or 200.00 seconds. Words: 25 × 74 = 1,850 against 9 × 31 = 279, a reduction of 84.92%. The pre-read costs 1,850 / 240 = 7.7083 minutes a reader, and seven readers spend 53.96 reader-minutes against the meeting's 315 attendee-minutes, which is 17.13%. The cut is a page subset and not a rewrite, so the forwarded pre-read can't disagree with the presented version.

P11. Read the blended score (US). A slide scores 0.94 on the seven-rule machine check and 0.62 on the committed rubric self-score, blended at 0.6 machine plus 0.4 rubric. Give the composite. Then give the machine score that would be needed to pass at 0.80 with that same rubric score, and say what the pair of numbers is telling you.

Solution. The composite is 0.6(0.94) + 0.4(0.62) = 0.564 + 0.248 = 0.8120, which passes at 0.80. To reach 0.80 with a rubric of 0.62, the machine score needed is (0.80 − 0.4 × 0.62) / 0.6 = 0.9200, so this slide is barely above the line and is there on machine points alone. The pair says the page is technically immaculate and argumentatively weak. It's aligned, decluttered and correctly formed, and it makes a claim the author themselves scored at 0.62. That's the specific failure the blend exists to catch. The repair goes into the title and the claim, and the formatting can stay as it is.

P12. Diagnose the wrong container (India). A board asks one question: should the 95 Tier-2 stores close? The analyst builds the full 25-page archetype. State the diagnosis with the arithmetic, name the artefact that should have been built, and give the one situation in which the 25-page version would still be right.

Solution. One claim on 25 pages is 25 pages a claim. The same answer as a two-page decision memo is 2 pages a claim, a reduction of 92.00%. Circulated cold to nine readers the deck spends 9 × 7.7083 = 69.38 reader-minutes and produces no discussion, because the reader agreed on page 2 and read on out of politeness. The right artefact is a decision memo: one page of recommendation with the ask and the date, one page of evidence carrying the break-even arithmetic, and the supporting exhibits attached. The 25-page version becomes right the moment the audience is hostile. The regional head whose expansion is being closed will challenge the allocation basis, and then the seven added appendix exhibits and the method page moved into the body are what the meeting is actually about.

Applied mini-project

Build one archetype end to end from a supplied brief, on one of the two templates, and submit it with per-slide machine scores and a storyline check. Budget about four hours, of which no more than ninety minutes goes into building slides.

The brief. Choose the India brief or the US brief, and use the dataset given for it in the practice set above. For India, Tez Dabba's board wants a view on the network, and the decision on the table is whether to approve closure of the Tier-2 cohort at the March review. For the US, Meridian Valley's chief operating officer commissioned a twelve-week study of whether the operating suite can absorb more cases without building rooms, and wants the readout.

Step 1: write the title stack, in a text file, before opening any tool. Give it about 45 minutes. You want nine action titles for the consulting deck, or the five section headings plus three to five findings titles for the readout. Every title is a sentence with a verb, and every title that makes a quantitative claim carries its figure. No exhibit, no layout and no file exists at this point.

Step 2: run the read-only-the-titles test on yourself, then on somebody else. This takes about 15 minutes. Read the stack aloud in order. Mark the point where the argument jumps. Then delete each title in turn and ask whether the argument still stands. A title whose deletion changes nothing isn't holding anything up, so it comes out. A deletion that breaks the chain has found your hinge page.

Step 3: build the body pages on the template. Allow about 60 minutes. Four elements a page and no more: the action title, one exhibit, a source-and-date line carrying the definition the number depends on, one supporting sentence. Let the claim each title already makes choose its exhibit's form. The shape of the data doesn't get a vote.

Step 4: sweep the appendix and write the index. That's about 20 minutes. Every exhibit you built and didn't use goes back, and every entry in the index is phrased as a question a reader would hold. Include the sourcing page. Then run the four-minute drill. Have somebody ask the six hardest questions about the work, and answer each with a page number said out loud.

Step 5: write the executive summary or the recommendation page, last. Budget about 20 minutes. For the consulting deck, that's one governing thought and three or four key lines, each resting on one headline figure, and an ask that names an approval and a date. For the readout, it's the recommendation with its value and any overlap between next steps disclosed on the page.

Step 6: score and submit. This one takes about 20 minutes too. Run the seven-rule check on each slide and record the machine score. Commit your own rubric self-score for each slide before revealing the rubric. Submit the deck, the per-slide machine scores, the blended composites at 0.6 machine plus 0.4 rubric, and a one-page storyline check listing your titles in order with the hinge page named.

What to hand in. The deck, in both its pre-read form and its presented cut, with the page count of each. The title stack as plain text. The appendix index. The per-slide scores. And one paragraph naming the exhibit you built and did not use, with the reason it went to the appendix.

Rubric. Twenty-four points, passing at 19 with no zero on any row.

Any row whose evidence is missing scores zero, however the deck looks. A beautiful deck whose numbers don't reproduce scores zero on row 8 and fails, and that's intended. Presentation is the last thing graded here.

Reading & resources

Core, do these.

• Barbara Minto, The Pyramid Principle: the source of the governing thought and the key-line structure that the executive summary page implements. Read the sections on introductions and on grouping. The material on deduction and induction repays a second pass once you have built a deck. [Paid] [Intermediate]
• Gene Zelazny, Say It With Charts: written for this very artefact, and the tightest published statement of the idea that the message chooses the chart form. The chapter organisation maps almost one to one onto the exhibit decisions in the worked examples above. [Paid] [Beginner]
• Cole Nussbaumer Knaflic, Storytelling with Data: the best single treatment of the action title and of what to remove from an exhibit. Its companion exercise volume is the better buy if you want repetitions rather than argument. [Paid] [Beginner]
• Stephen Few, Show Me the Numbers: the reference for tables and graphs, and where to go when you want the reason behind a rule rather than the rule. Particularly useful for the readout's findings pages, which lean on tables more than a consulting deck does. [Paid] [Intermediate]

Going deeper.

• Jean-luc Doumont, Trees, Maps, and Theorems: covers documents, slides and speech in one compressed frame, and is unusually good on the difference between a document that is read and one that is presented. Short and worth rereading. [Paid] [Advanced]
• Edward Tufte, The Visual Display of Quantitative Information: read it for the standard of care rather than for the prescriptions, several of which the field has since argued with. The data-ink argument is what "declutter" means. [Paid] [Intermediate]
• Alberto Cairo, How Charts Lie: the clearest short treatment of chart deception, including the axis and dual-axis problems that the corpus grammar rules out. Useful when you are building a hostile-audience version and need to anticipate the challenge. [Paid] [Beginner]
• Nancy Duarte, Resonate: the audience-first case made from a design direction rather than an analytical one. Read it for the audience material. The presentation-as-story frame is a poor fit for a governance readout and a good one for a persuasion deck. [Paid] [Beginner]
• Garr Reynolds, Presentation Zen: the argument for restraint on a presented slide, and the reason the presented cut runs at about 31 words a page when the pre-read runs at 74. [Paid] [Beginner]

Standards, sources and free material.

• The US Securities and Exchange Commission, A Plain English Handbook: free from the commission's site, written for disclosure documents, and the cheapest available improvement to a dense supporting sentence. [Free] [Beginner]
• W3C Web Content Accessibility Guidelines: the contrast minimum for normal-size text is 4.5 to 1, which is the standard the seven-rule colour check is written against (WCAG 2.2, checked September 2026; confirm, since guideline versions advance and the ratios can move).
• Your own firm's engagement letter and document-classification policy: the confidentiality mark, the retention rule and the permitted distribution list are set there and nowhere else. Read the actual policy once, and don't copy the mark from an old deck (organisation-specific; check with your own risk or legal function, since classification schemes differ and change).
• Operating-room efficiency measurement: block utilisation, turnover time and first-case on-time start definitions differ between institutions and between published benchmark sets, and the 25-minute turnover benchmark used above is illustrative rather than authoritative (checked September 2026; confirm against the benchmark set your own institution subscribes to before quoting any figure).

Do this, not just read. Take the last deck you built or received. Hide every page except the titles and read them in order, out loud. Then find its appendix, pick the hardest question you can think of about the work, and time how long it takes you to reach the page that answers it. Under ten seconds is an appendix. Over a minute is storage.

Teach it back & journal

Feynman prompt. Explain to a competent colleague who has never built a client deck why the same analysis, presented in two different containers, produces two different meetings. Use the arithmetic, and leave the adjectives out. A 25-page deck in a 45-minute slot is 72 seconds a page once introductions and reserved discussion are taken out. That's why the document is cut into a front section sized to the meeting and an appendix sized to the questions. A nine-page presented cut runs at 200 seconds a page, a pace a person can actually hold. Then make the harder point, which has nothing to do with pages. The archetype is a set of promises: the answer is on page 2, every number has a source line, every next step has an owner and a date, and everything you might ask is in the appendix and reachable in one instruction. A reader who knows the promises reads faster and trusts more, and a reader who finds one promise broken slows down on all of them. Finish with the case against your own argument. A single claim on 25 pages is 25 pages a claim against a decision memo's 2, so you choose the container after you count the claims. If your colleague can tell you unprompted why the executive summary is written last, you have taught the structure and not just the page count.

Journal prompt. Find the last deck you built, or the last one somebody sent you, and audit it against the archetype. Your taste doesn't count here. Read only the titles, in order, with the pages hidden, and write down the argument they make. If they make none, note where the first jump happens. Then count three things: how many body pages carry more than one claim, how many exhibits carry a source line with a date, and how many next steps carry an owner by role and a date. Write the three counts down as fractions, because "most of them" is how a deck stays broken. Now find the finding you left out. There's almost always one, and the hard part of the exercise is naming it: the number that did not fit the story, the service that was the exception, the cohort whose data was inconvenient. Write down what it was and why it went. Be specific about whether it was cut for length or for comfort. Those feel identical at 11pm, and they aren't the same thing at all. Finish by writing the one-sentence governing thought that deck should have had, and compare it with the sentence on its first page. If the two differ, you've found the gap between what you concluded and what you shipped, and closing that gap is what the archetype is for.

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