The Meta-Skill: When Frameworks Mislead

Branch: Strategy Frameworks (Consulting galaxy) · Node CN3.03 · ~8 focused hours · Green Belt (Consultant): mastery gate ≥85%

You have, by this point in CN3, run a proper Five Forces analysis, drawn a BCG growth-share matrix with real numbers on both axes, and built a 7S diagnostic that actually distinguishes structure from systems. You did the mechanics correctly: CN3.01 and CN3.02 made sure of that. And yet you have almost certainly also sat in a room (or built the deck yourself) where every framework was filled in correctly, the slides looked exactly like the ones in the textbook, and when someone finally asked "so what should we actually do differently on Monday morning," nobody could answer, because the deck had never actually found out why the client's specific number was moving. That gap, perfect framework execution paired with zero diagnosis, is not a rare failure. It is the default outcome of treating a framework as the starting point of thinking rather than as one candidate lens applied to it, and it is common enough, and expensive enough, to be a named idea in the strategy literature: Richard Rumelt calls its cousin "bad strategy," and this module is built on his central insistence that real strategy work always starts with diagnosis, never with a template.

Here is the whole module compressed into one sentence you will spend eight hours earning the right to believe: a framework is a hypothesis about what matters here, never a law about what is actually there, and the instant you let its shape substitute for diagnosis, you have stopped doing strategy and started performing it. Five Forces, the BCG matrix, 7S, Ansoff, blue ocean, every one of them was built by someone who noticed a pattern in some situations and generalized it into a reusable lens. Every one of them therefore also encodes a hidden bet about which variable is the hinge of the problem, and every one of them will mislead you, fluently and confidently, the moment you apply it to a situation where that bet doesn't hold, producing a deck that is technically correct and diagnostically empty.

By the end you will be able to do something the two framework sets before this one deliberately did not teach you. Given a fresh business situation, you'll name the framework a less experienced team would reach for by reflex, state in one sentence the assumption that framework requires without saying so, and test whether it actually holds here. When it doesn't, you'll build a bespoke, MECE structure keyed to the real mechanism, stress-test it against a rival explanation, and compress the result into one sentence a smart twelve-year-old could follow. That is the Green Belt: not knowing more frameworks than the next person, but refusing to let any framework think for you.

Learning objectives

By the end you can:

1. State Rumelt's three-part "kernel" of strategy (diagnosis, guiding policy, coherent action) and explain, from it, exactly which part framework-as-theater fakes and why that specific fake is so easy to miss.
2. Recognize the four faces of bad strategy (fluff, failure to face the challenge, mistaking goals for strategy, bad strategic objectives) inside a real deck, memo, or argument, including your own.
3. State, for each major framework taught in CN3.01/CN3.02 (Five Forces, generic strategies, value chain, 3C, BCG growth-share, Ansoff, 4P, 7S, jobs-to-be-done, blue ocean, scenario planning), the one hidden assumption it requires to be true, without re-deriving how to run any of them.
4. Turn a vague mandate ("fix the margin," "we're losing share") into one falsifiable, mechanism-level candidate diagnosis, using why-laddering and the "what would have to be true" test.
5. Build a bespoke, MECE issue tree keyed to a specific diagnosed mechanism, borrowing fragments of named frameworks as lenses rather than adopting any one of them whole.
6. Stress-test a bespoke structure with the rival-hypothesis check, the branch-removal check, and the one-sentence clarity test, and know what it means when a structure fails one.
7. Apply a stated three-part decision rule for when the canonical framework is exactly the right tool, and recognize reflexive framework-refusal as its own failure mode.
8. Refuse framework-as-theater in the room: propose a sharper, falsifiable bespoke alternative without grandstanding, and manage the credibility cost of an unfamiliar-shaped right answer.
9. (Productivity objective: R10 duality.) Use AI to widen your menu of candidate lenses fast and to red-team your own bespoke structure, while catching AI's own reflex to hand back a framework-shaped answer whether or not it fits, which is the identical failure this module trains you to catch in people.

The duality, stated once (R10). Objectives 1–8 are the gated skill: diagnosing before prescribing, by hand, checked against the data, that is what the mastery quiz and the module's mapped challenge reward. Objective 9 is the productivity payoff you keep: knowing that a chat assistant is exactly as prone to framework-as-theater as an eager junior analyst, for the identical structural reason (AI0.01's payoff line, a model optimizes for plausible, not true, and "run a Five Forces" is the overwhelmingly plausible-shaped answer to "analyze this business"). AI can widen your menu of lenses faster than you can alone; it cannot tell you which lens fits your specific mechanism. Only your own diagnosis, checked against your own data, earns the gate.

Prerequisites & connections

Builds on. CN3.01 (Competitive & Industry Frameworks) and CN3.02 (Corporate, Growth & Customer Frameworks) taught you to run Five Forces, generic strategies, the value chain, 3C, the BCG growth-share matrix, Ansoff, 4P, 7S, jobs-to-be-done, blue ocean, and scenario planning correctly. None of those mechanics gets re-taught here, so you'll need to be able to build each one unaided. CN1.01 (Issue Trees & MECE) supplies the structural tools, the branch-removal and overlap/gap tests, which you'll now reuse on structures you build yourself instead of ones a template hands you. CN1.02 (Hypothesis-Driven Problem Solving) supplies the day-one-hypothesis habit and the minimum-analysis instinct that diagnosis-first thinking depends on. CN0.01 (The Consulting Mind) supplies the problem-solving cycle, and what you learn here sharpens one stage of it. If your first taste of this idea was the Finance galaxy's elective E11.03 (The Consulting Case-Interview Lab), you've already met it in a single paragraph. That lab's closing section names the meta-skill ("diagnose before you prescribe... a five-forces recital on a pricing case wastes everyone's time"), and its case CL-11.3 (Kite Foods Profitability) lets you feel the difference once. Here, that one paragraph grows into eight full hours: the kernel formalized, the hinge-assumption catalog for every framework you now know, the diagnostic procedure, the stress tests, and worked examples where every number reconciles.

Feeds forward. CN4.01 (Reading an Industry Like a Strategist) hands you a generic method for industry reading. Bring the hinge-assumption habit to it straight away, because a generic method run as a checklist can turn into theater as easily as any named 2×2. CN5.01/CN5.02 (business-model teardown, market attractiveness) and the CDD toolkit of CN5.03 depend on a bespoke thesis, not a templated one. A commercial due-diligence report that runs the same five slides regardless of the deal is the same disease, at engagement-cost stakes. CN6.01/CN6.02 (M&A) hand you synergy templates and portfolio quadrants that are as prone to hidden-assumption failure as the BCG matrix worked example below, and the same test applies there. CN7.01 names the parallel disease by its operating-model nickname, the "boxes-and-lines trap", an org chart redrawn without diagnosing why the current one is failing. CN7.03, the Partner Capstone, hands you a never-seen company and times you. No template survives contact with a company you have never seen, but the procedure you learn here does. For the answer-first presentation of whatever bespoke diagnosis you build, cross-link CN1.03 (The "So What": Synthesis & the Answer-First Storyline) and the Finance galaxy's M10.03 (The Communication Toolkit (Minto's Pyramid Principle)). Diagnosis and storyline are two different jobs done in sequence, and the work here stops where those two cross-links begin. (The Finance galaxy's M4.03, M4.04–M4.05, and E11.02 cover Porter, the moats canon, and M&A modeling respectively from the finance-analysis seat. The material here shares their canon but never re-derives their mechanics.)

Core teaching content

Why frameworks mislead: hypotheses wearing the costume of laws

Every named framework was built by noticing a pattern that held across many real situations and generalizing it into a reusable lens. So every one of them also encodes an implicit claim, one the whole framework rests on, about which variable is the actual hinge of the problem it's applied to. State the hinge, and you can test it. Leave it implicit, and the framework will mislead you with the same fluent confidence a well-trained model gives a fabricated citation. You already know how to run every framework below. Here's the one sentence each of them is betting on without telling you:

• Five Forces bets that the relevant unit of analysis is a bounded, definable industry, and that its structure is the primary driver of average profitability here, more than any one firm's specific execution. It misfires the moment the question is really about one firm's execution (Worked Example 1), or the industry's boundary is itself dissolving.
• Generic strategies (cost leadership / differentiation / focus) bet that the strategic choice reduces to a position along those two axes. Porter himself warned against forcing every firm into one box. The bet breaks for a firm whose edge is a system of reinforcing activities that resists a single-axis label.
• The value chain bets that cost and value are best understood by decomposing into the chain's generic sequential activities. It breaks when the real source of advantage is a configuration the standard vocabulary has no box for.
• 3C bets that company / customer / competitor exhausts what matters. It breaks when the binding constraint is regulatory, a platform shift, or internal (capital allocation, governance), none of which is a "C."
• The BCG growth-share matrix bets that market growth proxies for cash need and relative share proxies for cost position via the experience curve. It breaks in any business where share doesn't buy a cost advantage, and outside a shrinking set of scale-economics industries, that's most of the modern economy. It breaks completely the moment two people in the room can't agree what "the market" actually is (Worked Example 4).
• Ansoff bets that product and market are the right pair of axes for a growth choice. It breaks when the real lever is a business-model or channel change orthogonal to both.
• 7S bets that the organization's problem is internal (mis)alignment among its seven elements. It breaks when the puzzle was caused outside the firm (a competitor's move, a regulatory shift, a demand shock). Looking inward then burns the diagnostic window while the real clock keeps running (Worked Example 5).
• Jobs-to-be-done bets that the customer's own account of "the job" is the right lens on demand. It breaks when the binding constraint is supply-side (capacity, unit economics, regulation) and no amount of demand-side insight relieves it.
• Blue ocean's value-cost decoupling bets that the industry's current value-cost frontier is the right reference frame to jump away from. It breaks when the trouble is execution within the current frontier, where positioning against the frontier won't help.
• Scenario planning bets that the future's key uncertainty is knowable in advance and roughly discrete. It breaks when the uncertainty that matters is about the firm's own execution capability, which no scenario matrix resolves.

None of this says any of these tools is bad. Every one of them is often the right call, and Worked Example 6 uses one that way, on purpose. The claim is narrower and harder to skip past. "I ran it correctly" and "its hidden bet holds here" are two entirely different judgments, and so far you've only been trained in the first.

Diagnosis first: turning a vague mandate into a falsifiable question

Real mandates arrive vague: "margins are under pressure," "we're losing to a competitor," "growth has stalled." The most expensive habit in strategy work is reaching for a framework's shape before you've turned that vague mandate into something a specific piece of data could actually contradict.

Why-laddering means asking "why" repeatedly, but with a stopping rule: stop the moment you reach a claim you could test with data available this week. You're not after philosophical depth. You want the shallowest claim that, if false, would change what you recommend. Start from "margins are under pressure" and ask why. The reply is "cost per unit rose." Ask why again, and you reach a fork. Is the rise broad-based, pointing to industry input-cost inflation, or concentrated (one supplier, one plant, one product line)? That last fork is already testable, and you haven't needed a single named framework to reach it.

The "what would have to be true" test takes the obvious explanation, the one everyone in the room already half-believes, and forces it into the open as a prediction. If margin pressure really is industry-wide input-cost inflation, the decline should be broad-based across product lines, and it should roughly track the relevant commodity or wage index. State that prediction before you look, then go look. Suppose the decline is concentrated in one line, one region, one vintage instead. Then the obvious explanation has just failed a test nobody gave it before the deck went out, and you have your first real clue about where the mechanism actually lives.

Symptom versus mechanism is the distinction the whole procedure protects. A symptom is what you were told: "NPAs are up," "same-store sales are down," "win rate fell." A mechanism is the causal chain that produced the symptom. A framework's quadrant label ("this business is a Dog," "rivalry is intense") is very often the symptom wearing a costume. It restates that something is bad without saying why. Putting a symptom into a framework's vocabulary can feel like progress, simply because it now looks structured. It doesn't become diagnosis until it names a mechanism specific enough to be wrong.

Building the bespoke structure: from diagnosis to a tree that fits this problem

When you have a candidate diagnosis that's falsifiable and mechanism-level, or more often two or three competing ones, build a MECE tree keyed to that, and leave any template's generic branches out of it. Start from the metric that moved, and decompose only along the dimensions that could plausibly carry the mechanism you hypothesized: geography, channel, cohort, vintage, product, customer segment, time period. Pick them because your diagnosis implies one of them should show a concentrated effect, never because a template says to cut it that way. A cost bridge, a same-store-sales bridge, a roll-rate table by vintage: none of these has a famous three-letter name, and you build each one new, every time, from the specific mechanism you're chasing.

You don't have to work from nothing, either. Borrow fragments, never templates. Pull in a value-chain lens for the narrow question "where does this specific cost actually sit," or a quick 3C sanity check for "did a named competitor move recently," without adopting either framework's full structure as your organizing spine. A bespoke tree is a custom assembly of the sharpest fragments of several lenses. You choose each one because it bears on your specific hypothesis, never because it's "the" framework for problems that look like this.

The one-sentence test catches you if you've slipped back into theater. Compress your candidate diagnosis into one sentence a smart twelve-year-old could follow. Here's one that passes: "NPAs rose because collections quality broke down in one branch cluster after a vendor change". It names a mechanism, a location, and a trigger, and it would be straightforwardly false if the deterioration were actually broad-based. Now try "NPAs rose due to macro headwinds and competitive intensity in a maturing credit cycle". It's fluent, and it sounds like an analyst wrote it. It also fails, because it would say the same thing whether it were true or false. Nothing in it could be checked. If you can't compress your diagnosis this way, you don't have a diagnosis yet. You have a framework's output wearing a sentence's clothes.

{
  "skill": "Widening the menu of candidate lenses before committing to a bespoke structure",
  "use": "Ask AI to list every named framework or lens that could plausibly apply to a stated business situation, each with its one-sentence hidden assumption -- a fast way to check you haven't missed an obvious angle before you diagnose. Never let it pick the structure for you.",
  "tools": ["Chat assistants -- Claude, ChatGPT, Gemini"],
  "prompt": "Here is a business situation: <PASTE 100-150 WORDS, METRIC + CONTEXT, NO CONCLUSION>. List every strategy framework or lens that could plausibly apply here (name each one), and for each state in one sentence the hidden assumption it requires to be true. Do not recommend which one to use, and do not diagnose the situation yourself -- that decision and that diagnosis are mine.",
  "verify": "Test every assumption the model names against your own data before trusting the framework it belongs to. Treat the list as a menu, never a verdict -- a framework whose assumption you haven't personally checked stays unproven, no matter how confidently the model presented it (the AI0.01 payoff line applies exactly: plausible-sounding is not the same claim as true).",
  "diy": "You must already be able to name a framework's hidden assumption yourself (the section above) before you ask AI for the menu -- otherwise you cannot tell a genuinely relevant lens from a plausible-sounding irrelevant one.",
  "market": "IN"
}

Stress-testing the structure before you trust it

A bespoke tree earns trust the same way a framework's output should have had to, and never automatically.

The rival-hypothesis check. For every candidate diagnosis, name at least one plausible rival explanation, and state what data would distinguish the two. If you can't say what would distinguish them, you don't have a diagnosis yet. You have a story that happens to fit the facts you already looked at. That's a much lower bar than a diagnosis that has survived a real competitor.

The branch-removal check comes straight from the MECE work you did on issue trees, now aimed at a structure you built yourself. Delete any one branch of your tree. Does the tree still "explain" the movement in the metric? If it does, that branch was never holding anything up. Cut it, or work out what it should actually have been measuring.

Reconciliation. A bespoke tree's arithmetic has to add up, like every other number you'll meet here. The parts you've identified must sum to the whole you're explaining, to the rupee or the dollar or the percentage point. Approximately isn't good enough. Every worked example below is built this way on purpose. You get to see the reconciliation instead of taking the conclusion on faith, and you build the habit of demanding the same of your own trees before you ship them.

When the canonical framework is exactly right: the discipline cuts both ways

Refusing frameworks on reflex is its own failure mode, and it isn't a subtle one. Rebuilding from scratch a structure that a well-tested named framework already supplies is slower and more error-prone, and it frequently arrives at something close to the framework's own shape anyway. Contrarianism isn't diagnosis either. Before you reach for a bespoke tree, run this three-part decision rule:

1. Is the question a genuine, bounded structural question about a market or industry as a whole, and not about one firm's specific performance inside it?
2. Does the framework's hinge assumption plausibly hold here, checked against what you already know about this specific situation and not merely assumed?
3. Would a bespoke tree, if you built one anyway, end up re-deriving something close to the framework's own structure?

If the answer is yes to all three, use the canonical framework. Run it the way you were taught, and spend your remaining time on the analysis instead of reinventing its shape. Worked Example 6 below is this case: an industry-entry question that really is structural, where Five Forces is the right first move. It isn't there as a contrast for its own sake. The habit that refuses theater is the same one that recognizes when the theater accusation doesn't apply.

Refusing framework-as-theater in the room

Knowing a framework won't fit is only half the skill. The other half is saying so, in the room, to a partner or a client who wants the comfort of a familiar shape. That costs real credibility every time, so it's worth learning on purpose instead of leaving it to instinct.

Done well, the refusal doesn't sound like "frameworks are useless" or "I don't believe in 2×2s". That's grandstanding, it's unhelpful, and it's simply wrong, since you've just spent a whole section learning when they're the right tool. It sounds like a specific, respectful redirection, paired with a falsifiable alternative instead of plain skepticism: "Before we build the Five Forces slide, can I show you what the branch-level data says first? If the deterioration is as concentrated as I think it is, Five Forces will tell us the industry is basically fine while missing the one specific thing that's actually wrong with us, and we'll have spent the week answering a question nobody asked." Skepticism without an alternative is noise, and a partner will, correctly, override it. Skepticism paired with a sharper, cheaper, faster test is the Green Belt.

The social cost is real, so name it plainly. A familiar framework is comfortable because everyone in the room already recognizes its shape and can nod along without following new logic. A bespoke tree asks the room to follow an argument whose outline they haven't seen before, and that spends more of the presenter's credibility per slide. Rumelt's diagnosis-first ethic supplies the professional courage to spend that credibility anyway. A wrong-shaped answer, delivered on time and confidently, costs the client vastly more than an unfamiliar-shaped right one delivered the same week.

The AI mirror: why a chat assistant over-fits to famous frameworks

Recall the payoff line from your introduction to how modern AI works: a large language model optimizes for plausible, not true. Ask a bare chat assistant to "analyze this business's strategy," and the overwhelmingly most plausible continuation is some arrangement of Five Forces, SWOT, and the BCG matrix. That shape saturates the training corpus, from every MBA case write-up to every consulting-101 blog post. The model produces it with the same fluent confidence whether or not the situation's hinge assumptions hold. It's the same failure mode you've been studying, wearing a different face. The parallel is no coincidence. A model has read more framework-shaped case studies than any human analyst ever will. That makes it an excellent tool for widening your menu of candidate lenses fast, and an equally dangerous one for choosing among them. Choosing means checking a hidden assumption against your specific data, and a model has no privileged way to know your data. It only gets there if you give it every bit of the information you'd need to check the assumption yourself.

{
  "skill": "Catching an AI assistant's default framework-shaped answer when the real problem needs a bespoke structure",
  "use": "After AI diagnoses a specific, narrow business puzzle, explicitly test whether the answer names a falsifiable mechanism or is a generic framework recital dressed as one -- and if it's the latter, force a redo grounded only in your own data.",
  "tools": ["Chat assistants -- Claude, ChatGPT, Gemini", "Research/agentic tools -- background context only, never the diagnosis itself"],
  "prompt": "I asked you to diagnose <PUZZLE, WITH YOUR OWN DATA PASTED IN>. Before I accept this: does your answer name a mechanism specific to my situation and data, or could it apply to any company with a similar symptom? If it's the latter, redo it using ONLY the data I gave you, and name the ONE data cut that would prove or disprove your top hypothesis.",
  "verify": "Run the one-sentence test and the rival-hypothesis check on the model's revised answer yourself, exactly as you would on a junior analyst's first draft. An AI's diagnosis is exactly as vulnerable to framework-as-theater as a person's, for the identical structural reason (plausible, not true) -- its confidence is never itself evidence.",
  "diy": "You must already be able to run the one-sentence test and the rival-hypothesis check unaided (the sections above) -- AI can draft candidate mechanisms fast; only your own check tells you whether one of them is real.",
  "market": "US"
}

Common mistakes & how experts think differently

1. Template-itis, reaching for the first framework whose name you remember. A novice reads the prompt, recalls a familiar 2×2 and fills it in. An expert reads the prompt, asks what specific question is actually being asked, and only then checks which lens fits, if any does. Familiarity isn't evidence of fit.
2. Box-filling mistaken for insight. A quadrant "filled in" with language that would describe any company in the industry is decoration. The expert asks whether a rival firm's deck could contain the identical bullet. If it could, the bullet hasn't yet said anything about this client.
3. Ignoring the contradictions the framework itself surfaces. A deck that asserts "buyer power is low" on one slide and "we must cut prices to stay competitive" on the next, without reconciling them, has stopped thinking the moment the slides were finished. The expert treats an internal contradiction as the most valuable sentence in the deck, because it points straight at the diagnosis nobody has done yet.
4. Fluff mistaken for sophistication. Jargon-dense language that would say the same thing whether true or false reads, to an inexperienced ear, as expertise. The expert runs the one-sentence test on their own writing before anyone else does.
5. Mistaking a goal for a strategy. "Grow 20% and become the most trusted brand" states two ambitions and zero obstacles, zero tradeoffs, and zero mechanism. It isn't wrong. It just isn't a strategy yet. On hearing any goal, the expert's reflex is to ask what specifically stands in the way of it.
6. A list of priorities that never says no to anything. Eleven "strategic pillars," unranked, is a permission slip for every department to keep its budget. The expert insists on a ranking and a named sacrifice before calling anything a priority.
7. Failing to face the actual challenge. Naming the real obstacle sometimes assigns blame or admits a sunk cost, so teams reach for a framework's comfortable, blame-free vocabulary instead. The expert asks of any conclusion, "would this sentence survive if I had to say it to the person responsible for the outcome?"
8. Assuming a framework's hidden assumption holds without checking it. Take BCG's experience-curve bet, 7S's internal-locus bet, Five Forces' bounded-industry bet. Each is a testable claim and never a law. The expert states it out loud and checks it before trusting the framework's conclusion.
9. Refusing frameworks reflexively, as its own kind of contrarian theater. Rebuilding from scratch what a well-tested named framework already supplies wastes time and frequently arrives back at something close to the framework's own shape. The expert applies the three-part decision rule before reaching for a blank page, and is perfectly willing to say "Five Forces is exactly right here" when it is.
10. Building a bespoke tree that "explains" the data without checking whether a rival story explains it equally well. A structure that merely fits the facts you already looked at clears a low bar. The expert names the rival hypothesis, and the one data cut that would kill it, before presenting the tree as a finding.

Worked examples

Worked example 1: Refusing Five Forces theater at Amrit Rural Credit (India, ₹)

Amrit Rural Credit, a synthetic regional NBFC, sees its gross NPA ratio rise from 3.2% to 5.6% of a ₹1,000 crore loan book over one year. The CFO calls the jump "concerning" in the board memo, and says nothing more specific than that. A junior team's first instinct is to open with an industry-level Five Forces: rising rate cycle, competitive intensity from new digital lenders, regulatory tightening on NBFC capital norms. Every one of those forces is real and correctly described. None of them explains why this book, at this company, deteriorated by 2.4 points in one year while peers reported roughly flat NPAs over the same period.

Solution. Step 1, check the hinge assumption. Five Forces bets that industry structure is the primary driver of the metric you're explaining. So test it. If a sector-wide rate cycle or regulatory shift were the mechanism, the deterioration should show up broadly, across the whole book, roughly in line with how peers' NPAs moved. It didn't. Peers were flat. The hinge assumption has just failed a test nobody ran before the board memo went out.

Step 2, build the bespoke cut. Split the ₹1,000 crore book into its three branch clusters and compare prior-year and current-year NPA in ₹ and %:

Step 3: reconcile and read it. Check the totals first. 12.00 + 10.50 + 9.50 = 32.00, which is 3.2% of 1,000 and matches the reported prior figure. 13.00 + 11.20 + 31.80 = 56.00, or 5.6% of 1,000, which matches current. Now look at Cluster C. It's a quarter of the book, and it produced ₹22.30 crore of the ₹24.00 crore total increase, 92.9% of it, tripling from 3.80% to 12.72% while A and B stayed essentially flat. You're looking at a single-cluster story, and industry structure doesn't come into it. The actual mechanism would turn up in the bespoke second-level cut within Cluster C, by loan officer, by vintage month, by product. In this synthetic case, it's a collections-vendor change that took effect in Cluster C alone three months before the spike. Five Forces would have told the board the sector faces headwinds, which is true and answers nothing. The bespoke tree found the ₹22.3 crore.

Read it. The rival-hypothesis check applies directly. If the "sector headwinds" story were right, you'd expect roughly proportional deterioration across all three clusters, weighted by book size. Instead, 93% of it is concentrated in 25% of the book. That mismatch should have stopped the Five Forces slide before anyone built it.

Worked example 2: A same-store-sales bridge instead of a BCG matrix at Fairview Home Goods (US, $)

Fairview Home Goods, a synthetic mid-size US home-goods chain, reports same-store sales down roughly 6% for the quarter across its mature store fleet. A regional VP proposes running a BCG growth-share matrix across the company's three store formats (flagship, express, outlet) to decide which to "invest, hold, or harvest." But every format is a mature, low-growth business with a stable relative position. The matrix's own hinge assumption, that market growth proxies cash need and share proxies cost position, has almost nothing to bite on, because nothing about the puzzle is a growth-versus-share portfolio question. What you actually need to know is why an already-mature, already-classified set of formats suddenly moved.

Solution. Step 1, decompose the metric, not the portfolio. Same-store sales is a product of three levers: traffic (customers per store), conversion (share of visitors who buy), and average basket (dollars per transaction). This quarter, traffic moved −9.0%, conversion +2.0% and basket +1.3%.

Step 2, reconcile the arithmetic. (1 − 0.090) × (1 + 0.020) × (1 + 0.013) = 0.910 × 1.020 × 1.013 = 0.9282 × 1.013 = 0.9403, a −5.97% net change. That matches the reported "down roughly 6%" to the decimal.

Step 3, read the decomposition, not the label. Conversion and basket both improved, so the merchandising and pricing team did their job. The entire decline is a traffic problem, and only a traffic problem. Fewer people are walking into the stores, and once inside, they're converting better and spending more per visit than a year ago. A BCG matrix can't produce a specific, falsifiable finding like that even in principle. "Invest/hold/harvest by format" was never the axis this quarter's number moved along. The mechanism, in this synthetic case a nearby competitor's new format opening in the same trade areas, is inside a single lever and doesn't spread across the whole portfolio.

Read it. Why call this "framework-as-theater" and not just "the wrong tool"? Because running the BCG matrix here wouldn't have been incorrect mechanically. Each format really does have a growth rate and a relative share you could plot. You'd have got a technically valid quadrant chart that answered a question nobody asked this quarter ("how should we allocate capital across formats long-run?"), while the real, falsifiable, one-lever answer sat one level of decomposition away the whole time.

Worked example 3: Failing to face the challenge: the well-documented pattern behind two famous disruptions (US, historical, qualitative)

Not every worked example needs a fabricated number to make its point, and this one deliberately carries none. It's here because the "failure to face the challenge" face of bad strategy (Core teaching content, above) is easiest to recognize in hindsight, at a safe distance from any live engagement, in situations that are exhaustively public record. Two of the most widely taught cases in the strategy literature fit. One is Kodak's documented internal response to the rise of digital photography. The other is Blockbuster's decision around the year 2000 not to acquire or seriously engage with what became Netflix. Both are discussed at length in Rumelt's own book, in Christensen's disruption literature, and in each company's own subsequent public filings and retrospectives. Both companies later filed for bankruptcy, a fact of public record. Treat the specifics below as the narrative most published accounts give, not as the course's own research, and check them against the Reading & Resources list if you want the primary sources.

Solution. Step 1, name the pattern, not the company's failings generically. In both cases, the public record shows leadership had credible internal information about the coming shift well before it became existential. By most published accounts, Kodak had early digital-imaging research of its own. By the same accounts, Blockbuster's own executives evaluated the smaller subscription-by-mail rival and passed on engaging with it. In neither case was the information absent. What was absent, on the public record, was a diagnosis that faced the specific, threatening conclusion the information pointed to.

Step 2, see the framework's role. The retrospective business literature commonly reports that contemporary strategic reviews at firms in this position reached comfortable conclusions using standard tools. Portfolio logic reasonably classified the core, cash-generative business as the priority to protect. Competitive analyses at the time reasonably sized the new entrant as a niche threat and not a boundary-dissolving one. Each was correctly run, by the standards of the day. The diagnosis a Rumelt-style read demands is "our core business is being substituted away, and 'protect the core' is therefore not a guiding policy but a failure to face the challenge." In retrospect, accounts report that this was the sentence that didn't get written, or was written and not acted on with the urgency the threat warranted.

Step 3, connect it back to your own work. Don't take away "always assume disruption." The lesson is narrower. A framework's output can be a perfectly reasonable, well-supported conclusion given its own hidden assumptions, which here were roughly that the current industry boundary would persist. Reasonable-given-its-assumptions and correct are different claims. "Failure to face the challenge" hides in the gap between them, invisible to anyone who only checks whether the framework was run well.

Read it. This is the highest-stakes version of the pattern in Worked Examples 1 and 2. In those, a framework got misapplied to a narrow operational puzzle. Here, a comfortable, well-executed strategic read never got tested against its own hinge assumption at the moment it mattered most. The fix is the same at every scale. State the assumption, and go find out, before the failure becomes public record too.

Worked example 4: What a BCG "Cash Cow" hides at Konark Industries Group (India, ₹)

Konark Industries Group, a synthetic diversified Indian conglomerate, runs a standard portfolio review across four business units, plotting market growth against relative share:

Straight off the matrix, the board's instinct is to harvest Electricals (a "mature Cash Cow") and keep funding Renewables ("tomorrow's Star"). One number should have stopped the meeting. Textiles, the other labeled Cash Cow, earns 14.4% ROIC. Electricals, labeled identically, earns 6.7%. That's well below what a mature, self-funding cash cow should look like, and below the group's cost of capital.

Solution. Step 1, check the hinge assumption on the odd one out. BCG bets that low growth plus high relative share should predict a healthy, self-funding return. The whole case for "harvest it" rests on that. Electricals fits the growth/share box but not the return the box predicts. That mismatch is the tell. The matrix alone will never explain it, because the reason isn't on either of its two axes.

Step 2, build the bespoke cut by following the related-party transaction. Konark's disclosure notes (synthetic, internally consistent) show Electricals sells 40% of its output, ₹240 crore of its ₹600 crore revenue, to sister unit Renewables. It sells at a related-party transfer price 15% below the arm's-length price for the same product. Recompute at arm's length. The arm's-length value of that internal revenue = ₹240 cr ÷ 0.85 = ₹282.35 cr, an uplift of ₹42.35 crore that's currently counted in Renewables' books instead of Electricals'.

Step 3, reconcile both corrected ROICs. Corrected Electricals EBIT = 30.0 + 42.35 = ₹72.35 cr, so corrected ROIC = 72.35 ÷ 450 = 16.1%. That makes it the group's best-performing unit, and a healthy cash cow after all. Corrected Renewables EBIT = 10.0 − 42.35 = −₹32.35 cr, so corrected ROIC = −32.35 ÷ 300 = −10.8%. The business that looked like tomorrow's Star is, at arm's-length prices, destroying value. It's propped up entirely by an internal subsidy the matrix had no way to see.

Read it. The BCG matrix didn't lie about growth rates or relative share. Those inputs were correct. It misled by design, because its two axes can't see a transfer-pricing mechanism. And "harvest the Cash Cow, fund the Star" is the comfortable, textbook-sounding conclusion that a related party benefiting from the current pricing arrangement would want the board to reach without asking why. That's a lot of value to hide behind a quadrant label. The bespoke question, "why does a mature, high-share unit earn a below-cost-of-capital return?", is the one-sentence diagnosis the matrix skipped straight past.

Worked example 5: A win-rate bridge instead of 7S at Northbridge Analytics (US, $)

Northbridge Analytics, a synthetic B2B SaaS company, watches its competitive win rate fall from 38% (76 of 200 competitive deals won) to 24% (48 of 200) in one quarter. The VP of Sales proposes a full 7S organizational review. Are strategy, structure, systems, staff, skills, style, and shared values still aligned? It's a reasonable-sounding instinct, and it bets, without saying so, that the cause is internal.

Solution. Step 1, check the hinge assumption before scheduling the 7S workshops. If the cause were internal misalignment, you'd expect something internal to move. Headcount, tenure, sales-cycle length, tooling, or onboarding quality should show some kind of shift in the same window. They don't. Seller headcount (42 vs. 43), average sales-cycle length (54 vs. 55 days), and onboarding NPS (71 vs. 69) are all essentially flat quarter over quarter. Nothing internal changed enough to explain a 14-point win-rate collapse.

Step 2, build the bespoke cut instead, a loss-reason bridge. Of the 152 deals lost this quarter, 108 (71%) were coded price/bundle-versus-rival, 20 (13%) feature gap, and 24 (16%) other/no-decision. That's 108 + 20 + 24 = 152, so it reconciles. Now cross-reference the timing. The jump in "price/bundle" losses began the same week a named competitor rolled a previously paid add-on module into its base tier at no extra cost.

Step 3: reconcile and read it. Nothing about Northbridge's own seven S's moved. Something in the competitive landscape did: a specific, dated, external pricing action that shows up in 71% of this quarter's losses. 7S's hidden bet, that the problem lives inside the firm, fails the same test Worked Example 1's Five Forces bet failed. The data simply isn't shaped the way the framework's assumption requires.

Read it. This is the mirror image of Worked Example 1. There, a team reached for an external framework, Five Forces, when the mechanism was internal and firm-specific. Here, the reach is for an internal framework, 7S, when the mechanism is external and competitor-specific. The lesson goes beyond either case. Check whether the framework points in or out, and compare that with where the metric's own movement is concentrated, before you assume they match.

Worked example 6: When Five Forces is exactly the right first move: Chakra Quick Commerce (India, ₹)

Here's a contrast case, on purpose. Chakra Quick Commerce, a synthetic 10-minute-delivery operator profitable in its home city, is deciding whether to enter three new metro markets. This time, run the three-part decision rule from Core teaching content before reaching for anything.

(1) Is this a genuine, bounded structural question about a market as a whole? Yes. "Should we enter these markets" is an industry-structure question, not a read on one existing store's performance. (2) Does Five Forces' hinge assumption plausibly hold? Check it. Over the past 18 months, six well-funded new entrants have launched in the same four target metros. Average per-order delivery-fee discounting has deepened industry-wide from ₹0 to roughly ₹35. Dark-store real-estate lease rates in those metros are reported up about 20% as new entrants bid against each other for the same locations. Those are structural signals, and none of them is one firm's execution story. (3) Would a bespoke tree just re-derive Five Forces anyway? Almost certainly, yes. "Who are the new entrants, what do suppliers (dark-store landlords, delivery-partner labor pools) charge, how price-sensitive are customers, what substitutes exist (larger-basket weekly grocery trips, other apps)" is Five Forces' own question set, asked from scratch.

Solution. Step 1, accept the framework, on the evidence, not on habit. All three conditions hold, so the right move is to run a real Five Forces, done properly, on these four target metros specifically. Skip the generic recital. Measure rivalry intensity by the discounting trend above, supplier power by the lease-rate trend, and buyer power and substitutes by market-specific data.

Step 2, what refusing it here would cost. A bespoke tree would mean reinventing, slide by slide, a structure Five Forces already supplies. It would be slower, with more room to leave out a force by accident, most often substitutes, that the named framework would have forced onto the page.

Read it. The meta-skill you're building is "check the hinge assumption before committing to any structure, named or invented." It was never "prefer bespoke structures." Here, checking is what tells you to stop checking and start running Five Forces the way you were already taught.

Worked example 7: The AI mirror in practice: Solace Skincare's CAC "crisis" (US, $)

Solace Skincare, a synthetic single-brand D2C skincare company, sees blended customer-acquisition cost (CAC) rise from about $27.50 to $46.00 in six weeks, up 67%. A team member pastes the situation into a chat assistant: "Our CAC has risen sharply and our category is getting more competitive, what's our strategic problem?" The assistant returns a fluent Five Forces-and-SWOT writeup about rising rivalry in the DTC skincare space, increasing buyer power from price-comparison tools, and the threat of low-cost private-label substitutes. Every sentence is plausible. None of it was checked against Solace's own two-channel spend data, which the team never gave it.

Solution. Step 1, build the bespoke cut the AI was never asked to build. Solace runs two acquisition channels, interest-based targeting (Channel A, cheaper) and lookalike-audience targeting (Channel B, pricier). Before, 75% of orders came from A at $20 CAC and 25% from B at $50 CAC, so blended = 0.75 × 20 + 0.25 × 50 = $27.50. Then a platform policy change restricted Channel A's targeting reach. The mix shifted to 25% from A at $22 and 75% from B at $54, so blended = 0.25 × 22 + 0.75 × 54 = $46.00.

Step 2, reconcile and locate the real mechanism. Channel-level CAC barely moved. A went from $20 to $22, up 10%, and B from $50 to $54, up 8%. Nothing about "the category getting more competitive" shows up at the channel level. The mix moved. The cheap channel shrank from three-quarters of volume to one-quarter because of a platform-side targeting restriction, and no market-wide competitive shift was involved. The 67% blended jump is a weighted-average mix effect wearing a "rising rivalry" costume.

Step 3, run the same test on the AI's answer. Does the AI's Five Forces/SWOT writeup name a mechanism specific to Solace, or could it describe any DTC skincare brand's quarter? It could describe any of them. That makes it framework-as-theater, generated with the same fluent confidence whether or not it was actually checked. The reason is the one you met when you learned how modern AI works. The model produced the most plausible continuation of "competitive," not the true mechanism behind Solace's own number, because Solace never gave it the two-channel data that would have let it try.

Read it. Give the same tool the channel-level data and ask it to name one falsifiable mechanism, instead of "the strategic problem" in the abstract, and it becomes useful for widening the menu of hypotheses fast. Prompted the way the team actually prompted it, it reproduced, at machine speed, the very failure you've been learning to catch in people.

Practice set

Work each problem fully before you read its solution, and write before you peek. Where a computation appears, assume a calculator. Numeric answers are to about ±3%. All company figures are illustrative and synthetic unless explicitly marked as public record.

P1 (guided). Rumelt's kernel, and where theater hides. State Rumelt's three-part kernel of strategy in one line each, and say which part "framework-as-theater" most commonly fakes, and why that specific part is the easiest one to fake without anyone noticing.

Solution. Diagnosis: simplify the tangled situation down to its critical, addressable structure. Guiding policy: the overall approach chosen to cope with the diagnosed obstacles. Coherent action: coordinated, resourced steps that carry the policy out. Framework-as-theater fakes diagnosis. Once every box is filled and every force rated, a completed framework looks just like a finished diagnosis from the outside. Only checking whether any decision actually rested on one of its conclusions shows that none did.

P2 (guided). From vague mandate to falsifiable diagnosis. A mandate arrives: "margins are under pressure: fix it." Using why-laddering and the "what would have to be true" test, turn this into one falsifiable, mechanism-level candidate diagnosis (not a framework name).

Solution. Why are margins under pressure? Because cost per unit rose faster than price. Why did that happen? Stop laddering here, because it's already testable. If this were industry-wide input-cost inflation, the rise would have to be broad-based across product lines and roughly track the relevant commodity or wage index. The falsifiable candidate diagnosis: "if margin pressure tracks the commodity index and is broad-based across lines, it's a sector story; if it's concentrated in one line or one customer contract, the mechanism is elsewhere (mix, a specific renegotiated contract, or discounting) and a bespoke cut by line/contract is the next step." Either branch is checkable this week, and that's what makes it a diagnosis and not a restated symptom.

P3. The BCG label that shouldn't be trusted on sight. Business unit P has market growth of 4% and relative share of 2.1×, so BCG says Cash Cow. Its reported ROIC is 4%, below the group's 11% cost of capital, and its capital employed has grown 30% over two years while EBIT grew 2%. Identify the contradiction the label hides, and state the one question you'd ask before accepting "harvest it."

Solution. A real Cash Cow should need little capital and earn comfortably above the cost of capital, and the whole logic of "harvest, don't reinvest" depends on that. Unit P is below the cost of capital, and its capital base is growing fifteen times faster than its profit. That's the opposite profile. The one question to ask is "why is capital employed growing 15× faster than EBIT in a unit the framework says should be a self-funding harvest asset?" The answer, whether it's heavy capitalized costs, an internal-allocation distortion, or an asset that's really deteriorating, determines whether "harvest" is right at all.

P4. Right grain, wrong grain. A regional cement producer holds 8% share in one state against a leader with 45% share and is deciding whether to raise price there. A colleague proposes a full national Five Forces on the Indian cement industry. Is that the right first move? What would you propose instead?

Solution. No. The question is firm-specific and single-state, and it isn't an industry-wide structural question. Running Five Forces at the national-industry grain answers a question nobody asked while the actual pricing decision waits. Propose a bespoke cut instead: the plant's freight-adjusted cost position in that specific state versus the leader's, current capacity utilization / local demand-supply balance, and the leader's historical reaction pattern to past price moves in that state (did it match, undercut, or ignore?). A national Five Forces slide can't see that mechanism.

P5 (guided). First-level MECE cut for a bespoke tree. Contribution margin in Region X fell 4 points, from 34% to 30%, this quarter. Build a first-level MECE cut for the bespoke tree (branches only, not the full numbers) and say what data you would pull for each branch.

Solution. Four branches exhaust everything that can move a contribution-margin line. Price/mix covers list-price changes and product-mix shift toward lower-margin SKUs. Pull the price-realization and mix report for it. Input cost is COGS per unit, and you'd pull the standard-cost bridge. Discounting/promotions means off-invoice and trade spend, which needs the promotions ledger. Freight/logistics is per-unit shipping and warehousing, and the logistics cost report covers it. Check MECE by asking whether any other ledger line could move contribution margin outside these four. If you can't name one, the cut is complete.

P6. Reconcile a fresh roll-rate table (India, ₹). A ₹600 crore loan book's NPA ratio rose from 2.5% (₹15.00 cr) to 4.0% (₹24.00 cr). Cluster X holds ₹450 cr (prior NPA 2.20% = ₹9.90 cr; current NPA 2.40% = ₹10.80 cr). Cluster Y holds the remaining ₹150 cr. Compute Cluster Y's current NPA in ₹ and %, and state what share of the total deterioration it explains.

Solution. Total current NPA is ₹24.00 cr and Cluster X's is ₹10.80 cr, so Cluster Y's current NPA = 24.00 − 10.80 = ₹13.20 cr. That's 13.20 ÷ 150 = 8.80% of its book. Its prior NPA was 15.00 − 9.90 = ₹5.10 cr, or 5.10 ÷ 150 = 3.40%. Total deterioration = 24.00 − 15.00 = ₹9.00 cr, and Cluster Y's deterioration = 13.20 − 5.10 = ₹8.10 cr. That's 8.10 ÷ 9.00 = 90% of the total increase, concentrated in a cluster holding only 25% of the book. It's the same pattern as Worked Example 1, with fresh numbers.

P7. Reconcile a same-store-sales bridge in reverse (US, $). Traffic +3.0%, conversion −7.0%, basket +1.0%. Compute the net same-store-sales change and say which lever is driving it.

Solution. (1.030) × (0.930) × (1.010) = 1.030 × 0.930 = 0.9579, and 0.9579 × 1.010 = 0.9675, a −3.25% net change. Traffic and basket both improved, so the entire decline is a conversion problem, and then some. More people are visiting and spending slightly more per visit, but a shrinking share of them is completing a purchase. That points at a checkout, staffing, or in-store execution mechanism instead of a demand or pricing one.

P8. Spot the fluff and the goal-substitution. Four "so what" bullets from a real-looking deck: (a) "We must harness synergistic capabilities to unlock stakeholder value across the ecosystem." (b) "Branch cluster C's NPA rate tripled after the March collections-vendor change; the other two clusters were flat." (c) "Our strategic priority is operational excellence." (d) "Grow revenue 20% while becoming the most-loved brand in the category." Identify which pass the one-sentence/falsifiability test and which don't, and name the bad-strategy face each failing one commits.

Solution. Only (b) passes. It names a mechanism, a location, and a trigger, and it would be straightforwardly false if the deterioration were broad-based. (a) is fluff. It's jargon-dense and says nothing that would look different if false. (c) is mistaking a goal for a strategy, and arguably fluff too, since "operational excellence" names an aspiration with no obstacle and no mechanism attached. (d) is also mistaking goals for strategy. It states two ambitions together with no guiding policy connecting them and no obstacle named for either.

P9. The laundry-list problem. A company's annual plan lists eleven "strategic pillars" for the year, unranked, with no resource tradeoff stated between any two of them. What is missing, and rewrite two of the eleven into a single, prioritized guiding-policy sentence that says no to something?

Solution. What's missing is a ranking and a stated tradeoff. A plan that doesn't choose gives every department a permission slip to keep its existing budget, and that's no strategy. Model rewrite (illustrative pillars "expand into three new states" and "improve service-center response time"): "This year we will fund service-center response-time improvement ahead of new-state expansion, because retention economics in our existing footprint currently return more per rupee of capital than a new-state launch would. New-state expansion is paused, not merely deprioritized, until the response-time program hits its target." That sentence names a ranking, a reason, and an explicit sacrifice, the three things the original eleven-pillar list lacked entirely.

P10. Right lens, wrong direction. A regional airline's on-time performance has fallen sharply at exactly one hub; performance at every other hub is unchanged. The COO's staff propose a full 7S organizational review. Is that the right first move?

Solution. No. 7S bets the cause is internal misalignment across the whole organization. A symptom concentrated at a single hub, with every other hub unaffected, points at something local and specific, a hub-level operational mechanism, and away from firm-wide alignment. Propose a hub-level delay-cause bridge instead (weather, crew availability, ATC, mechanical, gate/ground-handling) at the affected hub. Benchmark it against the same cause codes at the unaffected hubs over the identical window. It's the same "check where the symptom is concentrated before choosing an internal-vs-external lens" test as Worked Example 5.

P11. MECE-test a proposed tree. A proposed bespoke tree for "why did contribution margin fall" has three branches: input cost, price/discounting, product mix. The company also changed its freight carrier and renegotiated warehousing this quarter. Is the tree MECE? If not, name the gap.

Solution. It's not MECE. Freight/logistics cost is missing as its own branch. Depending on how the company's ledger codes freight, it may sit partly inside "input cost" (if capitalized into COGS) or nowhere in the three stated branches (if it's a separate operating-expense line). Either way, a change in carrier and warehousing terms this quarter is the kind of concentrated, dateable event you're learning to chase, and the tree as stated can't even register it. Add a fourth branch, and re-run the branch-removal check to confirm all four now carry weight.

P12. The one-sentence test. A diagnosis paragraph reads: "Our performance this quarter reflects a complex interplay of macroeconomic headwinds, evolving customer expectations, and an increasingly dynamic competitive landscape, all of which have created a challenging environment that requires us to remain agile and customer-centric as we navigate forward." Compress it to one sentence, or explain why it cannot be compressed.

Solution. It can't be meaningfully compressed, because it doesn't survive compression. Strip the jargon and nothing falsifiable remains. "Things are complicated and we should stay adaptable" is true of every company in every quarter. This is fluff by Rumelt's definition. It would read identically whether the underlying quarter was a triumph or a disaster, and that's the test for whether a sentence contains a diagnosis at all.

P13. When to trust the canonical tool. A US regional bank is deciding whether to enter the buy-now-pay-later (BNPL) space as a new business line. Apply the three-part decision rule: is this a genuine structural question, does a standard framework's hinge assumption plausibly hold, and would a bespoke tree just re-derive it anyway?

Solution. Yes to all three. Entering a new market is a textbook bounded structural question, not one existing business line's performance puzzle. A real BNPL industry of incumbents, new entrants, suppliers (funding costs, card networks), and substitutes (traditional credit, layaway) exists, so Five Forces' hinge assumption plausibly holds. And a from-scratch bespoke tree would very likely re-derive something close to Five Forces' own five categories. So run Five Forces, plus an Ansoff-style entry-mode framing, as the right first move. Here the canonical frameworks get used as intended, and nothing calls for building something bespoke.

P14. Diagnose the AI's over-fit. A learner pastes: "Our single-SKU candle brand's contribution margin fell from 42% to 31% this quarter, what's our strategic problem?" into a chat assistant, which returns a full Five Forces and SWOT writeup about the candle industry's competitive intensity. Diagnose the failure and state the better first move.

Solution. The puzzle is a single-SKU, single-quarter margin-bridge question. Input cost, price/discounting, freight, and mix are the only things that can move a contribution-margin line this fast for one product. It isn't an industry-structure question. Neither Five Forces nor SWOT could explain it even in principle, because an eleven-point single-quarter drop for one SKU can't be "rivalry intensity" at that speed and grain. The better first move is to build the same four-branch bespoke bridge as P5 and P11, using the SKU's own cost and pricing ledger, before asking any question about the wider industry at all.

P15. Write the refusal script. A partner says: "Just give me the Five Forces slide, the client expects to see one." You believe the branch-level data points somewhere else entirely. Write the one or two sentences you would actually say.

Solution. Model script: "I can build that slide by Thursday, but before I do, let me show you the branch-level cut I already have. If the deterioration is as concentrated as it looks, Five Forces will tell the client the sector's fine and miss the one thing that's actually broken, and we'll have spent the week answering a question they didn't ask. Five minutes, then you decide which slide goes in." Look at what the script does. It proposes a fast, falsifiable alternative instead of bare skepticism. It names the specific cost of getting it wrong. And it hands the decision back to the partner instead of refusing outright.

P16. Capstone synthesis. A synthetic scenario: a chain of 40 fitness studios sees membership churn rise from 4% to 9% monthly, concentrated almost entirely among members who joined via one referral-partner channel in the last two quarters, while members from all other channels churn at the old 4% rate. Produce (a) a one-sentence diagnosis, (b) a guiding policy, (c) one coherent action, and self-check against Rumelt's kernel.

Solution. (a) Diagnosis: "Churn rose because the referral-partner channel is delivering members with a fundamentally worse fit for the studio model, not because retention broke down generally." It's falsifiable: if true, non-referral-channel churn should stay near 4%, and it does. (b) Guiding policy: "We will treat the referral-partner channel as a targeting and screening problem to fix at the source, not a retention problem to fix after the fact. We will not spend retention budget on symptoms of a channel-fit issue." (c) Coherent action: renegotiate or pause the referral-partner deal's lead-qualification terms, and route referral-channel signups through the same trial-class screening step used for other channels, before spending on win-back campaigns. Now self-check. The diagnosis names a specific, falsifiable mechanism. It doesn't just restate a symptom. The guiding policy rules something out (spending retention budget on symptoms) as much as it commits to something. The action is a single, resourced, sequenced step, not a list. All three kernel parts are present.

Applied mini-project

Diagnose, refuse, rebuild.

It's modeled on the "predict, verify, diagnose" loop of the project in How Modern AI Actually Works, applied here to structures instead of AI outputs. You'll take three business situations, the two below plus one you supply yourself from something you've read, worked on, or observed, and run the full procedure on each, end to end.

The two supplied situations:

1. A logistics company's on-time delivery rate has fallen 6 points nationally over two quarters, and a team has already drafted a 7S organizational review to present next week.
2. A consumer-electronics brand's gross margin has compressed 3 points in its home market only (every other market flat), and a team is proposing a full generic-strategies review (cost leadership vs. differentiation) to decide "what we stand for."

Your own situation. A real or realistic business puzzle you have encountered (at work, in the news, in a case you've read) stated as a metric that moved plus enough context to work with (100–150 words, no conclusion).

Do this, for each of the three, in order:

1. Name the reflex. State the framework a less experienced team would reach for by default (it may be the one already drafted, as in the two supplied situations).
2. State the hinge assumption. Name, in a single sentence, the assumption that framework requires to be true here.
3. Test it: predict before you check. State what data pattern would have to be true for the reflex framework's assumption to hold, then check it against whatever data the situation gives you (or state plainly what you'd need to pull and what result would settle it).
4. Decide: keep or refuse. Using the three-part decision rule, decide whether the canonical framework belongs here or whether it's framework-as-theater waiting to happen.
5. If you refuse it, build the bespoke tree. A MECE first-level cut keyed to the actual mechanism, with the one-sentence test and the rival-hypothesis check both run against it.
6. State the kernel. One-sentence diagnosis, one guiding-policy sentence that rules something out, one coherent first action.

Deliverable and self-check. One worksheet covering all three situations, each with steps 1–6 completed. Score yourself 2 points for each of these. (a) You stated the hinge assumption as a testable claim, not a restated framework name. (b) You actually predicted before checking, writing the prediction before the "check" step and not reverse-engineering it after. (c) You applied the decision rule explicitly, with a reason, even when the answer was "keep the framework" (Worked Example 6 style). (d) Where you built a bespoke tree, it passes both the one-sentence test and the rival-hypothesis check. (e) All three kernel parts are present in the final synthesis, with the guiding policy stating something it rules out. Pass at 8/10, with no zero on any item.

Mapped challenge. In the app, the mapped challenge is the framework-recognition drill (Practice → Consulting tab). Given a fresh business situation, you name the one structure that's most useful and, where a named framework would mislead, say which one and why. It's the same judgment you just ran three times above, auto-checked at speed. Passing the mastery quiz plus this drill completes the node.

Reading & resources

Treat every item on this list as a lens to think with, never a template to fill in. It's the same habit you've been building, applied to its own reading list. Items are free unless marked, and level tags run from beginner to advanced.

The spine:

• Richard Rumelt, Good Strategy / Bad Strategy. The source of the kernel (diagnosis, then guiding policy, then coherent action) and of the four faces of bad strategy that everything here is built on. Read the first four chapters before anything else on this list. [Paid] [Intermediate–Advanced]

Structuring the diagnosis, once you've named it (cross-links, not re-teaching):

• Barbara Minto, The Pyramid Principle. How to present the bespoke diagnosis you've built, answer-first. That's the hand-off named explicitly above, to CN1.03 and to the Finance galaxy's M10.03 (The Communication Toolkit). [Paid] [Advanced]
• Conn & McLean, Bulletproof Problem Solving. The seven-step problem-solving cycle and the issue-tree methods CN1.01/CN1.02 already taught. Reread its chapter on hypothesis generation with the hinge-assumption lens on top. [Paid] [Beginner–Advanced]
• Ethan Rasiel, The McKinsey Way / The McKinsey Mind. Practitioner-level color on structuring under time pressure. [Paid] [Beginner]

The frameworks tested here, not re-taught (CN3.01/CN3.02's canon):

• Michael Porter, Competitive Strategy / Competitive Advantage, and Joan Magretta, Understanding Michael Porter. Read Magretta's chapter on what Five Forces and generic strategies are not claiming, the clearest published statement of the hinge assumptions named above. [Paid] [Intermediate]
• Hamilton Helmer, 7 Powers, and Bruce Greenwald, Competition Demystified. Two more structural lenses, each with its own hinge assumption. They're cross-linked to the Finance galaxy's M4.04–M4.05 (moats) and not re-taught here. [Paid] [Advanced]
• Lafley & Martin, Playing to Win. A guiding-policy-shaped alternative to a framework-shaped one, "where to play, how to win" as a pair of real choices, not a template. [Paid] [Intermediate]
• Kim & Mauborgne, Blue Ocean Strategy, and Clayton Christensen (jobs-to-be-done, The Innovator's Dilemma). Read for the hinge assumption each one bets on (Core teaching content, above), not for new mechanics. [Paid] [Intermediate–Advanced]

On the specific failure named here:

• Harvard Business Review (HBR). Search "framework" and "strategy process" for a steady stream of practitioner writing on this failure mode, from both sides. [Free/Paid] [Intermediate]
• Retrospective business-press and case-study coverage of strategic failures on the public record (Kodak, Blockbuster, and similar). Read it for the pattern (Worked Example 3), not for numbers to cite. The Rumelt and Christensen texts above are the sourced accounts.

Free practice:

• The MBA consulting-club casebooks (Wharton, Kellogg, Fuqua, Darden, INSEAD, LBS, Ross, Booth, Columbia), read three real interview transcripts specifically looking for the moment a candidate names or refuses a framework, and judge whether the refusal (or the acceptance) was earned. [Free] [Intermediate]
• caseinterview.com and the free official practice cases published by McKinsey, BCG, and Bain: same exercise. [Free] [Beginner–Intermediate]

Do this, not just read: take one framework-filled deck you can find (a public case study, a business-school write-up, or one of your own from CN3.01/CN3.02) and run the one-sentence test on its stated conclusion. If it survives, you've found a real diagnosis. If it doesn't, you've just caught, in the wild, the failure this whole topic exists to name.

Teach it back & journal

Feynman prompt. Explain to a colleague who has just learned to run a perfect Five Forces analysis why "I ran it correctly" and "it was the right tool for this problem" are two completely different claims. Use a real or invented example where a technically correct framework produced a deck that explained nothing about why a specific number moved. Then explain, in the same breath, why refusing every framework on principle isn't the fix either, using the three-part decision rule to show when the "wrong tool" accusation doesn't apply. If your colleague can, unprompted, state a framework's hidden assumption and propose one test for it before you've told them how, you've installed the meta-skill.

Journal prompt. For one week, watch for the moment (in your own work, in a meeting, in something you read) where a framework's shape is reached for before anyone has stated what specific, falsifiable question is actually being asked. Log each instance: what was the framework, what was the hinge assumption it required, did it actually hold, and what happened to the conclusion once someone checked (or didn't)? Then write a paragraph on this idea. A partner's judgment, compounded over hundreds of engagements, is mostly the accumulated experience of having been burned by this failure enough times to check for it by reflex. You can't get the hundred engagements here, but you can get the reflex early. Which of the checks you've learned would you have most needed this week: the hinge-assumption test, the one-sentence test, the rival-hypothesis check, or the branch-removal check? And what will you do to make running it automatic before you build the next deck?

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