The Executive Storyline: Minto Pyramid for Decks, SCQA and SCR

Presentation track, branch pr-storytelling (Storytelling) · Node PR4.01 · ~10 focused hours · Mastery gate ≥ 85%

Forty minutes, twenty-two slides, and the recommendation sits on slide 19. At an even pace that is 1.818 minutes a slide, so the sentence the meeting exists to settle arrives at minute 34.5 and leaves 5.45 minutes for everything the room wanted to say about it. Move the same sentence to slide 2 and the discussion gets 36.4 minutes, six and two-thirds times as much. Not one number in the deck changed.

The slot is not safe either. Board agendas run long, the item before yours overruns, and a forty-minute slot becomes eighteen minutes often enough that you should price it. Say that happens a third of the time. The chronological deck then delivers its recommendation in 67 out of 100 meetings, because eighteen minutes takes the room to slide 9 of 22. The answer-first deck delivers it in 100 out of 100, a 50% improvement in the probability that the thing you were hired to say gets said.

None of that turns on charisma, and all of it is priced in minutes. A storyline is an ordered set of claims. A deck without one is an ordered set of topics, and the difference shows up as minutes of discussion, as the probability a recommendation survives a shortened slot, and as the number of questions a reader can answer from the summary page alone.

The structure taught here comes from Barbara Minto's The Pyramid Principle, written for consultants in the 1970s and still the tightest account of how an argument should be ordered for a busy reader. Her frame has three ideas in it. Every document has one governing thought at the top. Every level below it exists only to answer the question the level above provokes. And the points at any one level must be a set: no overlaps, no gaps, one ordering principle. Adapted for a deck rather than a memo, that produces a spine you can test mechanically, which is why it can be graded by a drill rather than by taste.

There is a limit worth stating before the hours start. A pyramid does not make a weak argument strong. Structure a bad recommendation perfectly and you get a bad recommendation that is now easy to attack, which is arguably an improvement but is not what the learner is hoping for. What the structure buys is different: it makes the argument's weakest point visible to you before it becomes visible to the room, and it stops good analysis from being lost inside a sequence nobody can follow.

The two anchors below run the whole way through. Nirala Fresh (synthetic) is an Indian quick-commerce operator with 186 dark stores, losing ₹5.58 crore a month, whose board has already approved 75 more stores. Harborline Outfitters (synthetic) is a US specialty retail chain whose EBITDA margin has fallen from 11.4% to 5.77% and whose turnaround plan has five points, two of which are secretly the same point. Both carry real arithmetic, and the arithmetic is where the storyline gets its authority.

Learning objectives

You can:

1. Write a governing thought for a body of analysis and defend it against three tests: it is a claim someone in the room could contradict, it answers the question the reader actually holds, and an action with an owner and a date follows from it.
2. Distinguish a thought from a topic using the negation test, and reject candidate governing thoughts that are subject headings, restatements of the work performed, or truisms that survive every outcome.
3. Build a three-level pyramid (governing thought, key line of three to five claims, support) and state what belongs at each level and what does not.
4. Apply the MECE test horizontally across a key line, testing all pairs for overlap, checking exhaustiveness against the frame you declared, and naming the residual rather than hiding it.
5. Apply the vertical test between levels, by writing the question each box provokes and confirming that the boxes below answer that question and no other.
6. Write an SCQA opening for a stated audience, and find the complication by locating the number that moved, the decision that is due or the commitment that no longer holds.
7. Compress SCQA into SCR and say when the compressed form is right, pricing the saving in words and reader-seconds.
8. Order a key line inductively or deductively, compute the expected surviving value of each under stated rejection probabilities, and defend the choice.
9. Write an executive summary that is the pyramid's top two levels and passes a read-alone test against a named list of questions a decision-maker must be able to answer from it.
10. Diagnose and repair a failing storyline by running the two read-alone passes, naming which of the three standard breaks has occurred, and rebuilding the level that caused it.

Prerequisites & connections

Builds on. Nothing inside the storytelling branch, which starts here. You should be able to write a one-sentence takeaway and name a reader and a decision, which is what PR0.01 teaches. The work goes easier if you've already sketched a deck as a ghost deck in PR3.02, since a storyline is what a ghost deck is a sketch of. Neither is a gate. Any analysis of your own will serve as raw material: a teardown from M8.02, a case answer from the CN modules, a valuation, a market study.

Feeds forward. PR4.02 takes the key line built here and turns each of its claims into a slide title, then tests the title sequence as a whole. PR4.03 takes the finished pyramid and asks what a narrative arc adds to it. PR5.01 to PR5.03 apply the structure to the consulting deck, the board deck, the investor pitch and the research report, and each of those bends the pyramid in its own typical way. PR6.01 owns what happens when the room interrupts. PR7.01 runs the sequence end to end on an unseen brief. The executive one-pager and SCQA summary template that ships with the storytelling branch is built from the top two levels of the pyramid you construct here.

Six neighbours own material that touches this one, and the boundaries are drawn on purpose so that nothing below rebuilds them. CN1.03 owns answer-first synthesis for a case: the so-what ladder, the hypothesis tree and the interview-to-insight path that produce a governing thought in a client engagement. Its hand-off note draws this same line from the other side. What's owned here is the same logic applied to a deck, where the levels become slides, the key line becomes a title sequence and an appendix has to carry a contract. M10.03 owns the analyst's written deliverables, the one-page memo and the longer written note, where the pyramid is expressed in paragraphs instead of boxes. PR4.02 owns titles: the wording of a claim on a slide, the two-line rule, hedging, and the read-only-the-titles test applied to a whole deck. Everything below stops at the claim and hands the sentence on to PR4.02. PR2.03 owns the single chart's action title and the one-message rule inside an exhibit. PR0.01 owns the audience profile and the medium choice that come before any structure at all. Where a section names one of those ideas, it names it and moves on.

Core teaching content

1. The governing thought, and the test that it is a thought

At the top of any deck worth building sits one sentence. Everything else in the deck exists to make that sentence believable, and anything that doesn't is in the appendix or isn't in the pack at all.

Nirala Fresh shows how easily that sentence goes wrong. The finance team has three months of work behind it: a store-level P&L, a cohort study, a central-cost benchmark. The founder asks for the board deck. Three candidate sentences come back from three different analysts.

The first reads "Store economics and the path to profitability." That's a topic. It names a subject and asserts nothing about it, and the giveaway is that you can't disagree with it. A reader can disagree with a claim, be persuaded by evidence, and change a decision. Nobody has ever been persuaded out of a subject heading.

The second reads "We analysed 186 stores across three vintage cohorts and benchmarked central costs against four listed peers." That's a description of the work performed. It's worse than the topic, because it looks like content and is really a receipt. Boards don't buy receipts. And it answers a question nobody asked, which is what you did with your three months.

The third reads "Pausing all new store openings for two quarters takes Nirala Fresh to company EBITDA breakeven in month 5.9 on the 186 stores it already has, and preserves ₹88.3 crore of cash against the ₹20.4 crore deficit the approved plan produces."

Now that's a governing thought. Somebody can hold the opposite view, and one director will. It's falsifiable, because a slower maturation curve or a higher central cost breaks it. And an action follows. The board owns it, the date is this meeting, and there's a specific thing to stop.

The negation test is the fastest of the three checks and the one to run first. Write the opposite of your sentence and read it aloud. If the opposite is absurd, like "we should not analyse store economics", you have a topic. If the opposite is unintelligible, you have a fragment. If the opposite is a sentence a competent colleague could argue for over lunch, you have a thought.

Two failure modes slip past the negation test, so they need naming on their own. A truism passes it formally but not usefully. "Growth is strong but cash is tight" has a coherent opposite, yet no set of facts could ever refute it, because both halves are always true of something. A compound sentence with three clauses joined by "and" is usually three governing thoughts wearing one coat. Build a deck from it and you get three arguments running in parallel and finishing at different places.

There's a fourth property the three tests don't capture, and experienced writers apply it anyway. The governing thought has to answer the question the reader is actually holding. Nirala's board is holding the question "do we approve the 75 openings in the FY27 plan, and if not, what do we do instead?" Nobody in that room is waiting to hear whether the unit economics are good. A perfectly defensible claim about unit economics that never reaches the approval question is a good sentence in the wrong deck.

2. The pyramid: what belongs at each level

The structure has three levels and no more, for a deck of the size an executive audience will sit through.

At the top, the governing thought: one sentence, one claim, one action implied. Below it, the key line: three to five claims which, taken together, force the governing thought on anyone who accepts them. Below each key-line point, the support: the exhibits, computations and primary-source facts that make that one claim believable.

The arithmetic of the shape is worth seeing, because it settles the deck length before you build anything. A key line of three points with two supports each is ten boxes. Four points is thirteen. Five points is sixteen. Six points is nineteen and seven is twenty-two. That's the twenty-two-slide deck from the opening paragraph, and now you can see why it couldn't reach its recommendation before minute 34.5. Its key line was carrying seven points because nobody made it carry fewer.

Three to five is the working range for a decision deck, and a psychology finding about working memory has nothing to do with it. A key line of six or more almost always contains an overlap, and someone in the room will find it. Every extra point adds pairs to test: five points is ten pairs, seven points is twenty-one. Getting down to four forces the merge you were avoiding.

At the key-line level, each claim has to be the same kind as its siblings. Nirala's key line has three points, and each is an assertion with a number in it.

Notice what isn't at the key-line level. There's no point called "Background", none called "Methodology", and none called "Next steps". Those are deck furniture, with no argument in them, and they belong either in the appendix, in the summary's final clause, or nowhere. A key line whose first item is "Context" has already conceded that the argument doesn't start until item two.

Support is the level where analysts feel most comfortable, and it's where the most damage is done. An exhibit that's interesting without being probative pulls the room away from the claim it was meant to serve. One rule keeps support in line, and it comes with vertical logic, below. A support box proves its own parent, and if it also proves something else, that's a second deck.

4. Vertical logic: every level answers the question above it

Horizontal logic makes each level a set. Vertical logic turns the levels into a hierarchy, so you don't end up with three lists stacked on top of each other.

The relation between two levels works as a dialogue. A box at any level provokes a single question in the reader's mind. The boxes directly beneath it exist to answer that question. Write the question down. If the boxes below answer a different question, the level is misparented, and no amount of polish on the exhibits will fix it.

Nirala's governing thought provokes "why would pausing produce that?" All three key-line points answer it, in order: because the economics work, because the loss is mix and overhead, because the alternative burns the cash. Each key-line point in turn provokes its own question. Point one, that the economics work, provokes "how do you know?" Its support answers with the maturation curve, the ₹549.66 breakeven line and the fact that 111 of 186 stores sit above it. Point three, that the approved plan ends the year at minus ₹20.4 crore, provokes "how much and by when?" Here the support answers with the twelve-month cash bridge under both plans.

Three questions cover most of the cases. Why? is provoked by a claim that something is true. How? is provoked by a claim that something should be done. How much? is provoked by any claim carrying a magnitude, which in an analyst's deck is nearly all of them. A fourth question, so what?, runs the other way. It's what a reader asks of a fact, and its answer is the level above. A deck full of facts that nobody has asked so-what of is a deck with no top level, whatever slide 2 says.

The vertical test catches a very common defect, which is support that's true, interesting and irrelevant. Harborline's freight point claims $3.088m from an 8% renegotiation on a $38.6m spend. Beneath it sits a beautiful exhibit showing that freight cost per unit rose 41% over two years. That exhibit proves freight got expensive. But the parent claims 8% is recoverable, and the exhibit doesn't prove that. A procurement director will say so in the room. You can re-parent the exhibit under a different claim, or replace it with one that proves the parent: the three carrier bids, the mode-shift analysis, the contract expiry dates.

There's a cheap way to run the test on a finished deck. Cover the body of every slide and read only the top level and the key line. Then cover the key line and read one key-line point with only its own support visible. If you can't say, in a sentence, what question the support is answering, the reader won't manage it either. And they won't have your three months of context to help them.

7. Leading with the answer, and the two cultures that resist it

Answer-first ordering is the recommendation, then the reasons, then the evidence. Everything in the two previous sections assumes it, and most analysts arrive resisting it.

The resistance isn't irrational. It comes from two distinct places, and each needs its own response. First comes the scientific and academic culture, which orders a report as method, results, discussion. It does that because its reader is a peer reviewer whose job is to find the flaw in the method before being told the conclusion. If you trained through a research degree, you've had that order rewarded for years, and in that setting it's correct.

A decision meeting is different. Nobody in that room is auditing your method. They're deciding something, and they'll decide it with or without your method.

Then comes the hierarchical or high-context organisation. In it, stating a conclusion before the most senior person in the room has indicated a direction reads as presumption, however clear you meant to be. It's a real constraint, and telling people to be more direct doesn't solve it. It shows up in family-owned Indian businesses, in Japanese corporates, in parts of the public sector everywhere. Pretend otherwise and junior analysts get into trouble that has nothing to do with their analysis.

The move that works in the second setting keeps the structure and changes who owns the sentence. Attribute the question to the room, then answer it, then hand the decision back. "The question the board left with us in March was whether the opening plan is affordable at the current burn. Our answer is that it is not, and the alternative that costs least is a two-quarter pause. The board's call is whether the pause is two quarters or three." Your recommendation is still on slide 2. Now it arrives as an answer to their question. It no longer sounds like an instruction, and the final clause hands back the decision rights that the direct form seemed to take.

There's one fair exception to answer-first, and you should use it rarely enough that you notice yourself doing it. If the answer will be rejected on sight, because it contradicts something the decider has said in public or has personally sponsored, the build-up buys a hearing the bare conclusion wouldn't get. You already know what that choice costs. Your recommendation arrives at minute 34.5 instead of minute 3.6, discussion falls from 36.4 minutes to 5.45, and if the slot is cut the recommendation is never reached at all. Pay that price on purpose, for a reason you could state to your own manager, or don't pay it.

Answer-first doesn't mean stating the conclusion with more confidence than the evidence carries. Harborline's plan closes 93.2% of the margin gap and leaves 0.384 points unresolved. Said answer-first, that sentence gives both halves in the same breath. Leading with the answer and overstating it are separate acts, and mixing them up is how a good habit gets a bad reputation.

8. Grouping and ordering: deductive against inductive

A set of points can support a conclusion in two ways, and only two. Choosing between them is a real decision, with a cost you can compute.

A deductive group is a chain. Its first point states a situation, the second comments on it, and the third draws the consequence. Freight rates rose 41%. Our contracts expire in March. Therefore we must retender or absorb the increase. Each link depends on the one before, and if the premises hold, the conclusion follows necessarily.

An inductive group is a set of same-kind claims that share a plural noun. Harborline's five levers are inductive. Each is a lever, each is independent of the others, and the conclusion is a sum. Nothing has to be inferred. That plural noun gives it away. If you can write "the five levers", "the three constraints", "the four markets" over the group, it's inductive. If you can't, and the points only make sense read in order, it's deductive.

Inductive usually wins in a deck, and you can compute the reason. Give each of Harborline's five levers the probability that the room rejects it. Realised price is the most aggressive and the most likely to be pushed back, at 0.30. Store closures are politically hard but well evidenced, 0.15. Freight, with three bids in hand, 0.10. Corporate overhead, 0.20. Shrink and markdown, the newest and least worked, 0.35.

Under grouping, each lever survives or falls on its own, so the expected surviving value is the sum of each lever's value times its survival probability. That comes to $24.432m of the $32.123m plan. Under chaining, the argument survives only if every link holds. That happens with probability 0.27846, so the expected surviving value is $32.123m times 0.27846, or $8.945m. Against the same audience and the same evidence, the grouped structure is worth 2.73 times the chained one.

Hold on to the mechanism behind that ratio. It's the practical reason, and the arithmetic only proves it. When a chained argument loses a link, the whole recommendation goes with it. When a grouped argument loses a point, the recommendation shrinks. Harborline's worst single loss under grouping is the price lever at $12.253m, which is 38.1% of the plan. So the meeting carries on, and the remaining 61.9% still gets decided. Under chaining, a director who doesn't believe the freight number has just killed the store closures too, and neither of you will notice that's what happened.

Deduction still has its cases. Use it when the argument turns on a single causal chain the reader hasn't accepted yet, where the whole point is that A forces B forces C. Use it when the conclusion is counter-intuitive and the reader needs walking through it. And keep it short. In a deck, nobody will follow a deductive chain longer than three steps, and Minto's own advice is to summarise a long deduction inductively one level up.

Ordering within a group is a separate decision from grouping, and three principles cover almost everything. Time order follows a process or a chronology. You use it when the sequence is itself the argument. Structural order follows a division of the thing being discussed, by geography, by business unit, by process step. Use it when the reader already holds that division in their head. Degree order ranks by importance or by size, largest first. It's the default for a lever list, because it puts the money where the attention is.

Declare the principle and hold it. A five-point key line has 120 possible orderings, and, ties aside, only one of them follows your declared principle. So the storyline drill can grade order mechanically. It's also why a reader who can't infer your principle by point three stops trying and starts reading the slides as separate documents.

9. The executive summary is the pyramid's top two levels

An executive summary holds the governing thought plus the key line, written out. Don't treat it as a shortened version of the deck. It has to work as a standalone artefact for the director who won't open the deck at all.

Two properties make it standalone. Every claim carries its own number, because a claim without its number sends the reader into the deck to find it, which defeats the purpose. And the summary contains no forward references: no "as shown on slide 7", no "see appendix B", nothing that only resolves elsewhere.

Nirala's summary runs 148 words and reads in 40.4 seconds. Its first draft ran 37 words and read in 10.1 seconds, and it was a list of section headings: store economics, cohort analysis, central costs, capital plan, recommendation. It was faster to read, and useless. That's the trade the topic version always offers.

To test it, write the list of questions first, before the summary exists, naming what a director must be able to answer from the summary alone. For the Nirala board there are six.

1. What are we being asked to decide.
2. What changes if we say yes.
3. How much cash that saves, and by when.
4. The strongest argument against it.
5. What would have to be true for the recommendation to be wrong.
6. Who owns the next step and when they report back.

The topic version answers one of the six, which is 16.7%. Written as claims it answers all six, and that's what "readable alone" means in practice.

That last question, the one about ownership, is where most summaries stop a sentence early. A recommendation with no owner and no date is a suggestion, and suggestions don't get minuted. Adding eleven words turns the summary from a document the board reads into a document the board acts on.

Length matters more here than anywhere else in the deck. A summary that runs to 400 words has stopped being the top two levels and started including support. Now the reader has to do the selecting you were paid to do. If the summary won't fit in about 150 words, the usual cause is a key line of six or seven points, and the repair is upstream, back at the key line.

10. The appendix contract

Treat the appendix as a contract, and stop thinking of it as storage. The contract has one clause: every numeric claim the summary makes has exactly one exhibit behind it that a sceptic can check.

Put that way, the appendix becomes something you can finish building. A summary making eleven numeric claims needs eleven exhibits. Nirala's team arrived with twenty-three appendix slides, and four of them were opened in the room. It's tempting to conclude the appendix was 17.4% useful. That's the wrong measure, and it leads to the wrong fix, which is to build a smaller appendix next time and get caught out.

Score it against the contract instead. Of the eleven numeric claims in the summary, eight had a supporting exhibit and three did not: the maturation curve's source, the central-cost benchmark and the assumed transfer of orders between stores. Those three are 27.3% of the claims. Question any one of them in the room and you'd have got the answer no analyst wants to give. One exhibit per backed claim accounts for eight of the twenty-three slides, so the other fifteen were carrying nothing the summary had promised. So the appendix was too big and incomplete at the same time. That's the normal state of an appendix built up by accumulation, with no contract behind it.

Build order follows from the contract directly. Write the summary. Underline every number in it. For each underlined number, name the one exhibit that would satisfy a sceptic, and build that exhibit. Then, and only then, look at what's left over from the analysis and ask whether any of it answers a question the summary provokes but doesn't settle. Most of it won't. Leave that out of the pack, and keep the working file where you can reach it in ninety seconds.

Two conventions make the contract usable in a live meeting. First, index the appendix by claim, never by workstream. A director who challenges the ₹549.66 breakeven figure should get the exhibit named after that claim, and a folder called "unit economics" won't help them. Then put the source and date on every appendix exhibit, because the question that follows "where did that come from" is always "as at when".

11. Testing a storyline before anyone else does

There are three passes. Run them in this order, and each takes under two minutes on a finished deck.

The first pass reads only the governing thought. Alone, out of context, cold. It must be a sentence that could be contradicted. It must answer the question the reader holds, and the one you hold doesn't count. And an action must follow from it. This pass fails on the second test more often than on the other two, and the author can't see the failure, because they've spent three months with the question they chose.

Pass two reads only the key line, in order, with the governing thought and all support hidden. It must read as a set of same-kind claims, in a declared order, that together force the governing thought. Check three things while you read. See whether any point restates another in different words. Look for a step where the argument needs a claim that isn't there. And make sure the last point arrives at the conclusion and doesn't trail off into next steps.

Pass three reads the top two levels together as the summary, against the question list you wrote for the executive summary. It comes last because a summary can read beautifully while resting on a key line with a hole in it. You want pass two to find the hole, while it's still cheap.

Nirala's nine-slide draft shows what pass two catches. Its label titles ran 19 words in total, averaging 2.1 words each: "Store economics", "Cohort analysis", "Central costs", and so on. Read alone, they make no argument at all, because a list of nouns can't make one. Rewritten as claims, the same nine titles ran 153 words, averaging 17.0, and reading the whole stack took 41.7 seconds. Now set that against a deck body of 14 slides at about 95 words, which is 1,330 words and 6.0 minutes of reading. So the claim stack is 11.5% of the reading cost and carries the entire argument. A director who reads nothing else gets the case in 42 seconds.

Wording those titles well is a craft of its own, and it comes next. Keep the two jobs apart. This pass tests whether the claims form a coherent sequence. Whether each claim is put in the best available sentence, at the right length, with the right hedge and the right number, is a separate craft with its own rules.

12. Rebuilding a storyline that failed the test

Three breaks account for nearly every failure, and each has a different repair. Diagnosing which one you have is most of the work, because the repairs are cheap and the misdiagnosis is expensive.

The first break is a key line of topics. Pass two turns up a list of nouns and no argument. Repairing it is mechanical. Take each topic, find the one claim you'd actually make about it, and write that claim with its number. Nirala's "Store economics" becomes "a store crosses breakeven at 549.7 orders a day and 111 of 186 are past it". Then re-read the converted list. Turning topics into claims routinely shows that two of them are now the same claim, and the key line shrinks from five points to three. That shrinkage is the good outcome, even if it feels as though something went wrong.

The second break is a key line that doesn't force the governing thought. Every point is a claim, every claim is true, and a reader who accepted all of them could still reasonably reject the conclusion. It comes in two shapes. A gap means a necessary claim is missing. You repair it by adding the claim, and that usually means going back to the analysis. Going back to the deck won't fix it. A leap means the conclusion asserts more than the points support. You repair that by weakening the conclusion until it fits, which nobody enjoys and everybody should practise.

Harborline's original governing thought claimed the plan restored the 11.4% margin. Corrected, the levers reach 11.02%, so the accurate governing thought names 11.02% and the 0.384-point residual. And the storyline is stronger for it, because the residual is now a named piece of work and can't ambush anyone.

The third break is misparented support. Here the key line is fine, the conclusion follows, and underneath one of the points sits an exhibit proving something next door. Harborline's freight exhibit is the standard case: a 41% cost increase shown under a claim that 8% is recoverable. There are two repairs, and choosing between them is a judgement about what the deck is for. Re-parent the exhibit if the thing it proves is a claim the deck needs somewhere. Replace it if the parent claim really has no evidence. Then you've found a hole in the analysis, and finding it in your own rehearsal is worth more than the hour it costs.

When you retest a rebuilt storyline, start again from pass one, whichever pass failed. Repairs move things, and a key line repaired at point three has changed what point four is answering.

13. Where this sits against the case answer

The consulting galaxy owns answer-first synthesis for a case. The pyramid, the so-what ladder, the hypothesis tree and the habit of driving from a hypothesis to a defensible recommendation are taught there, in the setting of an engagement where the deliverable might be a conversation, a memo, a model or a deck.

Here you get the same logic under three constraints that only a deck imposes. First, the levels become an ordered sequence of slides, which forces the argument to be linear even where the analysis wasn't. The key line becomes a title sequence that a reader will meet alone, at speed, possibly on a phone. That puts a wording burden on each claim that a spoken case answer never carries. And there's an appendix. It's a place to put material without losing it, and so it needs the appendix contract.

In practice the seam works like this. If you can already produce a governing thought for a case, everything in sections 1 and 2 is revision for you. Spend the time on sections 3, 8 and 10, which are the deck-specific parts. If you can't, build it here on a deck and the case version will follow. Horizontal and vertical testing is the harder half of the skill. Vocabulary is the easy half.

14. Where an assistant helps, and where it takes the judgement you are graded on

An assistant helps on two of the tests here, and it's dangerous on the one that matters most. It's good at the mechanical parts of horizontal logic, like reading a key line and reporting which pairs of points might touch the same quantity, or which claim has no number attached. It can't decide what your governing thought should be. That decision is the one thing the room will ask you to defend, and you can't defend an answer you didn't choose.

{
  "skill": "Stress-testing a key line for overlap, gaps and misparented support before the deck is built",
  "use": "After you have written the governing thought and the key line yourself, paste the pyramid as plain text, one line per box with its number, and ask the assistant to run the horizontal and vertical tests as an adversary. Use its output as a list of things to check by recomputation, never as a rewrite.",
  "tools": ["Chat assistants (Claude, ChatGPT, Gemini)", "A plain text file holding the pyramid, so no slide exists yet", "A spreadsheet or Python session holding the numbers behind each claim"],
  "prompt": "Here is a pyramid. Governing thought: <ONE SENTENCE>. Key line: (1) <CLAIM + NUMBER> (2) <CLAIM + NUMBER> (3) <CLAIM + NUMBER> (4) <CLAIM + NUMBER>. Support under each point: <LIST>. Do five things and nothing else. 1) For every pair of key-line points, say whether they could be acting on the same underlying quantity, and name the quantity. 2) Name any claim I would need to accept the governing thought that is not in the key line. 3) For each support item, state which key-line claim it proves; if it proves a different claim, say which. 4) State the ordering principle my key line appears to follow, and name any point that breaks it. 5) Write the single strongest sentence a hostile reader could say against my governing thought. Do not rewrite my sentences.",
  "verify": "Every overlap the assistant names is a hypothesis, not a finding: recompute the two claims jointly in your own spreadsheet and see whether the combined figure differs from the sum, as Harborline's price and promotion lines do by $5.639m. Every gap it names must be checked against the frame you declared, because an assistant does not know your scope and will invent claims that are out of it. Every re-parenting suggestion must be tested by writing the question the parent provokes and reading the support against it. If the assistant's hostile sentence is one you cannot answer with a number you already hold, that is the finding of the session, and it is yours to fix in the analysis rather than in the wording.",
  "diy": "You must be able to build the pyramid, run the pair test and the vertical test, and write the SCQA opening cold, with no tool open, from a fact set and a stated audience. That is what the mastery gate examines and what the storyline drill grades on parent, child and order. An assistant that supplies your governing thought has taken the one sentence you will be asked to defend, and a governing thought you cannot defend is worse than none, because it will be delivered with confidence.",
  "market": "IN"
}

Common mistakes & how experts think differently

1. The key line is a list of the workstreams. It's tempting to write the key line from the project plan, because the project plan already exists and it feels organised. You end up with a deck whose second level names your team's activities. Your argument's claims don't appear, and the reader has to build the argument themselves. An expert writes the key line from the conclusion downward, then checks which workstreams produced no claim. A workstream with no claim in the key line is either an appendix or a piece of work that shouldn't have been done.

2. Two key-line points act on the same quantity. Overlap is tempting because two actions really do feel different. Raising a list price and cutting a promotional calendar are different meetings, different owners, different systems. They're still the same rupee or dollar. Experts test all pairs mechanically, ten pairs for a five-point key line, and ask of each pair what quantity it moves, in what units. Harborline's pair moved realised price per unit on the same SKUs, and the sum overstated by 46.0% of the true figure.

3. Exhaustiveness is claimed rather than framed. Saying a key line is MECE without saying what it's exhaustive of makes the claim unfalsifiable, and so worthless. An expert declares the frame in the same breath as the key line, in the form "these are the levers available inside the existing store base over eight quarters". Then the omission of shrink and markdown becomes visible in ten seconds, long before you're standing in the room.

4. The residual is padded away. When five levers reach 93.2% of a target, the tempting repair is to find another 6.8% inside the existing levers by adjusting an assumption. That option is nearly always available, and nearly always dishonest. An expert names the residual, sizes it at $2.354m or 0.384 points, and assigns an owner to closing it. A named residual survives cross-examination. A padded lever doesn't, and the padding usually turns up in the one lever the questioner knows best.

5. The situation is used to establish credibility. Openings that recite the company's history, the team's methodology or the data's provenance are asking the reader to accept the author before accepting the argument. It reads as confidence and works as delay. An expert opening states one thing the reader has already signed, and spends the credibility budget on the complication instead. For Nirala, that one thing is the March approval of 75 stores and ₹86.25 crore.

6. The complication is an adjective. "Competitive intensity is increasing" and "margin pressure continues" are complications the writer couldn't find, dressed up as ones they did. Ask whether a number moved, a date arrived or a commitment broke. If none of the three is true, there may be no complication. Then there's probably no decision either, and a short written update will do the job better than a meeting.

7. Support is chosen because it is the best exhibit rather than the right one. A striking chart that proves an adjacent claim is worse than a plain one that proves the parent, because it invites a question the parent can't answer. Experts write the question each box provokes on the box itself during the build, and delete any support that doesn't answer it. The 41% freight increase was the best-looking exhibit in Harborline's pack and the one that cost the most time in the room.

8. The deck leads with the answer and the summary does not. This is a common half-conversion. Slide 2 states the recommendation, but the one-page summary that goes to people who never see the deck is still a set of section headings. People who read only the summary are usually the ones who can't attend, and that correlates with seniority. Experts write the summary first, as the top two levels, and then expand it into the deck.

9. Answer-first is confused with confidence-first. Leading with the answer says the recommendation early. It doesn't say it more strongly than the evidence carries. A governing thought that claims a full margin restoration when the levers reach 11.02% is wrong, however direct it sounds. An expert states the conclusion and its residual in the same sentence, which is earlier than the alternative and more accurate too.

10. The appendix is built from what is left over. Sweep the working files into the back of the pack and you get an appendix that's large, unindexed and still missing the three exhibits that matter. That's a fair description of Nirala's twenty-three slides. Experts build the appendix from the summary's underlined numbers, one exhibit per claim, indexed by the claim, never by the workstream.

Worked examples

Seven examples, four Indian and three American, all of them built on the two synthetic anchors. Every figure comes from the computation restated in the Verify block below. Work each one yourself before you read the solution.

Worked example 1: Three candidate governing thoughts, two of them topics (India, ₹)

Nirala Fresh operates 186 dark stores in six cities. The order economics are: average order value ₹487, merchandise gross margin 21.4%, and ₹46 of pick, pack and last-mile cost per order. Each store carries ₹9,60,000 a month of rent, staff and power. Stores mature along a curve the operations team has fitted from three years of data. It runs 310 orders a day at month 3, 480 at month 6, 620 at month 9, 700 at month 12. From there it climbs to 890 at month 18, 1,090 at month 24, 1,240 at month 30, 1,310 at month 36, flattening at 1,340 by month 42. The three vintage cohorts are 48 stores at an average age of 30 months, 63 at 18 months and 75 at 6 months. Central cost runs ₹14.20 crore a month. Cash is ₹96 crore. In March the board approved 75 more stores at ₹1.15 crore of fit-out each.

Three analysts propose three governing thoughts. Choose one, and say what is wrong with the other two.

• (A) Store economics and the path to profitability.
• (B) We analysed 186 stores across three vintage cohorts and benchmarked central costs against four listed peers.
• (C) Pausing all new openings for two quarters takes Nirala to company breakeven in month 5.9 on the stores it already has, and ends FY27 with ₹88.3 crore of cash against the minus ₹20.4 crore the approved plan produces.

Solution. Step 1, compute what any of the three would have to rest on. Contribution per order is 487 × 0.214 − 46 = ₹58.218. A store's monthly contribution is that times 30 days per order per day, which is ₹1,746.54 for each order a day it does. Breakeven against ₹9,60,000 of fixed cost is 960000 / 1746.54 = 549.658181 orders a day, and the maturation curve crosses that level at month 7.492675.

Step 2, place the cohorts. At 30, 18 and 6 months the three cohorts run 1,240, 890 and 480 orders a day. Store-level EBITDA per store is 1240 × 1746.54 − 960000 = ₹12,05,709.60, ₹5,94,420.60 and −₹1,21,660.80. Cohort totals are ₹5,78,74,060.80, ₹3,74,48,497.80 and −₹91,24,560, so store-level EBITDA is ₹8.6198 crore a month. Against ₹14.20 crore of central cost, the company loses ₹5.5802 crore a month. That's ₹66.96 crore a year, and it gives 17.203684 months of runway on ₹96 crore.

Step 3, run the negation test on each candidate. The opposite of (A) is "not store economics and the path to profitability", which isn't a sentence. (A) is a topic. The opposite of (B) is "we did not analyse 186 stores". That's a sentence, but it's a claim about the team. It says nothing about the business, and no director's decision changes on it. (B) is a receipt. The opposite of (C) is "pausing openings does not get us to breakeven in month 5.9, or does not preserve that cash, or costs more than it saves". A competent director could hold that position and argue it. (C) is a thought.

Step 4, check (C) against the reader's question. The board's live question is whether to proceed with the 75 approved openings. (C) answers it directly. (A) and (B) answer questions the board did not ask.

Step 5, verify the numbers (C) commits to. Under a pause, weighted orders across the three cohorts rise as they age, and company EBITDA crosses zero at t = 5.890564 months. Cumulative twelve-month EBITDA is +₹3.442372 crore. Maintenance capital at ₹6 lakh a store a year is ₹11.16 crore. So cash ends at 96 + 3.442372 − 11.16 = ₹88.282372 crore.

For the approved plan you need a few more assumptions. Its 75 stores open evenly, so 75t/12 of them are trading in month t at an average age of t/2. A store younger than three months is held at the curve's first fitted point of 310 orders a day. And central cost ramps in a straight line to ₹15.20 crore a month by month 12. On those assumptions cumulative EBITDA is −₹16.751598 crore, fit-out is ₹86.25 crore and maintenance is ₹13.41 crore on the 223.5 stores trading on average, so cash ends at −₹20.411598 crore. The gap between the two is ₹108.693970 crore.

Read it. The minus ₹20.4 crore is what turns (C) from an opinion into a governing thought, because it's the number a director can't walk past. Notice what (C) doesn't claim, too. It doesn't say the pause is the best possible plan, only that it beats the approved one on cash and reaches breakeven without new capital. A governing thought that claimed optimality would need a comparison against every alternative, and the analysis doesn't contain one.

The checklist to apply. Write the opposite of your sentence and read it aloud. Check that the opposite is arguable and not absurd. Confirm the sentence answers the question your reader is holding, in their words. Confirm an action, an owner and a date follow. Make sure every number in it comes from a computation you can show. A rounded figure lifted from an earlier slide doesn't count.

Worked example 2: The SCQA opening, and where the complication was hiding (India, ₹)

Same company, same figures. Write the opening for the board pack, and identify the complication.

Solution. Step 1, reject the draft opening. The team's first draft reads "Nirala operates 186 dark stores in six cities. Order volumes have grown strongly. Store contribution is improving with cohort age. We have analysed the FY27 capital plan." Every clause is true and none of it is a situation, because a situation is something the reader already accepts and would nod at. A board doesn't nod at "order volumes have grown strongly". It wonders which volumes, and against what.

Step 2, find the situation the board has already signed. In March the board approved 75 new dark stores and ₹86.25 crore of fit-out. Every director voted for it. That's the platform.

Step 3, look for the complication in the three standard places. Start with a number that moved: the FY26 cohort matured more slowly than the plan assumed. At 480 orders a day it sits 69.658181 orders a day below the 549.658181 breakeven line, which is 1.492675 months of ageing away. Next, a decision that became due: the first ten leases of the FY27 programme are signed in six weeks. Last, a commitment that no longer holds: the approved plan spends 86.25 / 96 = 89.843750% of the cash on hand on fit-out alone, and twelve months of it ends the year at −₹20.411598 crore. The third is the strongest, because it says the company can't execute the plan its own board passed.

Step 4, write the four moves. Situation: in March the board approved 75 new stores and ₹86.25 crore of fit-out, which is 89.8% of the cash we hold. Complication: twelve months of that plan ends FY27 at minus ₹20.4 crore of cash, because the FY26 cohort is maturing about six weeks behind the curve the plan assumed. Question: what replaces the approved opening schedule. Answer: a two-quarter pause on all new openings, which reaches company breakeven in month 5.9 on the existing 186 stores and ends FY27 with ₹88.3 crore.

Step 5, price the compression. Run together as prose for the pack, those four moves are 101 words:

In March the board approved a plan that adds 75 new dark stores and spends ₹86.25 crore of fit-out capital, which is 89.8% of the cash we hold today. Since then the FY26 cohort has matured more slowly than the plan assumed, and twelve months of that plan now ends the year at minus ₹20.4 crore of cash. The question for Thursday is therefore what now replaces the approved opening schedule. Our answer is a two-quarter pause on all new openings, which reaches company breakeven in month 5.9 on the existing 186 stores and ends FY27 with ₹88.3 crore of cash.

At 220 words a minute that reads in 101/220 × 60 = 27.5455 seconds. Drop the question, and the same four facts run 54 words and 14.7273 seconds:

The plan we approved in March spends ₹86.25 crore of fit-out, or 89.8% of our cash. Twelve months of it ends FY27 at minus ₹20.4 crore. We are pausing all new openings for two quarters instead, so breakeven arrives in month 5.9 on the 186 stores we have and FY27 closes with ₹88.3 crore.

The cut is 47 words, 46.534653% of the text, and 12.8182 seconds a reader. Across nine recipients that's 1.922727 person-minutes. That isn't why you do it, though. You do it because 54 words get read on a phone before the meeting and 101 words get opened afterwards.

Read it. Finding the complication took no new analysis. All the team did was divide one number it already had by another number it already had. That's the usual case, so start the search with the commitments the reader has already made, and come to the data after.

The checklist to apply. Write the situation as one sentence the reader has already signed. Test the complication against the three sources: a number moved, a date arrived, a commitment broke. Confirm the question is the reader's question and is narrower than the one you started the analysis with. Confirm the answer is your governing thought word for word. Then write the SCR version and see whether you lost anything the reader needed.

Worked example 3: A five-point key line, two points that are one (US, $)

Harborline Outfitters runs 214 US specialty retail stores. FY26 revenue is $612.4m at a 38.2% gross margin, so gross profit is $233.9368m. SG&A is $198.6m, so EBITDA is $35.3368m, a margin of 5.770216%. Three years ago the margin was 11.4%. The turnaround team proposes five actions.

1. Raise list prices 3% on 340 SKUs carrying $284m of realised revenue. Worth $8.52m.
2. Cut the promotional calendar from 22 events to 12, moving promoted volume from 41% to 26% at an average 22% discount. Worth $9.372m.
3. Close the 38 stores below $1.6m of sales: $47.2m of revenue at a four-wall contribution of minus $3.9m, with 18% of sales transferring to nearby stores at a 34% four-wall margin, and $11.4m of one-time exit cost.
4. Renegotiate a $38.6m freight spend, with three bids in hand, for an 8% saving.
5. Cut corporate overhead by $5.4m.

Test the key line horizontally and rebuild it.

Solution. Step 1, size the gap the key line has to cover. Target EBITDA at an 11.4% margin is 612.4 × 0.114 = $69.8136m, so the gap is 69.8136 − 35.3368 = $34.4768m, which is 5.629784 points of margin.

Step 2, test all pairs for exclusivity. Five points give C(5,2) = 10 pairs. Nine of them touch different quantities. Only the pair (1, 2) doesn't. Both move realised price per unit on the same 340 SKUs. That's the overlap, and you have to net it, because adding the two figures counts the same money twice.

Step 3, net it properly. Realised revenue on those SKUs is $284m with 41% of volume promoted at 22% off, so list-equivalent revenue is 284 / (0.59 + 0.41 × 0.78) = 284 / 0.9098 = $312.156518m. Raise list 3% and it becomes $321.521213m. Cut promoted volume to 26% and realised revenue before any volume response is 321.521213 × (0.74 + 0.26 × 0.78) = $303.130200m, an effective realised price increase of 6.735986%. At an assumed elasticity of minus 0.8, units fall 5.388789%. Revenue comes in at $286.795154m, and COGS falls from $175.512m to $166.054029m. Gross profit goes from $108.488m to $120.741125m, so the two actions together are worth $12.253125m.

Step 4, size the error. The naive sum was 8.52 + 9.372 = $17.892m. The overstatement is $5.638875m, which is 46.019895% of the true figure and 31.516181% of what the plan claimed. Merge points 1 and 2 into one key-line point called realised price, and state the combined figure.

Step 5, compute the surviving points. Store closures avoid $3.9m of four-wall loss and add 47.2 × 0.18 × 0.34 = $2.88864m of transferred gross profit. So the lever is $6.78864m against $11.4m of exit cost, a payback of 1.679276 years. Freight is 38.6 × 0.08 = $3.088m. Overhead is $5.4m. The four corrected levers total $27.529765m, which is 4.495389 points and covers 79.850116% of the gap. Uncorrected, the five covered 96.205680%. No wonder nobody had questioned the plan.

Step 6, test exhaustiveness against the declared frame. The frame is the EBITDA bridge from 5.77% to 11.4% using levers available inside the existing store base over eight quarters. Shrink runs at 2.4% of revenue against a peer level of 1.5%, a gap of 0.9 points or $5.5116m. The markdown rate has moved from 7.1% to 9.8%, a gap of 2.7 points or $16.5348m. Neither had a key-line home. Taking a third of the shrink gap and a sixth of the markdown gap as capturable inside eight quarters gives 1.8372 + 2.7558 = $4.593m.

Step 7, restate the key line. Five points again, but a different five.

One convention has to be written on the slide. Don't leave the reader to assume it. In the table, the points column prices every lever against the FY26 revenue base of $612.400000m, and that's what makes the bars sum to the gap. Two of the levers also move that base. Closing the stores removes 47.2 × 0.82 = $38.704000m of revenue net of the 18% that transfers, and the price action adds 286.795154 − 284 = $2.795154m. A margin quoted after those denominator effects is higher than 11.02%. So the bridge holds the base fixed and gives the denominator its own bar, where nobody can mistake it for part of a lever.

Read it. The corrected plan reaches 11.015605% on the fixed FY26 base, short of 11.4%, and covers 93.172118% of the gap. And the residual of $2.354035m is named, so nobody has absorbed it into a lever. Add the bars and the residual and they sum exactly to the gap. That's your arithmetic check that the key line is now both exclusive and exhaustive on its own terms. Look at the ordering, too: the table runs by size, largest first, which is the declared principle. Freight moves from fourth in the draft to fifth, because now it's being ranked, where the draft only listed it.

The checklist to apply. List the pairs and name the quantity each pair moves. Net any pair that moves the same quantity, and recompute rather than subtracting an estimate. Declare the frame in one sentence. Check the frame for levers with no key-line home. Sum the points and the residual and confirm they equal the gap exactly. Re-sort under one declared ordering principle.

Worked example 4: The exhibit that proves the wrong claim (US, $)

Harborline's corrected key line goes into the deck. Under key-line point 5, freight, sits the exhibit the logistics team is proudest of: a line chart of freight cost per unit over 24 months, rising 41%. Under the deck as a whole sits a second exhibit: a bar chart of the five levers, sorted by size, drawn before the price merge and therefore showing six bars totalling $37.76m.

Diagnose both and specify the replacements.

Solution. Step 1, write the question each parent provokes. Key-line point 5 claims that 8% of a $38.6m freight spend is recoverable, worth $3.088m. The question it provokes is "how do you know 8% is available?" But the rising-cost chart answers a different question, which is "did freight get expensive?" It's true, it's well drawn, and it proves the wrong thing. A procurement director will accept the 41% and then ask where the 8% comes from, and the deck has no answer visible.

Step 2, specify the replacement. The exhibit that proves the parent is the bid comparison. It shows three carrier bids against the incumbent rate on the top twelve lanes, with the weighted saving computed and the contract expiry dates shown. Those dates matter, because a saving you can't access for eighteen months isn't in an eight-quarter frame. Sort it by lane spend, largest first. Give it an action title that states the claim: eight percent of the freight spend is available on lanes whose contracts expire inside the plan window. As for the 41% chart, it moves to the appendix, indexed under the claim it does support, which is the reason freight is on the key line at all.

Step 3, diagnose the lever chart. A sorted bar of six levers totalling $37.76m shows the double count as though it were money. Two of its bars, $8.52m and $9.372m, are the same $12.253125m seen twice. Any reader adding the bars gets a plan $5.638875m larger than the one that exists.

Step 4, specify the replacement. The message is a decomposition of a change, from a 5.770216% margin to a target, so the form is a bridge: a starting bar at the current EBITDA, one bar per lever in the declared order, a residual bar, and a closing bar at the target. Six intermediate bars, in points of margin: realised price 2.00084, store footprint 1.10853, corporate overhead 0.88178, shrink and markdown 0.75000, freight 0.50425, and residual 0.38440. Those six sum to 5.629784 points, which equals the gap exactly. That identity is the reason a bridge is the right form. A reader can add the bars and arrive at the target, and a plain sorted bar won't let them do that.

Step 5, apply the colour rule. Five lever bars in one neutral tone, the residual bar in a second tone that reads as still unresolved, and the two end bars in the darker anchor tone. Colour carries the distinction between what the plan delivers and what it doesn't, and that's the one distinction the chart exists to make. There's no third colour, because there's no third category.

Read it. The two defects are the same defect at two levels. A support exhibit proved a claim adjacent to its parent, and a summary exhibit added quantities that overlap. Both survive a design review, because both are well made. Neither survives the question "what does this prove, and which claim above it is that?"

The checklist to apply. For every exhibit, write the parent claim above it and the question that claim provokes. Confirm the exhibit answers that question and not a neighbouring one. Where an exhibit shows a total, confirm the components do not overlap. Where the message is a decomposition, use a bridge whose bars sum to the stated change and show that they do. Use colour for one distinction only.

Worked example 5: The support exhibit under the first key-line point (India, ₹)

Nirala's key-line point 1 claims that the unit economics already work: a store crosses breakeven at 549.658181 orders a day, reaches that at month 7.492675, and 111 of 186 stores are past it. Two exhibits are proposed. The first is a pie chart of FY26 revenue by city. The second is a line chart of the maturation curve with a horizontal reference line at breakeven and the three cohorts marked. Choose, and compute what each shows.

Solution. Step 1, compute the city data, since it is the exhibit being rejected and the rejection has to be earned. Orders a day by city are Delhi NCR 46,700, Mumbai 37,250, Bengaluru 27,480, Hyderabad 16,720, Pune 13,090 and Chennai 10,350, summing to 151,590, which ties to the cohort table. Annual revenue is 151,590 × 365 × 487 = ₹2,694.5880 crore. The shares are 30.80678%, 24.57286%, 18.12784%, 11.02975%, 8.63513% and 6.82763%, summing to 100.000000%.

Step 2, price the pie. Six slices at those shares are 110.9044, 88.4623, 65.2602, 39.7071, 31.0865 and 24.5795 degrees. The two smallest differ by 6.5070 degrees, and no reader can judge that by eye. For that reason the chart rule rules out a pie above three parts. Worse, the chart is irrelevant to its parent. City mix says nothing about whether a store's economics work, so this is the misparenting defect from the previous example, coming in through a different door.

Step 3, specify the right exhibit. A line chart with months since opening on the horizontal axis and orders a day on the vertical, plotting the nine fitted points from 310 at month 3 to 1,340 at month 42. A horizontal reference line at 549.658181 orders a day, labelled with what it is. Three markers on the curve at the cohort ages: FY24 at (30, 1,240), FY25 at (18, 890), FY26 at (6, 480). The curve crosses the reference line at month 7.492675, which is the single number the exhibit exists to deliver.

Step 4, write the action title. A dark store covers its fixed cost at 550 orders a day, which it reaches at month 7.5. That's a sentence, it carries the number, and a reader with different data could contradict it. You'll get the finer points of wording with slide titles. Here the title only has to state the parent claim, and a title that just names the axis doesn't do that.

Step 5, check the exhibit against its parent. The parent claims the economics work and that 111 of 186 stores are past breakeven. The curve proves the crossing point. The three markers prove where the cohorts sit relative to it. The count of 111 comes from the cohort table, so either the marker labels carry the store counts or a second small exhibit does. Go with the second option, because it keeps the chart to one message.

Read it. Nobody picked the pie because they thought it proved the point. It got picked because it existed, it was colourful, and the slide looked thin. Most misparented support starts out that way. Your defence is to write the parent claim at the top of the slide before you choose anything to put underneath.

The checklist to apply. State the parent claim above the exhibit before choosing a form. Match the form to the message: change over an ordered variable takes a line. Rule out a pie above three parts, and compute the angle difference between the two smallest slices if you need to convince someone. Put the reference line the claim depends on onto the chart and label it. Keep one message per exhibit and split the second message into a second exhibit.

Worked example 6: Grouped against chained, priced (US, $)

Harborline's five corrected levers can be presented two ways. As a group: five independent levers that sum to $32.122765m. As a chain: freight contracts expire in March, so retendering must happen now. Retendering releases cash that funds the store exits. The store exits release the management time for the shrink programme. Therefore the plan must run in that order. The chief executive prefers the chain because it feels like a plan, where the group feels like a list. Price the choice.

Solution. Step 1, assign rejection probabilities. Take these from the team's own read of the board. Realised price gets 0.30, because it's the most aggressive and touches the brand. Store footprint gets 0.15. Freight, with three bids in hand, gets 0.10. Corporate overhead gets 0.20. Shrink and markdown gets 0.35, being the newest and least worked.

Step 2, compute the grouped expectation. Each lever survives on its own, so the expected surviving value is 12.253125 × 0.70 + 6.788640 × 0.85 + 3.088 × 0.90 + 5.4 × 0.80 + 4.593 × 0.65, which is 8.577188 + 5.770344 + 2.779200 + 4.320000 + 2.985450 = $24.432181m.

Step 3, compute the chained expectation. A chain survives only if every link holds. 0.70 × 0.85 × 0.90 × 0.80 × 0.65 = 0.278460. The expected surviving value is 32.122765 × 0.278460 = $8.944905m.

Step 4, take the ratio. 24.432181 / 8.944905 = 2.731408. The grouped presentation is worth 2.73 times the chained one against the same board, the same evidence and the same five actions.

Step 5, look at the worst case rather than the average. Under grouping, the largest single loss is the price lever at $12.253125m, which is 38.144677% of the plan, so a board that rejects it still decides the other 61.855323%. Under chaining, a board that rejects any single link has rejected the plan, and the meeting ends with nothing approved and no record of what was agreed.

Step 6, keep what the chain was right about. The dependencies are real. Freight contracts do expire in March, and management time is finite. Those belong in the implementation sequence, which is a support-level exhibit under the overhead point or a one-line schedule at the end. They don't belong in the key line, whose job is to make the recommendation survive contact with a room.

Read it. The 2.73 assumes the five rejection probabilities are independent. That's hard on the chain and generous to the group, since board scepticism is correlated. Correlation raises the chain's survival probability and leaves the grouped expectation where it was, because that expectation is a sum of the five separate terms. So the advantage narrows. Treat 2.73 as an upper bound on the benefit, and don't read it as an estimate. Nobody doubts the direction, only the magnitude.

The checklist to apply. Write the plural noun over your key line and see whether it fits. If it does, the group is inductive and each point should stand alone. If it doesn't, check whether you have a chain, and if you have, count the links. Assign a rejection probability to each point from your own read of the room. Compute both expectations. Move real dependencies into the implementation sequence and keep them out of the argument.

Worked example 7: A nine-slide storyline that fails pass two (India, ₹)

Nirala's draft board pack has nine slides. Read alone, in order, the titles are "Executive summary", "Business overview", "Store economics", "Cohort analysis", "Central costs", "FY27 capital plan", "Cash and runway", "Risks" and "Next steps". Run the second pass, diagnose the break, and rebuild.

Solution. Step 1, run the pass. Nine label titles run 19 words in total, an average of 2.111 words each. Read in order they make no argument, because a sequence of nouns can't make one. The deck fails pass two on the first criterion, before any question of gaps or ordering comes up.

Step 2, name the break. This is break one, a key line of topics. The repair is mechanical: for each topic, find the one claim the analysis supports, and write it with its number.

Step 3, convert. The nine become these.

• Pausing new openings for two quarters ends FY27 with ₹88.3 crore rather than minus ₹20.4 crore.
• A dark store covers its fixed cost at 550 orders a day and reaches that at month 7.5.
• 111 of our 186 stores are already past breakeven and the other 75 average six months old.
• The FY26 cohort is 70 orders a day short, which is 1.5 months of ageing rather than a closure decision.
• Closing the 75 youngest stores falls behind a pause within 3.5 months and is ₹3.36 crore a month behind by month 12.
• Central cost of ₹14.20 crore a month is sized for 300 stores and consumes 165% of store-level profit.
• The approved plan spends 89.8% of our cash on fit-out before any of it earns.
• A two-quarter pause reaches company breakeven in month 5.9 without new capital.
• Finance reports the revised plan and a central-cost target to the board on 15 October.

Step 4, check the new arithmetic in those claims. Store-level EBITDA is ₹8.6198 crore a month and central cost is ₹14.20 crore, so central consumes 14.20 / 8.6198 = 164.7370% of store-level profit, which rounds to the 165% in the claim. The close-against-pause crossover is at t = 3.511265 months and the month-12 advantage is ₹3.358154 crore a month. Both rest on the stated assumption that closing 75 stores takes ₹1.10 crore a month out of the central base. That assumption carries much of the result, so it belongs on the slide. Hold central cost at ₹14.20 crore and closure never gets ahead for long. Then the crossover moves forward to month 1.492675, and the month-12 advantage rises to ₹4.458154 crore. The FY26 shortfall is 69.658181 orders a day and 1.492675 months.

Step 5, run pass two again on the converted list. The topics are gone, and now the failure is break two, a gap. Read in order, the nine claims still don't force the conclusion. Claims two, three and four all support the same point. Claims six and seven both point at cash without connecting. Collapse them to a key line of three and demote the rest to support. Those three say that the economics work, that the loss is cohort mix and central cost, and that the approved plan runs out of cash where a pause does not. Nine slides become one summary, three key-line slides, four support slides and a closing action line.

Step 6, price what the conversion bought. The nine claim sentences run 153 words, an average of 17.000, and read in 41.7273 seconds. A fourteen-slide deck body at about 95 words a slide is 1,330 words and 6.045455 minutes of reading, so the claim stack is 11.503759% of the reading cost and carries the whole argument. A director who reads only the titles gets the case in under a minute.

Read it. Two of the nine original titles, "Business overview" and "Risks", produced no claim at all when you tried to convert them. That's the useful part of the exercise. A topic that yields no claim is either an appendix exhibit or work that shouldn't have been in the pack, and finding two of them in nine is a normal hit rate for a first draft.

The checklist to apply. Hide every slide body and read the titles in order. If they're nouns, convert each to its one claim with its number before doing anything else. Re-read the converted list and merge claims that support the same point. Check that the last claim arrives at the conclusion. Note every topic that yielded no claim, and move or delete it.

Verify

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

Nirala unit economics: contribution per order 487 × 0.214 − 46 = 58.218000. Per order-day per month 58.218 × 30 = 1,746.540000. Breakeven 960000 / 1746.54 = 549.658181 orders a day. Breakeven age 7.492675 months on the stated curve.

Nirala cohorts: orders a day 1,240 / 890 / 480. Store EBITDA per store 1,205,709.60 / 594,420.60 / −121,660.80. Cohort totals 57,874,060.80 / 37,448,497.80 / −9,124,560.00. Store-level total 86,197,998.60 = ₹8.619800 crore. Central ₹14.200000 crore. Company −₹5.580200 crore a month and −₹66.962402 crore a year. Runway 17.203684 months. Stores at or above breakeven 111 of 186 = 59.677419%. FY26 shortfall 69.658181 orders a day and 1.492675 months. Fit-out 75 × 1.15 = ₹86.250000 crore = 89.843750% of ₹96 crore.

Nirala scenarios: pause breakeven t = 5.890564 months. Pause month-12 company EBITDA +₹4.479870 crore. Pause cumulative twelve-month EBITDA +₹3.442372 crore, maintenance ₹11.160000 crore, cash ₹88.282372 crore. Approved plan, with 75t/12 stores trading in month t at an average age of t/2, a store below month 3 held at 310 orders a day and central ramping in a straight line to ₹15.200000 crore. Monthly EBITDA −4.849273 / −4.118345 / −3.387418 / −2.918471 / −2.449525 / −1.980579 / −1.377221 / −0.712007 / +0.015063 / +0.803990 / +1.654774 / +2.567414 crore. Cumulative −₹16.751598 crore, fit-out ₹86.250000 crore, maintenance 223.5 × ₹0.06 crore = ₹13.410000 crore, cash −₹20.411598 crore. Difference against the pause ₹108.693970 crore. Covenant walk at ₹8.305000 crore of capital a month gives cash ₹26.466389 crore at month 6, ₹16.784168 crore at month 7 and −₹0.522776 crore at month 9. Close-against-pause crossover t = 3.511265 months and month-12 advantage ₹3.358154 crore, with closure assumed to remove ₹1.100000 crore a month of central cost. At an unchanged central base the crossover is t = 1.492675 months and the advantage ₹4.458154 crore. Central against store-level profit 14.20 / 8.6198 = 164.7370%.

Nirala cities: orders a day 46,700 / 37,250 / 27,480 / 16,720 / 13,090 / 10,350, summing to 151,590, which ties to the cohort table. Revenue 151,590 × 365 × 487 = ₹2,694.5880 crore. Shares 30.80678 / 24.57286 / 18.12784 / 11.02975 / 8.63513 / 6.82763, summing to 100.000000%. Pie angles 110.9044 / 88.4623 / 65.2602 / 39.7071 / 31.0865 / 24.5795 degrees. Smallest pair differ by 6.5070 degrees.

Harborline: gross profit 612.4 × 0.382 = 233.936800. EBITDA 233.9368 − 198.6 = 35.336800, margin 5.770216%. Target 612.4 × 0.114 = 69.813600. Gap 34.476800 and 5.629784 points. List-equivalent 284 / 0.9098 = 312.156518. After 3% 321.521213. Realised before volume 303.130200. Effective price +6.735986%. Volume −5.388789%. Revenue 286.795154. COGS 175.512000 → 166.054029. Gross profit 108.488000 → 120.741125. True lever 12.253125. Naive 8.520000 + 9.372000 = 17.892000. Overstatement 5.638875 = 46.019895% of the true figure and 31.516181% of the naive one.

Harborline levers: stores 3.9 + 47.2 × 0.18 × 0.34 = 3.9 + 2.888640 = 6.788640, payback 11.4 / 6.78864 = 1.679276 years. Freight 38.6 × 0.08 = 3.088000. Overhead 5.400000. Shrink gap 0.9 × 612.4 = 5.511600, markdown gap 2.7 × 612.4 = 16.534800, capturable 5.5116/3 + 16.5348/6 = 1.837200 + 2.755800 = 4.593000. Naive five-line total 33.168640 = 5.416172 points and 96.205680% of the gap. Corrected four 27.529765 = 4.495389 points and 79.850116%. Corrected five 32.122765 = 5.245389 points, 93.172118%, margin 11.015605%. Residual 2.354035 = 0.384395 points. Bars plus residual 34.476800 = 5.629784 points, equal to the gap. Points are taken on a revenue base held at 612.400000. The closure lever removes 47.2 × 0.82 = 38.704000 of revenue net of the transfer and the price lever adds 286.795154 − 284 = 2.795154, so a margin quoted after the denominator moves is above 11.015605%.

Harborline ordering: grouped expectation 8.577188 + 5.770344 + 2.779200 + 4.320000 + 2.985450 = 24.432181. All-survive probability 0.278460. Chained expectation 32.122765 × 0.278460 = 8.944905. Ratio 2.731408. Largest single lever 12.253125 = 38.144677% of the plan.

Reading and words: slot 40 / 22 = 1.818182 minutes a slide. Recommendation at slide 19 is minute 34.545455 with 5.454545 minutes left, against slide 2 at minute 3.636364 with 36.363636 left, a ratio of 6.666667. An 18-minute slot reaches slide 9.900000. With the slot cut a third of the time, delivery probability 0.666667 against 1.00, an uplift of 50.000000%. Summary 148 words and 40.3636 seconds. Topic version 37 words and 10.0909 seconds. Six director questions answered 1 (16.6667%) against 6 (100.0000%). Nirala SCQA 101 words and 27.5455 seconds. SCR 54 words and 14.7273 seconds. Saving 47 words, 46.534653%, 12.8182 seconds, 1.922727 person-minutes across nine readers. Harborline note SCQA 132 words and 36.0000 seconds. SCR 79 words and 21.5455 seconds. Saving 53 words, 40.151515%, 14.4545 seconds, 2.168182 person-minutes across nine readers. Label titles 19 words, mean 2.1111. Claim titles 153 words as printed in worked example 7, mean 17.0000, read in 41.7273 seconds, 11.503759% of a 1,330-word body.

Appendix and scoring: 11 numeric claims with 8 backed leaves 3 unbacked, 27.272727%. Appendix 23 slides with 4 opened is a 17.391304% hit rate, and one exhibit per backed claim leaves 23 − 8 = 15 slides carrying nothing promised. In the practice set, 5 of 14 claims unbacked is 35.714286%, 22 of 31 slides support no claim, and the meeting hit rate is 16.129032%.

Practice set

Twelve problems, alternating markets. Work each one before you read the solution. You'll need a calculator, and numeric answers within 2% of the stated figure count as correct unless a problem says otherwise. Where a problem asks for a sentence, the test is whether a reader could disagree with it.

P1 (guided). Governing thought on repriced economics (India). Nirala has repriced its basket. Average order value is now ₹512 at a 20.8% merchandise margin, with the same ₹46 of fulfilment cost per order and the same ₹9,60,000 a month of fixed cost per store. Compute the new breakeven and the month it is reached, then write a governing thought about the FY26 cohort.

Solution. Contribution per order is 512 × 0.208 − 46 = ₹60.496, so a store earns 60.496 × 30 = ₹1,814.88 a month for every order a day it does. Breakeven is 960000 / 1814.88 = 528.960592 orders a day, reached at month 7.049156 on the same curve. Against the old figures that's 20.697589 fewer orders a day and 0.443520 months earlier. A governing thought: "the repricing moves store breakeven from 550 to 529 orders a day, which brings the FY26 cohort forward by about thirteen days and does not change the recommendation to pause openings." Notice the last clause. A number that moves without moving the decision belongs in the deck as a support fact. Say so out loud, and nobody will promote it to a governing thought.

P2 (guided). Find the overlap (US). A four-point key line proposes: raise list prices 2% on SKUs carrying $305m of realised revenue at a 38.2% gross margin, worth $6.10m; cut promoted volume from 36% to 22% at an average 25% discount, worth $10.675m; renegotiate freight; consolidate two distribution centres. Test the pairs, net any overlap at an assumed elasticity of minus 0.7, and state the error in the draft.

Solution. Six pairs. Only the first pair moves the same quantity, which is realised price per unit on the same SKUs. List-equivalent revenue is 305 / (0.64 + 0.36 × 0.75) = 305 / 0.91 = $335.164835m. After a 2% list rise, $341.868132m. With promoted volume at 22%, realised revenue before any volume response is 341.868132 × (0.78 + 0.22 × 0.75) = $323.065385m, an effective price increase of 5.923077%. Units fall 4.146154%. Revenue comes in at $309.670597m, and COGS falls from $188.490000m to $180.674915m. The true combined lever is $12.485682m against a naive sum of $16.775m, so the draft overstates by $4.289318m, or 34.353893% of the true figure. Merge the two into one key-line point.

P3 (guided). Find the complication for a different reader (India). Nirala's working capital lender requires cash of at least ₹25 crore at all times. Under the approved plan, fit-out runs at 75/12 = 6.25 stores a month at ₹1.15 crore each, and maintenance capital at ₹1.1175 crore a month. Monthly EBITDA runs −4.85, −4.12, −3.39, −2.92, −2.45, −1.98, −1.38, −0.71, +0.02, +0.80, +1.65 and +2.57 crore across months 1 to 12. Find the complication and write the SCR opening for the lender.

Solution. Track cash month by month from ₹96 crore, taking out ₹8.305 crore of capital each month, which is ₹7.1875 crore of fit-out plus ₹1.1175 crore of maintenance. On the unrounded EBITDA path it reaches ₹26.46639 crore at month 6, ₹16.78417 crore at month 7 and turns negative at month 9, at −₹0.52278 crore. The covenant floor of ₹25 crore is breached in month 7. That's five months before the plan's own cash trough and two months before the company runs out of money.

For the lender, the SCR opening runs like this. "The facility requires ₹25 crore of cash at all times, and we hold ₹96 crore today. The FY27 opening plan the board approved in March takes cash below that floor in month 7 and below zero in month 9. We are pausing all new openings from this month, which holds cash above ₹88 crore through FY27."

Notice how the complication changed with the reader. For the board, the complication was minus ₹20.4 crore at year end. The lender's is a covenant breach five months earlier. A deck that gave the lender the board's version would be answering the wrong question with the same data.

P4. When the levers over-cover (US). Harborline's board sets the target at 9.5% instead of 11.4%. The five corrected levers are worth $32.122765m. Compute the new gap and coverage, then say what should change in the storyline.

Solution. Target EBITDA is 612.4 × 0.095 = $58.178m, so the gap is 58.178 − 35.3368 = $22.8412m, which is 3.729784 points. Those five levers cover 140.635189% of it. Two things change. Your governing thought no longer needs all five levers, so drop the least certain one, and the remaining four are worth $27.529765m and still cover 120.526789%. And the residual disappears. Now the storyline carries a different caveat, which is execution risk on four simultaneous programmes, where before it was a funding gap. A key line kept at five points when four clear the target is padding, and a director will ask why the shrink programme is in a plan that doesn't need it.

P5. Which exhibit proves the parent (India). The key-line claim is: 111 of our 186 stores are already past breakeven and the other 75 average six months old. Four exhibits are offered. (a) A line chart of total company orders a day over 24 months. (b) A bar chart of store count by city. (c) A scatter of store age against orders a day, one point per store, with a horizontal line at 549.7 and the count above the line labelled. (d) A table of monthly EBITDA by cohort. Choose and justify.

Solution. (c). The claim has two parts, a count above a threshold and an age for the rest, and the scatter carries both: every store is a point, the threshold is drawn, and the 111 count is read off directly. (a) proves the company is growing, which is a different claim. (b) proves nothing about economics. (d) is the closest competitor, and it's really useful under the second key-line point about cohort mix. But it reports three cohort averages and no distribution, so it can't support a claim about how many individual stores are above a line. Re-parent (d). Don't delete it.

P6. Reprice the ordering choice (US). Take Harborline's five levers, worth 12.253125, 6.788640, 3.088000, 5.400000 and 4.593000 in that order, with rejection probabilities of 0.40, 0.20, 0.10, 0.25 and 0.30. Compute the grouped and chained expectations and the ratio.

Solution. Grouped: 12.253125 × 0.60 + 6.788640 × 0.80 + 3.088 × 0.90 + 5.4 × 0.75 + 4.593 × 0.70 = $22.827087m. Chained: all five survive with probability 0.60 × 0.80 × 0.90 × 0.75 × 0.70 = 0.226800, so the expectation is 32.122765 × 0.2268 = $7.285443m. The ratio is 3.133246. Raise the probability that any single point is rejected and the gap widens. That's the counter-intuitive part, and the one that matters in practice. The more hostile the room, the more the grouped structure is worth.

P7. Score a summary against its question list (India). Write the six questions a Nirala director must be able to answer from the summary alone, then score this version: "Store economics. Cohort analysis of 186 stores across three vintages. Central cost benchmarking against four listed peers. FY27 capital plan and store opening schedule. Cash position and runway. Recommendation and next steps for board discussion and approval."

Solution. You need these six:

• What are we deciding.
• What changes if we say yes.
• How much cash does it save and by when.
• What is the strongest argument against.
• What would have to be true for this to be wrong.
• Who owns the next step and when do they report.

The topic version answers one, the first, and only by implication, so it scores 1/6 = 16.6667%. It runs 37 words and reads in 10.0909 seconds, against 148 words and 40.3636 seconds for the claim version, which answers all six. Thirty seconds is the entire price of the difference, and it's the cheapest thirty seconds in the pack.

P8. Audit an appendix against its contract (US). A Harborline summary makes 14 numeric claims. The appendix has 31 slides, of which 9 support a summary claim and 5 were opened in the meeting. Compute the contract score and the two conclusions.

Solution. Five claims of the fourteen have no supporting exhibit, which is 35.714286% of the claims exposed. Twenty-two of the thirty-one appendix slides support no summary claim, and the meeting hit rate was 5/31 = 16.129032%. Those two conclusions point in opposite directions, and both are correct. The appendix is too large, because 22 slides carry nothing the summary promised. It's also incomplete, because 5 claims have nothing behind them. Rebuilding to the contract gives 14 exhibits, indexed by claim. As for the hit rate, it won't improve. It doesn't matter either, since an appendix is insurance. You don't judge insurance by how often you claim on it.

P9. Convert a title stack (India). These nine label titles run 19 words in total: "Executive summary", "Business overview", "Store economics", "Cohort analysis", "Central costs", "FY27 capital plan", "Cash and runway", "Risks", "Next steps". Convert each to a claim with its number where the analysis supports one, and report what happens.

Solution. Seven convert. Store economics becomes the 550-orders-a-day breakeven at month 7.5. Cohort analysis becomes 111 of 186 past breakeven with the other 75 at six months. Central costs becomes ₹14.20 crore a month consuming 14.20 / 8.6198 = 164.7370% of store-level profit. FY27 capital plan becomes 89.8% of cash spent on fit-out before any of it earns. Cash and runway becomes minus ₹20.4 crore at year end against ₹88.3 crore under a pause. Executive summary becomes the governing thought. Next steps becomes the owner and the date. Two don't convert. Business overview and Risks yield no claim the analysis supports, so the first is deleted and the second is folded into the recommendation as its two conditions. The converted set runs 153 words, averages 17.000 a title, and reads in 41.7273 seconds.

P10. Spot the mixed ordering principle (US). A five-point key line reads "North America underperformed", "Europe grew 14%", "the second half was weaker than the first", "the direct channel outgrew retail", "Asia is now 19% of revenue". Name the principle each point follows, state the defect, and re-sort.

Solution. Points one, two and five follow a structural principle, by geography. Point three follows time. Point four follows a different structural principle, by channel. That's three principles in five points, so a reader can't predict what comes next and stops trying. Re-sort under one principle and demote the others to support. Keep geography on the key line, sorted by size of the revenue effect, and move the half-on-half comparison and the channel split beneath the geographies they explain. If the channel story really is the argument, do the reverse: channel on the key line, geography beneath. You can't have both.

P11. Price a compression (India). An SCQA opening runs 118 words. The SCR version runs 61. Reading is 220 words a minute and the note goes to 14 people. Compute the saving three ways and say when you would keep the question.

Solution. The cut is 57 words, which is 48.305085% of the text and 57/220 × 60 = 15.5455 seconds a reader, or 3.627273 person-minutes across fourteen readers. Keep the question when the complication could provoke two different questions and you're choosing between them. Keep it, too, when the audience is mixed enough that one half would otherwise spend the meeting waiting for a question you were never going to answer. Drop it when the reader has been living with the problem, because then the question is already in their head and writing it down reads as delay.

P12. Diagnose the break (US). Harborline's draft governing thought reads: "the five-point plan restores the 11.4% EBITDA margin by FY28." The corrected levers total $32.122765m against a gap of $34.4768m. Name the break, name the repair, and write the corrected sentence.

Solution. The key line is made of claims, so it isn't break one. Each support item proves its parent, so break three is out too. That leaves break two in its leap form: the conclusion asserts more than the points support. Those levers deliver 5.245389 points against a 5.629784-point gap, so the plan reaches 11.015605% and leaves $2.354035m, or 0.384395 points, unaccounted for. So weaken the conclusion until it fits, and name the residual as work. Corrected, it reads: "five levers take EBITDA margin from 5.77% to 11.02% by FY28, which is 93.2% of the way back to 11.4%; the remaining 0.38 points depends on a decision about the B-tier stores that this plan does not make." It's longer and less satisfying, and it's the only version that survives the finance director.

Applied mini-project

Build a complete pyramid for a supplied brief, write its SCQA opening, and submit the structure to the storyline drill, which checks parent, child and order mechanically. Budget about two hours. Pick one of the two briefs. Both have the same shape, and they differ in market, currency and the size of the problem.

Brief A: Sunhaven Dairy (India, synthetic). Sunhaven sells chilled dairy through 4,200 retail outlets in three states. FY26 revenue is ₹418 crore. The outlet base splits into three tiers: the top 900 outlets carry 52% of revenue at a 31.2% gross margin, the middle 1,500 carry 33% at 25.1%, and the tail of 1,800 carries 15% at 17.8%. Cold-chain and servicing cost is ₹1.53 lakh per outlet per year regardless of tier. Overhead is ₹41.8 crore. The managing director wants a recommendation on whether to cut the tail before the FY28 route plan is signed in six weeks.

Brief B: Kestrel Components (US, synthetic). Kestrel distributes industrial fasteners to 3,100 accounts. FY26 revenue is $284.6m. The top 210 accounts carry 46% of revenue at a 22.4% gross margin, the middle 640 carry 34% at 18.9%, and the long tail of 2,250 carries 20% at 12.1%. Service cost is $14,800 per account per year regardless of size. Overhead is $19.4m. The board wants a recommendation before the annual pricing letter goes out in five weeks.

Neither brief tells you the answer, and both contain the arithmetic that decides it. Run the numbers before writing a sentence.

Step 1: compute the tier economics (about 25 minutes). Build the contribution by tier: revenue times gross margin, less servicing cost times outlet or account count. Check that your tier revenues sum to the stated total. Then check that your blended gross margin is the one the tiers imply, and not one you assumed. For Sunhaven the blended margin works out at 27.177000% and EBITDA at ₹7.539860 crore, which is 1.803794% of revenue, with the tail contributing −₹16.379400 crore. For Kestrel the blended margin is 19.150000% and EBITDA is −$10.779100m, which is −3.787456% of revenue, with the tail contributing −$26.412680m. If your figures differ, find the error before you go on, because every sentence you write afterwards inherits it.

Step 2: write the governing thought (about 15 minutes). One sentence, with its number, its action and its date. Run the negation test on it and write down the opposite. Then write the strongest sentence a sceptic could say against it. For both briefs that's some version of the argument that the tail feeds the middle tier and cutting it costs more than the arithmetic shows. If you can't answer that sentence with a number you hold, say so in the storyline. Don't hide it.

Step 3: build the key line (about 25 minutes). Three to five claims, each with its number, all of the same kind, under one declared ordering principle which you write down. Then run the pair test: list every pair, name the quantity each pair moves, and net any pair that moves the same one. Then declare your frame in one sentence and check the frame for anything with no key-line home.

Step 4: attach the support (about 20 minutes). Under each key-line point, name the one or two exhibits that prove that claim, and write the question the parent provokes on the same line. Specify each exhibit as a form plus a message plus a title claim: what's on each axis, what's sorted how, what reference line the claim needs. A picture won't do. You aren't building the exhibits here. You're contracting for them.

Step 5: write the SCQA opening and its SCR compression (about 20 minutes). Situation the reader has already signed, complication from one of the three sources, question in the reader's words, answer equal to your governing thought. Then cut the question and compress, and count both versions in words and reader-seconds.

Step 6: write the summary and run the read-alone test (about 15 minutes). The top two levels, about 150 words, every claim carrying its number, no forward references, and a final sentence with an owner and a date. Write the six questions your reader must be able to answer from it, then hand the summary to someone else and score it. Anything below five out of six goes back to step 3, because a summary that fails is usually a key line that failed first.

Deliverables. A one-page pyramid with each box's question written beside it, from the governing thought down through the key line to the support. Then the SCQA and SCR openings with their word counts, the executive summary, the exhibit contract naming one exhibit per numeric claim, and the computation file that produced every figure.

Rubric. Twenty points, passing at 16 with no zero on any row.

Any row whose evidence is absent scores zero, however well the page reads. A pyramid that reaches 16 with a wrong arithmetic tie isn't a pass, because row 17 gates the rest. A storyline built on figures that don't reconcile is a well-organised error.

Reading & resources

Core (do these).

• Barbara Minto, The Pyramid Principle: the source. Read the first part, on the pyramid, the introduction structures and the ordering of grouped ideas, before anything else in this list. The later material on deduction and induction repays a second reading once you've built two or three pyramids of your own. [Paid] [Intermediate]
• Gene Zelazny, Say It With Charts: written for the same consulting audience, and the tightest statement of the idea that the message chooses the exhibit, which is what the vertical test enforces. [Paid] [Beginner]
• Cole Nussbaumer Knaflic, Storytelling with Data: read it chapter by chapter alongside the chart-craft branch. Read its material on the big idea and the storyboard here, though, because it's the most usable published bridge between a pyramid and a deck. [Paid] [Beginner]
• Jean-luc Doumont, Trees, Maps, and Theorems: a compressed treatment of professional communication that arrives at a very similar structure from an engineering direction, and is short enough to read twice. Its treatment of the message-first opening is the best short version in print. [Paid] [Advanced]

Going deeper.

• Nancy Duarte, Resonate: the audience-centred case for structure, coming at it from design instead of from analysis. Useful as a corrective if the pyramid starts to feel like a filing system. [Paid] [Beginner]
• Chip Heath and Dan Heath, Made to Stick: on why some claims survive retelling and others don't. That matters at the top of a pyramid, where one sentence has to travel through three people without changing. [Paid] [Beginner]
• Alberto Cairo, How Charts Lie: for the support level, and specifically for the habit of asking what a chart proves rather than what it shows. [Paid] [Beginner]
• Richard Rumelt, Good Strategy Bad Strategy: the chapter on the kernel (diagnosis, guiding policy, coherent action) is a different three-level structure that maps closely onto situation, governing thought and key line, and reading the two against each other sharpens both. [Paid] [Intermediate]
• Stephen Toulmin, The Uses of Argument: the claim, grounds, warrant frame is older than the pyramid and explains why the vertical test works. Heavy going, worth it if you like knowing why. [Paid] [Advanced]

Free and practical.

• The US Securities and Exchange Commission, A Plain English Handbook: free from sec.gov, written for disclosure documents, and the cheapest available improvement to the sentences your claims are written in.
• Any listed company's investor-day deck, downloaded from its own investor relations page: read the titles alone, in order, and write down the argument they make. Do this on three decks from the same sector and you'll have calibrated the read-only-the-titles test on real artefacts in under an hour. Investor-day decks are useful here because the argument is usually deliberate. Quarterly result packs usually aren't.
• Your own last deck: cover the bodies, read the titles, and count how many are nouns. That number is your starting point, and it is the only baseline that will matter to you.

Do this, not just read. Take one page of any analyst report you can find and reverse-engineer its pyramid. Write the governing thought you think it's arguing, the key line you can recover from its section headings, and the support under each. Most reports won't yield a clean pyramid, and finding out where the structure gives way teaches you more than reading a good one.

Teach it back & journal

Feynman prompt. Explain to a colleague who has never heard the word pyramid why two decks containing the same analysis can be worth different amounts, without using the words structure, framework or narrative. Use the arithmetic. A recommendation on slide 19 of 22 arrives at minute 34.5 of a forty-minute slot and leaves 5.45 minutes of discussion. On slide 2 it arrives at minute 3.6 and leaves 36.4. If the slot is cut to eighteen minutes a third of the time, the first deck delivers its recommendation in 67 meetings out of 100 and the second in all 100. Then make the harder point. Neither number is about slide order. Both are about whether somebody decided, before building, what single sentence the deck exists to defend. Finish with the pair test. Show your colleague that a list-price rise worth $8.52m and a promotion cut worth $9.372m aren't worth $17.892m, and that finding the $5.639m of double count took ten pairs and twenty minutes. If they can tell you, unprompted, why a plural noun over a key line is a useful test, you've taught the idea. The arithmetic was only the way in.

Journal prompt. Take the last deck you built or received and run the second pass on it: hide every body, read the titles in order, and write down the argument they make. Count how many of the titles are nouns. Then write out, as best you can reconstruct it, the governing thought that deck was defending, and be straight with yourself about which it is. Most people are writing that sentence for the first time, and that's the finding, so don't count it as a failure. Next, take the key line you've recovered and run the pair test on it, naming for each pair the quantity both points move. Note any pair where you can't answer, because an unanswerable pair is usually an overlap you haven't looked at yet. Then write a paragraph on which of the three breaks that deck had, and what it would have cost to find it before the meeting instead of during it.

Finish with one habit small enough to survive a bad week. Keep a text file, opened before the deck software, containing one sentence and its opposite, and a rule that the software stays shut until you can defend the sentence against a colleague who holds the opposite. That habit is worth more than the vocabulary. It's also the only part of this that transfers to a memo, a model walkthrough, an email and a two-minute answer in a corridor.

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