Checklist, Temperament + Capstone

Phase 6 · Week 56 · ~20 hours

Where it sits. You now know how you will fail. That leaves exactly one question open: given that you now know how you will fail, what do you actually do about it? The answer is not "try harder" or "be more rational." Willpower loses to a live market and a falling price. The answer is machinery: a written checklist derived from real mistakes, a decision journal that lets you grade your reasoning honestly, an environment engineered so the right behavior is the default, and the deliberate cultivation of temperament over intellect. This is the final module of the philosophy phase and its capstone, the point where the whole first year of the program (weeks 1–52) turns from knowledge you have into a process you run. First came the schools; then the decision science (inversion, base rates, expected value, Kelly, ruin-avoidance); then the bias catalog, which named the twenty-plus biases and the failure mode each one produces at the moment of decision. You will build the first version of your own investment checklist, run a structured behavioral self-audit on yourself, and, for the M7 milestone, take one real company end to end through the Mental-Model & Moat Checklist you finalize here. It is what makes everything before it survive contact with your own psychology.

Suggested hour budget. Core content: 7h · worked examples (calibration math, the mistake-loop, the behavior gap, the self-audit): 2.5h · practice set: 3.5h · the capstone (checklist v1 + full Mental-Model & Moat Checklist on one company): 5h · flashcards + mastery check: 2h.

App link. The work here feeds the Philosophy & Behavior branch and wires directly into the tools you will live inside for the rest of the program: the decision-journal template (thesis, disconfirming evidence, base rate, confidence, "what would change my mind"; the same entity the Phase 9 portfolio simulator forces on every trade), the calibration dashboard (predicted-vs-actual, scoring reasoning over outcome; it starts collecting data the moment you file your first journaled thesis), the behavioral self-audit questionnaire (a scored instrument that surfaces your two or three dominant biases and suggests the checklist items and environment fixes that counter them), and your personal checklist as an editable, versioned document the app carries forward and resurfaces at every case challenge. The Mental-Model & Moat Checklist you finalize here becomes the standing rubric behind the "analyze this business" case engine (Phase 8).

Learning objectives

By the end you can:

1. Make the intellectual case for a written checklist: explain Gawande's distinction between errors of ignorance and errors of ineptitude, why checklists attack the second, why they catch "the dumb, avoidable, recurring stuff" rather than the exotic, and cite the surgical, aviation, and ICU evidence that a five-to-nine-item list changes outcomes even for world-class experts.
2. Run the personal mistake-to-checklist loop: take a real investing mistake (yours or a documented one), find its root cause rather than its surface symptom, and write a single pointed checklist question that would have caught it before the decision, in the manner of Pabrai's and Spier's mistake-derived lists.
3. Distinguish a killer checklist from a dead one: build a list that is short, specific, veto-shaped, and run as a pause at the end of your analysis (do-confirm), not a hundred-item scoring sheet or a mechanical buy-signal.
4. Operate a decision journal as the core instrument of a process: record thesis, disconfirming evidence, the base rate, a numeric confidence, and your "what would change my mind" before the outcome is known, never edit it after, and explain why this single discipline is the engine of calibration (the bridge to M9.04).
5. Compute and interpret calibration: bucket your probabilistic predictions, judge whether you are over- or under-confident, compute a Brier score, and state why scoring reasoning over outcome is the only honest scoreboard.
6. Design your environment (Spier): specify the information diet, the peer group, the physical and structural rules (the checking-the-price cure, the "don't sell for a set number of days" rule, starting from least-biased sources) that make discipline structural rather than a daily act of will.
7. Argue temperament over intellect: explain Buffett's claim that above ordinary intelligence the binding constraint is temperament, name the components (patience, decisiveness, equanimity in drawdowns, independence, intellectual honesty), quantify why avoiding large permanent losses matters (the drawdown-recovery asymmetry and the behavior gap), and say honestly how much of temperament is trainable versus innate.
8. Complete a behavioral self-audit and assemble Checklist v1: work a structured questionnaire that identifies your own dominant biases from the M6.04 catalog, convert each into a specific checklist item or environment fix, and assemble your personal investment checklist v1 from the program's master template plus your own logged items. Then run the finalized Mental-Model & Moat Checklist end to end on one real company.

Prerequisites & connections

Builds on. M6.04: the bias catalog is the raw material of the self-audit. Every checklist item and environment rule here exists to disarm a specific tendency you met there. Anchoring leads to stale price targets and confirmation to one-sided research. Loss aversion means riding losers and panic-selling. Overconfidence shows up as oversizing, and herding as buying at the top. M6.03: process-over-outcome, base rates and the outside view, expected value, and ruin-avoidance are the decision-quality standards the journal records and the calibration score measures. "What would change my mind" is Bayesian updating made operational. M6.01–M6.02: the schools you resonated with become lines in your checklist. Their masters are quoted here as temperament as well as technique: Graham's "the investor's chief problem is likely himself," Klarman's margin of safety, Marks's second-level thinking, Pabrai's asymmetry. M0.04: the knowledge system installed in week 3 (atomic notes, company files, the checklist, the decision journal). Here the checklist and journal stop being templates and become instruments you've filled with your own mistakes. M2 and M3 supply the analytical content the checklist gates. You can't check "have I run the quality-of-earnings screen" or "is ROIC durably above WACC" until you can do those things by hand.

Feeds forward. M7 milestone (the gate, week 52, ~1 year in): checklist v1 filed and behavioral self-audit written. Phase 7 (Macro): the checklist's macro-overlay line and the "what's priced in" habit carry straight into the dashboard you build. Temperament in drawdowns is the behavioral counterpart to Marks's cycles. M8.04: the deep dive's Day 3 runs your checklist and your pre-mortem, and the Mental-Model & Moat Checklist finalized here is the spine of the deep-dive workbook. M9.04: the process capstone closes the loop you open now. It runs from checklist to journal to quarterly review to calibration as one repeatable machine, over ≥10 journaled decisions and a full simulator season, and the calibration data you begin logging now is what M9.04 grades. M9.01–M9.03: position sizing, when to sell, and your mistake taxonomy all draw on the journal you start filling here. Competency C9 ("investor's temperament and process") is largely built and certified here and in M9.04. From now on, the behavioral edge stops being a topic and becomes your operating system.

The short version. You won't out-discipline a falling price, and you won't out-think your own biases in the heat of a decision. So you build the safeguards into the process. That means a short, mistake-derived checklist run as a final pause, and a decision journal written before outcomes and never edited after. It means an environment engineered so the sane choice is the easy one, and a temperament you cultivate on purpose. In investing, the person who reliably avoids the big, dumb, recurring errors beats the person with the higher IQ and no system.

Core teaching content

4.3 Building the checklist from your own mistakes: the personal loop

The masters didn't download their checklists. They bled them, one item per scar.

Mohnish Pabrai, after reading Gawande, set out to build an investment checklist the way medicine built its surgical one, empirically, from documented failures. He studied a large set of investing mistakes. Some were his own. Others were the disclosed errors of great investors like Buffett and Munger, reconstructed from 13F filings, letters, and post-mortems. For each one he asked a single forensic question: what, specifically, caused this permanent loss of capital, and what question, asked before the decision, would have caught it? Each answer became one line. The checklist grew to roughly seventy items, and it kept growing. Every item traced back to a real mistake that lost real money. Not a textbook risk. A grave with a name on it.

Guy Spier, who with Pabrai famously bid $650,100 for the 2008 charity lunch with Warren Buffett, describes the same practice in The Education of a Value Investor. After every serious error he asks what went wrong at the level of process. Then he adds a rule so that particular error can't recur silently.

The method is a loop you'll run for the rest of your investing life. You don't have many mistakes of your own yet, so learn it now on other people's, then feed it your own from week 53 forward. It has four steps:

1. Isolate a real mistake with a real loss (or a real near-miss). Keep it concrete, a specific decision with a specific loss or avoided loss. "I should be more careful" isn't a mistake. "I bought the NBFC and it fell 80%" is.
2. Find the root cause, not the symptom. This step is what separates a useful loop from a diary. Keep asking "why" until you hit something structural and recurring, something that isn't specific to that one company. The stock falling is the symptom. "I did not check the asset-liability maturity profile" is closer. "I trusted a reported growth number without asking how the growth was funded" is the root. "I anchored on a beautiful growth story and never sought the disconfirming cash-flow evidence" is the behavioral root, and that one will recur across completely different companies.
3. Write one pointed question that would have caught it before the decision. Phrase it as something you have to answer. A platitude like "be careful about leverage" won't do. Write something like "For any lender: does the asset-liability maturity profile show a negative gap in the short buckets, and how much of funding is short-term (commercial paper)? If yes, what happens when refinancing markets freeze?"
4. Judge whether it is general enough to keep. A good item recurs across many companies. A hyper-specific one just clutters the list. If the same root cause could sink three unrelated businesses, it gets a permanent line. If it couldn't, it belongs in a company file and stays off the master checklist.

Worked instance (India). Take the 2018 collapse of India's infrastructure-and-finance shadow-banking complex, where IL&FS went first and DHFL followed. You take the case apart in the forensic-cases capstone, and again with the NBFCs and HFCs. On the surface the mistake was "owned an NBFC that defaulted." The symptom-level lesson, "avoid NBFCs," is useless, because excellent NBFCs exist. Push to the root. These lenders funded long-dated, illiquid infrastructure and mortgage assets with short-term borrowings, heavily commercial paper. That left them structurally dependent on rolling over their funding every few months. When sentiment turned and the commercial-paper market froze, the assets couldn't be sold fast enough to meet the maturing liabilities. It was a classic asset-liability mismatch, the same mechanism as a bank run. So the root cause of an investor's loss wasn't bad luck. It was failing to check the funding structure of a business whose entire risk lived in its funding structure. That gives you a permanent, general checklist item. It would also have flagged a US savings-and-loan in 1989 or Northern Rock in 2007. One scar, one line, transferable across decades and continents.

Worked instance (US). A generation of investors lost money in Valeant Pharmaceuticals around 2015. Valeant was a serial acquirer that reported dazzling "adjusted" earnings, while its GAAP numbers, its debt, and its dependence on price hikes and an opaque specialty-pharmacy channel told a different story. Symptom: "the stock fell 90%." The root cause was believing an adjusted-earnings narrative from a debt-fueled roll-up without reconciling it to cash flow and organic growth, and without asking whether the growth engine was durable or a treadmill of ever-larger acquisitions and price increases. Checklist line: "For any serial acquirer: what is organic growth stripped of acquisitions? Do GAAP earnings and free cash flow corroborate the adjusted story? Is the business a compounding machine or an acquisition treadmill that stops the moment deals or price hikes stop?" That single item came out of one American disaster, and you can carry it like a lantern into any roll-up in any market.

Look at what the loop gives you. The checklist that comes out is yours. It's built from evidence and aimed at how you behave. It lists the specific ways you, and the masters you learned from, actually lose money, phrased as questions you have to answer before you're allowed to act. That's far more powerful than a generic list of good ideas. Think of the difference between "diversify" on a poster and a scar that flinches at a familiar shape.

4.4 The decision journal: the core instrument

If the checklist is the machine that stops you skipping steps, the decision journal is the machine that tells you whether your thinking was any good. Of all the habits in the course, it pays back the most. Almost no one keeps one. The people who do compound an advantage that's nearly impossible to copy, because it rests on a record only they have. That record is a time-stamped account of what they believed before they knew how it turned out.

What problem does it solve? Human memory is a reconstruction. Whatever happened next edits it continuously and unconsciously. After a stock triples, you "always knew it was special." After it halves, you "had a bad feeling about it." That's hindsight bias, and it's fatal to learning, because it destroys the raw material learning needs, an accurate record of your prior state of mind. If you can't remember what you thought and how sure you were, you can't tell a good decision from a lucky one, or a bad decision from an unlucky one. You'll "learn" the wrong lessons with total confidence. The journal defeats hindsight bias by the only method that works: write your reasoning down before the outcome exists, and never edit it afterward.

A decision-journal entry is written at the moment of decision. At minimum it contains:

• The thesis, in a few sentences. What is this business, why is it a good business, and why is it mispriced? What are the two or three things that must go right? This is your answer to the five questions, compressed.
• The disconfirming evidence, written by you, against yourself. The strongest version of the bear case. What would someone who thinks you're wrong say? You can't skip this, and it isn't decoration. Writing the counter-case in your own hand is the most direct antidote to confirmation bias you have. If you can't state the bear case better than a bear could, you don't understand the situation.
• The base rate. What's the outside view? Take companies like this one, with the same growth claim, the same leverage, the same roll-up structure, the same turnaround promise. What fraction actually delivered? Mauboussin's Base Rate Book is the place to look. You're pricing your specific story against the historical batting average of stories like it.
• Your confidence, as a number. Not "I'm fairly confident." A percentage. "I think there is a 65% chance the thesis plays out over three years." Repeat that one number across dozens of decisions and calibration becomes something you can measure, as the next section shows.
• "What would change my mind": the pre-committed kill criteria. The specific, observable events or data that would tell you the thesis is broken: the metric that must hold, the number that must not deteriorate, the disclosure that would end it. You write them now, while you're calm and have no position at stake. Later, when you're down 30% and desperate to rationalize, you can consult a decision made by your clearer-headed earlier self. It's Bayesian updating, loaded in advance. It's also your defense against the sunk-cost and loss-aversion reflex, the urge to hold a broken thesis because selling means admitting a loss.
• The sizing rationale. Why this size and not larger or smaller? It's conviction weighed against downside, and it's the bridge to position sizing later on.

Two rules make it work, and both are in the course's working agreements. First, write before you know. An entry composed after the fact is worthless. Second, never edit after the outcome. You may append a later review, dated separately, but you never touch the original. The original is a fossil of your past reasoning, and its value is that it can't flatter you.

The calibration payoff. The journal is a training instrument, and what it produces is calibration, the match between your stated confidence and reality. Over dozens of entries you can ask the one question that measures forecasting skill fairly. Of all the times I said I was 70% sure, did the thing happen about 70% of the time? Say your 70%s come true 45% of the time. Then you're systematically overconfident, which is the most common and most expensive flaw an analyst has. Now it's a fact you know about, and you can correct it by widening your ranges and shrinking your position sizes. If your 90%s come true 90% of the time, you've earned the right to your convictions.

Calibration scores reasoning, and the outcome doesn't enter into it. A well-reasoned 65%-confidence thesis that happens to fail is a good decision with a bad outcome. Your P&L can't tell the difference, and neither can your memory or your ego. Your journal can (working agreement #8). You'll make this loop rigorous later, running the whole thing from checklist to journal to quarterly review to calibration as one machine over ten-plus decisions. The data collection starts here, this week, with your first real journaled thesis. Five years of that record compounding is an edge no one can take from you.

4.6 Environment design: making discipline structural

Willpower is a terrible strategy. The great investors took decades to accept that, and beginners refuse it. In the moment, you won't reliably override a bias you've read about, a falling price on a screen, a confident CNBC anchor, or a friend's hot tip. The whole point of Phase 6 is that these forces operate below deliberate reason. The durable solution is to engineer the moment so the sane choice is the default and the insane one takes effort. This is environment design, and Guy Spier's The Education of a Value Investor is its manifesto.

Spier's central realization was that he was a product of his environment. The frantic, status-soaked, information-saturated world of a New York money manager was corrupting his judgment. More self-control wasn't the answer. The highest-leverage thing he could do was change his environment so it stopped demanding so much self-control in the first place. He physically moved his operation from New York to Zurich to escape "the vortex." You may not move cities, but you can build the same structural defenses:

• Curate the information diet. Not all information helps. Most financial media is noise engineered to be addictive, priced in already, and behaviorally toxic, and it manufactures urgency where patience pays. Spier's rule of order goes deepest: consume information in order of increasing bias. Start with the primary source (the annual report, the 10-K, the numbers) and form your own view before you read a single broker note, bull thesis, or bear tweet. Once someone else's conclusion is in your head, you can't un-know it. Anchoring and confirmation will have you assembling evidence for it without noticing. The order in which information reaches you decides how much it biases you.
• Cure the checking-the-price addiction. Compulsively checking a stock price feels like research. It's a slot machine, and it feeds loss aversion, recency, and the itch to act. The price tells you what the market feels today. It tells you nothing new about the business, which is the thing you actually own. Spier's structural fix is to look at prices rarely and on a schedule, as infrequently as you can bear, and to route decisions through pre-committed rules in place of live reactions. One famous rule: once you decide something, don't reverse it for a set cooling-off period. Spier uses rules like refusing to sell a falling stock for a defined number of days, which forces the decision out of the amygdala and back to the analysis. You're putting friction between impulse and action.
• Choose your peers deliberately: the mastermind. You become the average of the people you think alongside. Spier credits an enormous share of his development to surrounding himself with better role models on purpose. Buffett and Munger were distant exemplars, and Pabrai was a real peer. He also credits a small "mastermind" group of serious, honest investors who improve his thinking and check his rationalizations. The corollary is defensive. Keep your distance from people who trade on tips, chase momentum, or turn investing into a status game, because their behavior is contagious and you won't notice the infection. In India the specific hazard is the WhatsApp-group and television-tip ecosystem. In the US it's FinTwit and the meme-stock crowd. Same disease, different accent.
• Pre-commit and remove temptation. Automate what you can. Write the kill criteria down before you're emotional. Keep the checklist physically in front of you at the pause point. Make the good behavior require no willpower and the bad behavior require breaking an explicit rule you wrote when you were sane. Ulysses tied himself to the mast before he heard the sirens. He didn't plan to resist them by force of character, and neither should you.

Temperament is who you are, and it's also what you build around yourself. A person of ordinary self-control inside a well-designed environment will out-behave a person of heroic self-control inside a casino. Design the room.

All four practices share a name in the literature, and it's worth knowing because it tells you what kind of lever you're pulling. Richard Thaler and Cass Sunstein call it choice architecture, in Nudge (2008). It means the arrangement of the context in which a choice is made. That arrangement is never neutral, and it changes what people choose without removing any option. Since there's no such thing as a neutral presentation, someone is always designing your defaults. The only question is whether it's you.

The part that carries the weight is the default. Whatever happens when a person does nothing is what most people get, because status-quo bias and the cost of deciding both push toward inaction. Retirement-plan enrolment is the standard demonstration. Switch a scheme from opt-in to opt-out and every option stays available, yet participation moves dramatically, purely because inaction now produces something different. The lesson transfers directly. Your investing defaults are the systematic investment plan that runs on salary day whether or not you feel confident, and the rebalancing that happens on a date instead of on a view. They're also the price alerts you didn't set and the app you didn't install.

So read the four practices as choice architecture. They aren't self-improvement projects. Curating the information diet changes what your attention falls on by default. The pause point puts the checklist at the spot where the decision gets made. The mastermind changes the default opinion in the room. Each one shifts the burden from your willpower in the moment to a structure you built while you were calm. Set your defaults on purpose, and the good behaviour becomes what happens when you do nothing at all.

4.7 Temperament over intellect

Now for the claim that organizes this entire phase, stated most bluntly by the person with the best track record to back it. Investing, Warren Buffett has said more than once, is not a game the 160-IQ mind wins from the 130-IQ mind. "Once you have ordinary intelligence, what you need is the temperament to control the urges that get other people into trouble." Ben Graham said the same thing sixty years earlier, and more sharply, in The Intelligent Investor (1949): "The investor's chief problem — and even his worst enemy — is likely to be himself."

Smart people aren't being falsely modest here. Nor are they saying intelligence is worthless. You need enough of it to do the analysis in Phases 1 to 5, and the course isn't easy. The claim is about the binding constraint. Above a threshold most serious people clear, extra IQ points don't improve investment results. They can even hurt, because intelligence is superb at building sophisticated rationalizations for what you already want to do emotionally. The graveyard of investing is full of brilliant people. Long-Term Capital Management had two Nobel laureates. The constraint that binds is the ability to behave well when behaving well is psychologically hardest. So what goes into that ability?

• Patience: the willingness to do nothing. Most of the time the right action is no action. No company clears the bar, so you wait, in cash, looking foolish, for months or years. Buffett's "20-punch card" asks you to imagine you get only twenty investment decisions in your whole life. Munger's line is "the big money is not in the buying and selling, but in the waiting." In both, sitting on your hands is a choice you make, and nothing to be ashamed of. Terry Smith's third rule is literally "do nothing." The market pays you to be patient and taxes you for being busy.
• Decisiveness: the willingness to act, hard, when the pitch is fat. The mirror image of patience. Sometimes the analysis is done and the asymmetry is enormous. That's usually in the middle of a panic, when the news is worst and the price is lowest. Temperament then is the ability to act aggressively against your own fear and against the crowd. Patience without decisiveness is just paralysis. The temperament you want is patient for years and then, rarely, ferocious.
• Equanimity in drawdowns: the willingness to suffer without flinching. This is the hardest and the most valuable. Own equities long enough and you'll watch a good business fall 30%, 50%, occasionally more, for reasons that have nothing to do with your thesis. If the thesis is intact, you have to be able to hold, or add, without panic. Prospect theory explains why that's so hard. Losses hurt roughly twice as much as equivalent gains feel good, so a 50% drawdown feels like an emergency demanding action, far worse than giving back some gains. The action it demands is selling, and selling is usually the wrong move. The investors who compounded didn't avoid drawdowns. They endured them without selling the bottom.
• Independence and intellectual honesty. The willingness to reach a different conclusion from the crowd, which is Marks's second-level thinking. And, harder, the willingness to face facts that hurt: to seek disconfirming evidence, to update when wrong, to say "I made a mistake" out loud. Here the ethics thread that runs through the course turns inward. Do the managers you study own their mistakes, or do they spin? You have to hold yourself to that same standard first. An investor who can't be honest with their own journal can't be trusted with a thesis.

Why the stakes are asymmetric: the drawdown arithmetic. Temperament matters more than it "should," because losses and gains are asymmetric in their arithmetic as well as in their psychology. To recover from a loss of size L, you need a gain of L / (1 − L):

A 50% loss isn't "half as bad" as a 100% loss. It demands a doubling just to get back to even. That arithmetic is the spine of Buffett's Rule No. 1, "never lose money," and of Marks's and Klarman's obsession with permanent loss over volatility. A deep hole is disproportionately hard to climb out of. So avoiding the big, permanent loss is worth more than any brilliant winner, and it's a temperament-and-process problem far more than an analytical one.

The behavioral side makes it worse. Bad temperament makes people sell into the hole, locking in the −50% when they could have endured it. Studies of real investor returns have consistently found that the average fund investor earns several percentage points a year less than the very funds they own. DALBAR's long-running QAIB analysis is the most cited of them. The shortfall comes purely from mistiming, buying after a fund has risen and selling after it has fallen. Magnitudes vary by study and period, so treat the direction as solid and the exact number as illustrative (as of mid-2026). That gap between what investments returned and what investors earned has a name, the behavior gap. It's temperament, denominated in money.

Over an investing lifetime it dwarfs whatever a great analyst earns over a merely good one.

The two worked reminders, one each side of the world. In the US dot-com collapse, Amazon's stock fell roughly 90% from its 1999 peak. The business was intact and eventually became one of the great compounders of the century. Almost no shareholder captured that return, though, because sitting through a 90%-down position for years takes a temperament that's vanishingly rare. The analysis was the easy part. In India, think of a quality compounder like Asian Paints, or HDFC Bank in its long ascent. The investors who held one through the multiple 30 to 50% drawdowns of 2008, 2013, 2020 earned the compounding. The ones with identical analysis and weaker nerves sold in the panic and bought back higher, or never bought back at all. Same information, same spreadsheets, opposite results. Temperament decided all of it.

Can temperament be built, or are you stuck with it? It's partly innate and substantially trainable, and the training is what you're assembling here. You can't will yourself calm in a crash. You can build scaffolds that produce calm behavior even from an anxious mind. The checklist removes the dumb errors, so your losses are at least earned. The journal's pre-committed kill criteria let a calm past self make the sell-or-hold decision for a panicked present self. Environment design lowers the emotional load, so there's less to withstand. And the base-rate habit and position sizing make sure no single mistake can ruin you.

That last one is what makes equanimity possible. It's easy to be calm about a loss that can't kill you, and impossible to be calm about one that can. Temperament, in the end, is downstream of process. Build the process, and the temperament has somewhere to stand.

One input to a lifetime of investing isn't covered by any checklist or measured by any calibration score. Without it, every habit above is aimed at a target that keeps moving. John Bogle told the story of Joseph Heller at a billionaire's party, and Heller's answer is the cleanest statement of the idea. He had something the host would never have, which was enough. Bogle made it the title of a 2008 book. The reason it belongs beside temperament is structural. This isn't a sermon.

Define enough as a terminal number, written down in advance: the capital at which you stop needing the portfolio to grow, expressed as the wealth that funds your required spending. Work it out from the spending, because spending is the thing you actually know. Wealth is the wrong end to start from. Take a household that needs ₹6,00,000 a year in today's money and applies a 3.5% real withdrawal rate. It requires 6,00,000 ÷ 0.035 = ₹1,71,42,857, or about ₹1.71 crore. Change the withdrawal rate and the number moves a long way: ₹2.00 crore at 3%, ₹1.50 crore at 4%, a band ₹0.50 crore wide. That's a wide band for one assumption. So the assumption goes in the same sentence as the number, and you re-check it when your spending and your horizon change. A move in your portfolio isn't a reason to.

You write it in advance because of the failure mode it prevents. Call it the goalpost problem: reaching a wealth target reliably produces a new, higher one, because the target was never anchored to a need. You've already met the mechanism, in the bias catalog. Hedonic adaptation resets the reference point to your current position, so the new level stops feeling like an achievement within months. Social comparison supplies a fresh peer group at every level of wealth, and there's always someone ahead. And the more capital you have, the more capital a given percentage gain represents, which makes the next increment feel more important. None of that is a character flaw. It's the same reference-point machinery working as described, on a variable nobody thought to fix in advance.

The consequences are concrete, and they're why this belongs next to the drawdown arithmetic. Without an enough line, three things follow. Risk never comes down, because there's no point at which the portfolio has done its job. So a person keeps running a capital-growth allocation into a life stage that can't absorb a 50% drawdown. Leverage becomes tempting late, when it is least survivable and when the marginal rupee is worth least. And the process stays switched on for good, which is how a good analyst reaches a ruinous decision through a run of individually reasonable ones. Put simply, an unbounded objective turns every one of this phase's disciplines into an engine with no governor.

Writing the number down does three things no amount of temperament does on its own. It turns an emotional question into an arithmetic one you can revisit once a year. It makes de-risking a scheduled event, where it would otherwise feel like an admission of defeat. And it gives the enterprising path an exit, which is what separates a strategy from a compulsion. Put the number in the decision journal, dated, with its withdrawal-rate assumption and its spending assumption stated. Treat any upward revision as a decision that has to be justified in writing, like any other.

4.8 The behavioral self-audit and assembling Checklist v1

You have the pieces now. The capstone puts them together, and it starts by turning the abstract bias catalog into a specific, truthful map of your weaknesses. The list has twenty biases. Only two or three of them are strongest in you, and those are the ones that will cost you money. You need to know which.

The behavioral self-audit protocol. Work the structured questionnaire in the practice set (Problem 8) and the capstone, and tell yourself the truth. It's a mirror, and lying to it wastes the exercise. For each major bias in the catalog, you rate your susceptibility on evidence. Self-flattery doesn't count. The only admissible evidence is your own past behavior, so even a near-beginner can do this from their financial life before the course. Think of the stock you bought because a friend was making money on it, the loss you held because selling felt like admitting defeat, the position you wouldn't cut because you were "waiting to get back to even." The audit produces three outputs: (1) a ranked list of your top two or three dominant biases; (2) for each, the specific failure mode it produces in your decisions; and (3) for each, a countermeasure, either a checklist item that forces the disconfirming step or an environment rule that removes the temptation. You'll see a full example worked through below.

The anatomy of Checklist v1. Your personal checklist has two layers, and you assemble both this week:

• The analytical spine: the Mental-Model & Moat Checklist (the reusable artifact finalized here). It's the shared, business-facing list every serious analyst should run. Its sections cover circle of competence; business quality and unit economics; the moat through all four lenses (five sources, Greenwald barriers, 7 Powers, CAP/ROIC-persistence); industry structure (Porter + capital cycle); management and capital allocation; valuation and margin of safety ("what must be true?"); risk, inversion, and pre-mortem; and a behavioral self-audit tab. It's organized around the five questions, so running it means answering them. You don't invent this one. It's written for you, and you'll grow it with your own items over the years.
• Your personal behavioral overlay: the mistake-derived items. The handful of lines that are yours. They come from your self-audit, which turns your dominant biases into questions that force the counter-move. From week 53 onward they also come from your own logged mistakes, run through the mistake loop. This layer starts small and real, three to six items today, and grows one scar at a time. It's short on purpose. A killer list of six behavioral vetoes you actually run beats a hundred you admire and skip.

How to run it, so it is a pause and not a ritual. At the pause point, before you initiate, add to, or exit a position, you stop and run the list as a do-confirm check. Every item is a question you answer from work you've already done. One failed item isn't fatal by itself, but it's a flag you have to resolve in writing. A cluster is different. Say you can't prove the moat. On top of that the valuation is demanding and management's incentives point the wrong way. Failures like those, across independent sections, add up to a veto. You aren't scoring the company up to a "buy" number. You're confirming you didn't skip a step that has bankrupted people before, and you're logging in the journal which flags you chose to accept, and why. Then you make the decision. The checklist never makes it for you.

So here's the full kit. A short, mistake-derived, do-confirm checklist, run at a defined pause. A decision journal written before outcomes and never edited after. A calibration scoreboard that grades reasoning. An engineered environment. And a temperament you've cultivated on purpose, standing on top of all that process. Assemble it in the capstone and you'll carry it for the rest of your investing life, improving it one scar and one journal entry at a time. More than any spreadsheet skill, it's what will still make you elite in twenty years.

Common mistakes & how experts think differently

Treating the checklist as a buy-signal scoring sheet. The beginner wants the checklist to make the decision: tick enough boxes, get a "buy." It can't do that. A checklist is a do-confirm pause that catches errors of omission. Ticking every box only confirms you didn't skip a killer step. Whether the thing is a good investment is a separate question. Experts run it looking for a reason to stop. A clean checklist and a bad thesis is still a pass.

Making it so long that it covers everything, which makes it dead. The instinct is to add every possible consideration until the list runs to a hundred items. Long checklists don't get run. They get skipped under the very pressure they were built for. Gawande's designers fought for brevity because a five-to-nine-item list that gets used beats an exhaustive one that gets admired. Experts are ruthless. Only killer items, the ones commonly skipped and catastrophic, get a line on the master list. Everything else goes in a sub-checklist or an artifact.

Copying someone else's checklist instead of bleeding your own. A downloaded checklist is a good starting spine, and the Mental-Model & Moat Checklist is one. But the items that save you are the ones that come from your own recurring failures, in your words. Pabrai and Spier built theirs from documented mistakes with real losses attached. A checklist you didn't earn won't flinch at the shapes that actually hurt you.

Stopping at the symptom in the mistake loop. "The stock fell, so avoid that sector" teaches you nothing you can carry elsewhere. Root causes like "I trusted a growth number without asking how it was funded" or "I outsourced my conviction to a tip" are behavioral, and they recur across unrelated companies. Experts push "why" until they hit something general enough to be worth a permanent line.

Editing the journal after the outcome, or not keeping one. The journal's whole value is that it's a fossil of your pre-outcome reasoning, out of reach of hindsight's flattering edits. Touch it after the fact and you've destroyed the one record that could tell a good decision from a lucky one. The most common version of this mistake is subtler. People don't keep the journal at all and run on memory instead, which is hindsight bias with no defense.

Confusing a good outcome with a good decision. This is the master error the whole apparatus exists to fix. A well-reasoned bet can lose, and a reckless one can win. Grade yourself on P&L over any short horizon and you'll reward luck and punish good process. You'll "learn" to become a worse investor, with total confidence. Experts grade process and calibration. The journal and the Brier score are what make that possible (working agreement #8).

Believing temperament is fixed, so why bother. "I'm just an anxious person" is an excuse, and it diagnoses nothing. Temperament in a crash is largely downstream of process. Pre-committed kill criteria let a calm past self decide for a panicked present one. Sizing that can't ruin you makes calm possible. An engineered environment lowers the load. Experts don't white-knuckle their emotions. They build scaffolds that get good behavior out of an ordinary nervous system.

Mistaking activity for diligence, and price-watching for research. Checking the price ten times a day, reading every hot take, trading around the position. It all feels like hard work, and it's the opposite. The price tells you the market's mood. It carries no news about your business. Experts curate a narrow information diet, look at prices rarely, and treat "do nothing" as the default correct action most of the time.

Reading others' opinions before forming your own. Once a broker's target or a bull's thesis is in your head, anchoring and confirmation organize your "independent" analysis around it without your noticing. Experts consume information in order of increasing bias, primary numbers first and opinions last, because the order of exposure sets the size of the contamination.

Worked examples

Worked Example 1: Reading your own scoreboard: calibration and Brier score

A learner has kept a decision journal for two years and now has 24 resolved predictions, each with a confidence stated in advance. Grouped by confidence bucket:

Step 1: build the calibration table. Actual hit rate = came-true ÷ count, per bucket. 90% bucket: 5/8 = 62.5%. 70% bucket: 7/10 = 70%. 50% bucket: 3/6 = 50%.

Step 2: diagnose. The 70% and 50% buckets are essentially perfectly calibrated, with actual ≈ stated. The 90% bucket is badly off. Things this person called near-certain (90%) happened only 62.5% of the time, a 27.5-point overconfidence gap concentrated at the top of the confidence scale. It's the most common and most expensive pattern there is: reasonable calibration in the middle, and dangerous overconfidence in the "sure things," which is where people size largest. The fix is to haircut high-confidence calls, treating your "90%" as 65 to 70%. Then, through Phase 9 sizing, never let a "sure thing" become a position large enough that being wrong 35% of the time can hurt you.

Step 3: one number for the record (Brier). Take four of the individual predictions, with p and outcome o: p = 0.90 (o=1), 0.90 (o=0), 0.70 (o=1), 0.50 (o=0). Squared errors: (0.90−1)² = 0.01; (0.90−0)² = 0.81; (0.70−1)² = 0.09; (0.50−0)² = 0.25. Brier = (0.01 + 0.81 + 0.09 + 0.25)/4 = 1.16/4 = 0.29. That's worse than the 0.25 you'd get by always saying 50%, and the one confident miss, the 0.81 term, did almost all of that. Feel the lesson here. A single overconfident wrong call does far more damage to your score than several humble 50/50s, and that's the mathematical case for humility at the top of the confidence scale. Track the Brier quarterly. A number that falls over the years is objective, luck-proof evidence your judgment is improving.

Worked Example 2: The mistake-to-checklist loop, worked end to end

The mistake (stylized, India). Two years ago the learner put 15% of the portfolio into a small-cap "next multibagger" a college friend was excited about. It was a company in a business the learner couldn't actually explain, and the price target came from the friend's WhatsApp group. Over eighteen months it fell 70% on an accounting-restatement disclosure. Run the four-step loop.

• Step 1: isolate the real decision. Bought a 15% position in a business I could not explain, on a price target I did not derive, in an industry outside my competence. Loss: 70% of a 15% position ≈ 10.5% of the whole portfolio.
• Step 2: root cause, not symptom. The symptom: "small-caps are risky." That's useless, since great small-caps exist. One level down: "I did not do my own work." The root is behavioral and recurring: I outsourced my conviction. I let social proof and an external price target stand in for a thesis I built and could defend. I bought a business outside my circle of competence, so I had no independent way to know the story was breaking. That root cause could sink me in a completely different sector next year, so it gets permanent lines.
• Step 3: write the pointed, answerable questions. (a) "Can I explain, in writing and in my own words, exactly how this business makes money and why it is mispriced, with a valuation I derived myself? If the source of my conviction is someone else, STOP." (b) "Is this inside my written circle of competence? If not, it is a pass regardless of the story." (c) "Did I read the primary filings and form my view before hearing anyone else's target?" That last one is Spier's ordering rule.
• Step 4: keep or discard. All three are general. They apply to every future idea, in any sector, in either market. They go on the personal behavioral overlay of Checklist v1. One 70% loss gave you three permanent lines, plus a specific behavioral weakness, outsourced conviction, flagged for the self-audit.

So the loop turned a painful, specific loss into general protection you can carry anywhere. That's the only way a mistake pays its tuition back.

Worked Example 3: Temperament in money: the drawdown arithmetic and the behavior gap

The setup. Two investors with identical analysis buy the same quality compounder: ₹10,00,000 each, 10,000 shares at ₹100. A market panic, say a 2020-style one, takes the price to ₹50. That's a −50% drawdown, and each position is now worth ₹5,00,000 on paper. The thesis is fully intact. Nothing in the business has broken. Over the next five years the price recovers and compounds to ₹250.

• Investor A (equanimity). Consults the pre-committed kill criteria in the journal and confirms nothing in the thesis has broken. The fall is sentiment, and the fundamentals haven't changed. So A holds. Final value: 10,000 × ₹250 = ₹25,00,000, a 2.5× on the original capital.
• Investor B (the behavior gap). Can't bear the −50% paper loss. That's loss aversion: it hurts about twice as much as the eventual gain will feel good. B sells at ₹50, realizes ₹5,00,000, and sits in cash, getting back in only after the stock has "proven itself" by climbing back to ₹150. At ₹150, ₹5,00,000 buys 3,333 shares. At ₹250 that's ₹8,33,333.

The cost of temperament, in rupees. A ends with ₹25,00,000 and B ends with ₹8,33,333. The gap of ₹16,66,667 is roughly two-thirds of A's final wealth. Identical analysis and opposite behavior in a drawdown created it. Had B simply stayed in cash after selling, the gap would be ₹20,00,000. The drawdown math explains it. Recovering the −50% took a +100% move, from ₹50 to ₹100. By selling, B turned a temporary 50% paper loss into a permanent realized one, and then paid up to rejoin. It's the behavior gap, DALBAR's "investors earn less than their investments," made concrete on one position. There's no spreadsheet skill B lacked that could have closed this gap. Only temperament, scaffolded by a journal with pre-written kill criteria, would have.

Worked Example 4: The behavioral self-audit, worked on one learner

A learner, "Ravi," runs the audit truthfully against his financial history from before the course. Here's the evidence he can actually point to. He checked prices compulsively and sold two good funds near the bottom of the 2020 crash. He held a losing stock for three years "to get back to even." He bought a hot IPO because everyone at work was in it. And he tends to read the bull case first and finds the bears "annoying." Rate each bias from 1 to 5 on behavioral evidence:

Top biases: (1) loss aversion, (2) confirmation, (3) herding. Failure modes and countermeasures:

• Loss aversion leads to panic-selling and riding losers. Countermeasure (environment + checklist): look at prices on a weekly schedule only, and add a hard checklist rule: "I may not sell any position on a down-day unless a pre-written kill criterion has been objectively triggered; otherwise the decision waits [cooling-off period]." Pre-commit kill criteria for every position in the journal.
• Confirmation leads to one-sided analysis. Countermeasure (checklist + Spier ordering): "Before deciding, I must write the bear case in my own hand, stronger than a bear would, and I must read primary filings before any bull/bear opinion." A failed box here is a hard stop.
• Herding leads to buying tops on social proof. Countermeasure (checklist, from the earlier mistake loop): "Is the source of my conviction a thesis I built and can defend, or someone else's excitement? If the latter, STOP." Also an environment fix: mute the tip channels.

Output: three ranked biases, three failure modes, three concrete countermeasures, two of which become checklist lines and three of which become environment rules. That's the self-audit deliverable of the capstone.

Worked Example 5: Base-rate blending in a pre-mortem: turning a hot gut into an honest confidence number (India)

You're about to file a decision-journal entry on Kaveri Retail Ltd (synthetic), an aggressive, debt-funded store-expansion story you find exciting. Your gut says "70% chance the three-year thesis works." That number is about to go in the journal, where it'll be graded for calibration and will drive position sizing. So check it two ways before it goes in. First take the outside view, which means a base rate. Then run a pre-mortem.

Step 1: the outside view (base rate). Use the base-rate habit you built with Mauboussin. Comparable aggressive, debt-funded retail rollups have historically hit their three-year growth-and-returns target only about 30% of the time (synthetic base rate). That's your anchor, the batting average of stories shaped like this one, before any of Kaveri's specifics get a chance to flatter you.

Step 2: the pre-mortem (a structured inside view). "It is three years later and this was a disaster. What happened?" You name three roughly independent failure paths and give each a probability. Same-store sales stall (25%). A refinancing or debt-rollover squeeze hits (20%). An accounting or governance shock arrives (10%). The thesis survives only if none of them fires, and that chance is the product of the complements: (1 − 0.25) × (1 − 0.20) × (1 − 0.10) = 0.75 × 0.80 × 0.90 = 0.54. So the pre-mortem's own inside view is about 54% success. That's already 16 points below the gut's 70%, because making yourself list the ways it dies is sobering by design.

Step 3: blend the two views. Kahneman's advice is to lean toward the outside view, so weight the base rate at 0.60 and the pre-mortem inside view at 0.40: 0.60 × 0.30 + 0.40 × 0.54 = 0.18 + 0.216 = 0.396, or about 40%.

Step 4: what the number does. You file 40% in the journal. Not 70%. The gut was 30 points too hot, and anchoring on the base rate and running the pre-mortem nearly halved it. That 40% is what your calibration record will grade later. Through Phase 9 it's also what sizes the position, so a 40%-confidence idea gets sized as the two-in-five bet it is. It felt like a near-certainty. It isn't one. The base rate is the leash, the pre-mortem keeps you truthful, and the blend is the number you can defend to your future self in three years.

Worked Example 6: Checklist error-reduction, in dollars: what catching omissions is worth (US)

Does a do-confirm pause actually pay? Turn the hard evidence from earlier into money. The WHO surgical trial (Haynes 2009) cut major complications from 11% to 7%, a relative reduction of (11 − 7) / 11 = 0.364, or about 36%. Now assume a checklist buys an investor the same relative protection against one recurring, avoidable, catastrophic error: the omission from the mistake loop, "bought a debt-fuelled roll-up without checking organic growth."

Step 1: the error rate, with and without the pause. A synthetic US investor makes 50 position decisions over several years. Without a do-confirm checklist, the avoidable-omission error strikes on 11% of decisions. With it, the rate falls to 7%, mirroring the surgical result. Notice that the checklist leaves the winners alone. All it does is remove a slice of the self-inflicted losers.

Step 2: put a loss on each error. When the omission fires, it costs $25,000 of permanent capital on average, a position that should have been vetoed. Expected avoidable loss without the checklist: 50 × 0.11 × $25,000 = $137,500. With the checklist: 50 × 0.07 × $25,000 = $87,500.

Step 3: the value of the laminated card. Savings = $137,500 − $87,500 = $50,000 over the 50 decisions, or $50,000 ÷ 50 = $1,000 per decision. You earn that by doing nothing more than reliably running a one-minute pause that catches an error you already "knew" to avoid.

The lesson. It's Gawande's argument in a spreadsheet. The checklist attacks ineptitude, and its payoff is the losers it deletes. It doesn't find you winners. A pause that feels beneath a skilled analyst is worth, in expectation, $1,000 a decision here. And unlike a stock pick, it's available on demand, every single time, for free.

Practice set

Do each problem fully in writing before you read the solution (working agreement #3). Problems 1 to 2 are guided, 3 to 8 are independent, and 9 to 12 are timed or synthesis.

Problem 1 (guided). Classify each as a read-do or do-confirm checklist, and state which type your master investment checklist should be and why: (a) a cake recipe; (b) an aircraft "before landing" list run by an experienced captain from memory then confirmed; (c) an IKEA assembly sheet; (d) your pre-decision investment checklist.

Solution. (a) Read-do, since you do each step as you read it. (b) Do-confirm, since it's done from experience and then confirmed at a pause point. (c) Read-do. (d) Do-confirm. You perform your full analysis in the order the company demands. Then, at the pause point before committing, you stop and confirm you didn't skip a killer step. Investing has no fixed sequence and depends on expert judgment, so a read-do "recipe" would be both wrong and unusable. The value is the confirming pause, which catches errors of omission without pretending to script the analysis or make the decision.

Problem 2 (guided). A learner writes this "lesson learned": "I lost money on an airline stock, so I'll never buy airlines again." Diagnose what is wrong with this as a checklist item, find a better root cause, and write a proper item.

Solution. It stops at the symptom, a sector label, so nothing in it transfers. Good airlines exist at the right point in the cycle, and the same underlying error will recur in other capital-intensive, cyclical, price-taking businesses. Push "why." Airlines are commodity-like, capital-heavy, high-operating-leverage, fuel/FX-exposed businesses with structurally weak pricing power. The likely root cause of the loss is buying a structurally poor, cyclical business on trailing earnings near a cycle peak, mistaking a good quarter for a good business. A proper, general checklist item: "For any cyclical or structurally low-return industry (ROIC rarely above WACC across the cycle): am I valuing on mid-cycle earnings, not the trailing peak? And do I have a specific reason this company escapes the industry's poor economics, or am I buying a bad business at a good moment?" That item now protects you across airlines, commodities, and shipping alike.

Problem 3. A learner's journal shows: of 20 predictions made at "80% confident," 11 came true. Compute the actual hit rate, state the diagnosis, and give the corrective action.

Solution. Hit rate = 11/20 = 55%, against a stated 80%, so there's a 25-point overconfidence gap. Diagnosis: this person is systematically overconfident in their high-confidence calls, which is the most common and most costly calibration error. Corrective action: haircut stated confidence, treating an instinctive "80%" as about 55 to 60%, and widen thesis ranges. Most important, reduce position sizes on high-conviction ideas (Phase 9), because overconfidence does its damage through oversizing. Keep scoring. Calibration can be trained, and it should improve as the feedback loop runs.

Problem 4. Compute the Brier score for these four resolved predictions: p = 0.80 (came true, o=1); p = 0.60 (true, o=1); p = 0.90 (did NOT happen, o=0); p = 0.50 (did not happen, o=0). Interpret against the 0.25 "always say 50%" benchmark.

Solution. Squared errors: (0.80−1)² = 0.04; (0.60−1)² = 0.16; (0.90−0)² = 0.81; (0.50−0)² = 0.25. Sum = 1.26, so Brier = 1.26/4 = 0.315. That's worse than the 0.25 know-nothing benchmark, and nearly all of the reason is the 0.90 confident miss, which contributes 0.81 by itself. Confident wrong calls are punished quadratically and can sink an otherwise reasonable record. That's the mathematical argument for humility right where you feel most certain.

Problem 5. (a) An investor suffers a −60% drawdown. What percentage gain is required to break even? (b) On ₹8,00,000 invested, the position falls 60% and the investor panic-sells to cash; the business was fine and the price later triples from its low. Compute the final wealth of the panic-seller versus a holder, and the behavior-gap cost.

Solution. (a) Required gain = L/(1−L) = 0.60/0.40 = 150%. (b) ₹8,00,000 falls 60% to ₹3,20,000 at the low. The seller locks in ₹3,20,000. A holder rides the low tripling: ₹3,20,000 × 3 = ₹9,60,000. The panic-seller, sitting in cash, keeps ₹3,20,000. Behavior gap = ₹9,60,000 − ₹3,20,000 = ₹6,40,000. That's four-fifths of the original capital, destroyed by selling the bottom. The analysis was fine. Even if the seller had bought back in after a 50% recovery from the low, they'd still end far behind. The exercise is deliberately punishing, to make the point.

Problem 6. Here is a learner's daily routine. Identify three environment-design flaws and give a fix for each: "I check my portfolio value on my phone about 15 times a day. Every morning I read three brokerage 'top picks' emails and watch business TV while I work. I'm in four stock-tip WhatsApp groups. When a stock I own drops 10% in a day I usually sell to 'cut the loss' before it gets worse."

Solution. (1) Compulsive price-checking (15×/day) feeds loss aversion and the itch to act. Fix: look on a fixed weekly schedule, and take the app off the home screen. (2) Consuming brokerage picks and TV first anchors and confirms before any independent view forms. Fix: Spier's ordering. Read primary filings and form your own view before any opinion, and cut most media as behaviorally toxic noise. (3) Tip-group membership + reflexive 10%-drop selling imports herding and turns drawdowns into panic sales with no reference to the thesis. Fix: mute or leave the groups, and pre-commit a cooling-off rule ("no sell on a down-day unless a written kill criterion triggered"). Notice that none of these fixes needs more willpower. They redesign the room so the sane behavior is the default.

Problem 7. Critique this decision-journal entry for completeness: "Buying HypotheticalCo: great company, wonderful brand, everyone loves the product, and the chart looks strong. Going in big." What is missing?

Solution. Almost everything that makes a journal useful. Missing: (1) A real thesis. That means how it makes money, why it's a good business (ROIC vs WACC), and why it's mispriced. "Great brand" isn't a thesis. (2) The disconfirming evidence or bear case, in the writer's own hand. (3) A base rate: of companies with this profile and valuation, how many delivered? (4) A numeric confidence level. (5) Explicit "what would change my mind" kill criteria. (6) A sizing rationale beyond "big," weighing conviction against downside. The entry also shows two red behavioral flags, social proof ("everyone loves it") and price-chasing ("the chart looks strong"), with no counterweight. As written, this entry cannot support calibration, cannot defend against a future panic, and cannot distinguish a good decision from a lucky one. It's a mood, not a journal entry.

Problem 8 (the self-audit mini). A learner reports these past behaviors: sold every stock the moment it showed a small gain but held losers indefinitely; was sure each of their last five stock picks "couldn't miss" and sized each at 20%+; never reads bear cases because "why be negative." Name their top two dominant biases, the failure mode of each, and one countermeasure each.

Solution. (1) Disposition effect / loss aversion: selling winners early to "lock in" a gain while holding losers to avoid realizing a loss. Failure mode: cutting compounders short and riding broken theses down. Countermeasure: a checklist rule that decouples selling from price direction ("sell only on a triggered kill criterion or a better opportunity, never because it's up or down"). (2) Overconfidence, shown by "couldn't miss" repeated, plus 20%+ sizing. Failure mode: oversizing convictions that a calibration check would reveal are 60/40s, so one miss does outsized damage. Countermeasure: haircut confidence, cap single-position size, and start logging confidence numbers to build the calibration record. Confirmation bias is a strong third ("never reads bear cases"), and the mandatory written bear case counters it.

Problem 9 (timed, killer-item selection, 8 min). From these candidate items, choose the ~6 that belong on a master do-confirm checklist and justify cutting the rest: (a) Is ROIC durably above the cost of capital? (b) Did I confirm the CEO's undergraduate university? (c) Is this inside my written circle of competence? (d) Have I run the quality-of-earnings screen? (e) What is the office's carpet colour? (f) For any lender, is there a short-bucket ALM mismatch / CP dependence? (g) Have I written the bear case and the kill criteria? (h) Is the stock up today? (i) What must be true for today's price to make sense (reverse DCF), and is that plausible? (j) Did I check today's Nifty level? (k) Is management's incentive/ownership aligned, and is capital allocation sane? (l) Is the position sized so that being wrong cannot ruin me?

Solution. Keep (killer items, commonly skipped AND catastrophic, general across companies): (a) quality/ROIC>WACC; (c) circle of competence; (d) QoE screen; (g) written bear case + kill criteria; (i) reverse-DCF "what's priced in"; (k) management incentives/capital allocation; (l) survival-sizing. Item (f) is a superb item, but it's sector-conditional. It belongs on the lender sub-checklist or sector cheat-sheet and fires only for financials. It stays off the universal master list. A defensible alternative answer keeps it if your circle is lender-heavy. Cut: (b) trivial and non-predictive; (e) noise; (h) and (j) price/index-watching, which only tells you the mood. None of these is commonly catastrophic when skipped, and none is general. Adding them is how a list becomes long, ignored, and dead. A line belongs only if skipping it has bankrupted someone and it applies broadly.

Problem 10 (timed scenario, 8 min). You own a position, thesis intact, now down 45%. (a) What gain is needed to break even? (b) Walk through, in order, the steps your process prescribes before you are allowed to sell. (c) Contrast the case where a kill criterion has triggered.

Solution. (a) 0.45/0.55 = 81.8%. (b) Your process decides here, and your feelings don't get a vote. (1) Open the original journal entry and read the pre-committed kill criteria you wrote when you were calm. (2) Check each one against current facts. Has anything structural in the thesis actually broken, or is this a price or sentiment move? (3) If nothing has broken, the drawdown isn't a sell signal. If anything it's a lower price for the same business. So the sound action is to hold, or to add if sizing and conviction warrant it, and to log the decision and your reasoning instead of reacting. (4) Enforce any cooling-off rule, so the decision isn't made on a down-day while loss aversion has hold of you. (c) Now suppose a kill criterion has objectively triggered: the metric that had to hold didn't, and the thesis is broken. Then you sell, because the reason you owned it is gone. The price falling has nothing to do with it. The 81.8% recovery math is now irrelevant, since holding a broken thesis to "get back to even" is the anchoring and sunk-cost trap the kill criteria exist to prevent. The price move never decides. The pre-written thesis test does.

Problem 11 (timed loop, 10 min). Run the four-step mistake-to-checklist loop on this documented pattern: an investor repeatedly bought serial acquirers ("roll-ups") on their reported adjusted EPS, and was burned when organic growth turned out to be near zero and debt ballooned.

Solution. (1) Isolate: bought roll-ups on adjusted EPS without testing whether growth was organic or acquired, and without checking the debt funding it. (2) Root cause: trusting a management-defined "adjusted" earnings narrative from an acquisition-dependent model, without reconciling it to GAAP earnings, free cash flow, and organic growth. That's a general, recurring analytical/behavioral failure, the habit of accepting the preferred metric of the promoter or management. (3) Item: "For any serial acquirer: what is organic growth stripped of acquisitions? Do GAAP earnings and FCF corroborate the adjusted story? Is this a compounding machine or an acquisition treadmill that stalls the moment deals or price hikes stop? And what is net debt / EBITDA and the maturity profile?" (4) Keep: it's general across every roll-up in any market (Valeant in the US, aggressive acquirers everywhere), so it's a permanent master-list line. It mirrors the Valeant instance from earlier, except that this time you built the item yourself.

Problem 12 (synthesis, 10 min). Two journaled decisions resolve. Decision X: a carefully reasoned 60%-confidence thesis, bear case written, kill criteria set, sized for the downside; the outcome was a loss. Decision Y: a stock bought on a tip with no thesis, oversized, no kill criteria; it tripled. Your friend says "clearly Y was the smart move." Rebut using the process-vs-outcome distinction, and state which decision you should learn from and how.

Solution. Y was a bad decision with a good outcome (luck), and X was a good decision with a bad outcome (variance). Grading on P&L, the way the friend does, rewards Y and punishes X. Repeat that and it trains you to drop your process and gamble, because you'll "learn" that reckless bets pay. The fair scoreboard is the journal and calibration, which grade reasoning. X's process was sound. A 60% bet is supposed to lose 40% of the time, and it was sized so the loss didn't hurt. So you keep the process and log it as a well-decided miss. Y's process was indefensible: no thesis, oversized, no kill criteria. The win is a warning, and it validates nothing. The correct lesson from Y is "I got lucky doing something that will eventually ruin me; stop." What you learn is to reinforce X's process and kill Y's. Duke's Thinking in Bets calls the friend's error "resulting" (working agreement #8).

Problem 13 (guided). You're about to file a journal entry on Deccan Turnaround Ltd (synthetic), an Indian restructuring story your model rates at "60% likely to work." Test the number. (a) The base rate for turnarounds of this profile is 20% (synthetic, the outside view). (b) Your pre-mortem names two roughly independent failure paths: the restructuring stalls (30%) and a liquidity crunch (20%). Compute the pre-mortem's implied success probability. (c) Blend the base rate and the pre-mortem inside view with a 0.60 weight on the base rate. What confidence do you file, and what was wrong with the gut 60%?

Solution. (b) Survival = the product of the complements: (1 − 0.30) × (1 − 0.20) = 0.70 × 0.80 = 0.56 → 56%. (c) Blend = 0.60 × 0.20 + 0.40 × 0.56 = 0.12 + 0.224 = 0.344 → ~34%. You file 34%. Not 60%. The gut anchored on the vivid inside story and ignored the brutal base rate of turnarounds, most of which fail. Blending the base rate with the pre-mortem pulled the number down about 26 points to something you can defend, and it'll size the position as the one-in-three shot it is. The weights are a judgment call. Leaning toward the outside view, as Kahneman advises, is the sensible default.

Problem 14 (guided). A learner's resolved journal, grouped by stated confidence. The 80% bucket has 15 predictions, 9 true. The 60% bucket has 10, 7 true. The 50% bucket has 8, 4 true. (a) Compute each bucket's actual hit rate and diagnose it. (b) Separately, compute the Brier score for these three individual forecasts: p = 0.8 (came true, o=1); p = 0.6 (true, o=1); p = 0.5 (did not happen, o=0). Compare to the 0.25 benchmark.

Solution. (a) 80% bucket: 9/15 = 60%, overconfident by 20 points, the classic top-of-scale error. 60% bucket: 7/10 = 70%, mildly underconfident, since actual exceeds stated. 50% bucket: 4/8 = 50%, well calibrated. Verdict: haircut the high-confidence calls, and leave the mid-range alone. (b) Squared errors: (0.8 − 1)² = 0.04; (0.6 − 1)² = 0.16; (0.5 − 0)² = 0.25; sum = 0.45; Brier = 0.45 / 3 = 0.15. That beats the 0.25 know-nothing benchmark, because no single confident-and-wrong call dragged it down. The trap that sinks a Brier is a high p with o=0, which can be worth up to 0.81 on its own. Both diagnoses point the same way. Save high confidence for when you've earned it.

Problem 15. A synthetic US investor makes 40 position decisions over several years. An avoidable-omission error, skipping the quality-of-earnings screen, strikes 10% of decisions without a do-confirm checklist and 4% with it. Each such error costs $30,000 of permanent capital. (a) Expected avoidable loss without the checklist? (b) With it? (c) The savings, and the per-decision value of running the pause.

Solution. (a) 40 × 0.10 × $30,000 = $120,000. (b) 40 × 0.04 × $30,000 = $48,000. (c) Savings = $120,000 − $48,000 = $72,000, or $72,000 ÷ 40 = $1,800 per decision. The checklist pays for itself entirely by deleting a slice of losers. It never touched the winners. The trap is dismissing a one-minute pause as beneath you, when in expectation it's worth $1,800 every time you run it, and it's free.

Problem 16. Two resolved decisions, graded on process using their pre-decision expected values (figures are the position's dollar P&L). Decision X: a well-researched, downside-sized bet: 60% chance of +$50,000, 40% chance of −$20,000; the outcome was the −$20,000 loss. Decision Y: an oversized tip with no thesis: 25% chance of +$40,000, 75% chance of −$30,000; the outcome was the +$40,000 win. (a) Compute each decision's expected value. (b) Which was the good decision, and which process do you keep? (c) Name the error a friend makes when he says "Y was obviously the smart move."

Solution. (a) EV(X) = 0.60 × 50,000 − 0.40 × 20,000 = 30,000 − 8,000 = +$22,000. EV(Y) = 0.25 × 40,000 − 0.75 × 30,000 = 10,000 − 22,500 = −$12,500. (b) X was the good decision, a positive-EV, well-sized bet that happened to land in its 40% loss branch, so you keep X's process. Y was a negative-EV gamble that happened to land in its 25% win branch, so kill Y's process. The win is a warning, and it proves nothing. (c) The friend is "resulting" (Duke), grading the decision by its single outcome. Over many repetitions X's process compounds and Y's ruins you. The journal grades your reasoning, and the one roll of the dice doesn't count.

Problem 17. Two analysts' four resolved forecasts. Analyst A: p = 0.9 (o=1), 0.7 (o=1), 0.6 (o=0), 0.8 (o=1). Analyst B: p = 0.9, 0.9, 0.9, 0.9 with outcomes 1, 1, 0, 0. (a) Compute each Brier score. (b) Who is better calibrated, and what behavioral pattern does B display? (c) Does either beat the 0.25 benchmark?

Solution. (a) Brier(A) = [(0.9 − 1)² + (0.7 − 1)² + (0.6 − 0)² + (0.8 − 1)²] / 4 = [0.01 + 0.09 + 0.36 + 0.04] / 4 = 0.50 / 4 = 0.125. Brier(B) = [(0.9 − 1)² + (0.9 − 1)² + (0.9 − 0)² + (0.9 − 0)²] / 4 = [0.01 + 0.01 + 0.81 + 0.81] / 4 = 1.64 / 4 = 0.41. (b) A is far better calibrated (0.125 vs 0.41). B is a serial overconfident guesser who stamps "90%" on everything, and the two confident misses, 0.81 each, wreck the score. (c) A beats 0.25 comfortably. B, at 0.41, is worse than know-nothing 50/50 guessing. The quadratic penalty on confident wrong calls is the trap, and B walked straight into it.

Problem 18 (timed/stretch, 12 min). Capstone synthesis on Nilgiri Compounders Ltd (synthetic), an Indian quality-growth name. (i) The base rate for names of this profile hitting their thesis is 25%. Your pre-mortem has two independent failure paths, a demand shock (20%) and a governance flag (25%), so compute the pre-mortem survival and then the confidence you file, blending 0.60 on the base rate. (ii) You buy anyway at ₹100/share, and a panic later takes it −55%. What gain from the low is needed to break even? (iii) You invested ₹5,00,000. A holder rides the low up 3×, but a panic-seller locks in the low and sits in cash. Compute both final values and the behavior-gap cost. (iv) Three years on, the thesis breaks and you lose money on a well-sized position. Was filing it a good decision?

Solution. (i) Survival = (1 − 0.20) × (1 − 0.25) = 0.80 × 0.75 = 0.60; blended confidence = 0.60 × 0.25 + 0.40 × 0.60 = 0.15 + 0.24 = 0.39 → file ~39%. (ii) Gain to recover = L/(1 − L) = 0.55 / 0.45 = 1.222 → 122.2%. (iii) At the low the ₹5,00,000 is worth 5,00,000 × (1 − 0.55) = ₹2,25,000; the holder rides 3× → 2,25,000 × 3 = ₹6,75,000; the panic-seller keeps ₹2,25,000 in cash; behavior gap = 6,75,000 − 2,25,000 = ₹4,50,000. (iv) Yes: a good decision with a bad outcome. A 39%-confidence bet is expected to lose about 61% of the time. If it was sized for that downside and the kill criteria were pre-committed and honored, the process was sound, so you keep it. Process beats outcome. Scoring: 4/4 in time = fluent; ≤2 = redo Worked Examples 5–6 and the journal and calibration sections before the mastery check.

Applied mini-project: Phase 6 Capstone: Checklist v1 + the Mental-Model & Moat Checklist end to end (the M7 milestone)

This is the capstone of the entire first year, and it's the M7 milestone. It has four parts and produces two durable artifacts you'll carry forward: your personal investment Checklist v1, and one company file run through the full Mental-Model & Moat Checklist with a filed decision-journal entry.

Choose your company. Pick one company you haven't studied deeply yet, from the course's starter bank or your own watchlist. A simple Indian company is ideal for a first full run, say Asian Paints, Pidilite, DMart, or Titan. A US name like Costco or Copart works too. You need 5 to 10 years of financials, the latest annual report / 10-K, and one earnings-call transcript. For India, get the financials from screener.in or the annual reports. For the US, use stockanalysis.com/TIKR or SEC EDGAR 10-Ks.

Part A: The behavioral self-audit (write it). Work the self-audit protocol and Worked Example 4's method on yourself, using evidence from your own financial behavior before the course. Produce your ranked top two or three biases, the failure mode of each, and a specific countermeasure (checklist item or environment rule) for each. This document is half of the M7 milestone.

Part B: Assemble Checklist v1 (build it). Combine two layers into one short, usable document. (1) The Mental-Model & Moat Checklist artifact, which is the analytical spine. Don't rewrite it. Adopt it. (2) Your personal behavioral overlay of 3 to 6 mistake-derived or self-audit items from Part A and the worked examples. Keep the overlay lethally short. State the trigger: run it before initiating/adding/selling. And confirm it's a do-confirm pause. It mustn't turn into a scoring sheet.

Part C: Run the full Mental-Model & Moat Checklist on your company (do it). Answer every question in every section in writing, with the "how to judge" cue applied and a number or citation behind each answer. Adjectives don't count. Where a section flags, log the flag and your resolution. Finish with the checklist's verdict mapped to the five questions.

Part D: File the decision-journal entry and the pre-mortem (commit it). Write the entry before you look at any price target or broker view. It needs the thesis (why a good business, why mispriced), the bear case in your own hand, the base rate, your numeric confidence, "what would change my mind" (kill criteria), and a sizing rationale. Then the pre-mortem: "It is three years later and this was a disaster. What happened?" File it, and don't edit it afterward. It's the first entry in the calibration record that gets graded later in the course.

Scoring rubric. 0 = absent/wrong · 1 = present but shallow · 2 = rigorous and specific.

Pass = ≥12/16 (75% of available 16 points), with the decision-journal criterion at 2. File both artifacts in your knowledge system. Together with the schools exams, this completes Phase 6.

Case Lab: run case CL-6.2 in the app's Cases tab for this module: a company that clears every analytical box yet trips a behavioral veto, taken end to end through the do-confirm checklist, a base-rate-blended pre-mortem, and a decision-journal entry with pre-committed kill criteria.

Reading & resources

Core (do these).

• Atul Gawande, The Checklist Manifesto: How to Get Things Right (2009). The whole book is short; the aviation (ch. on the Model 299), surgery/WHO, and construction chapters, and the read-do vs do-confirm and "killer items" discussion, are the spine of the opening sections on checklists. [Paid] [Beginner]
• Guy Spier, The Education of a Value Investor (2014). The environment-design bible: the move to Zurich, the checklist, and his explicit rules of order and behavior. [Paid] [Beginner]
• Annie Duke, Thinking in Bets (2018). Process-vs-outcome ("resulting"), decision journals, and calibration made practical; pairs with the journal and calibration sections. [Paid] [Beginner]

Deepening.

• Mohnish Pabrai: interviews and lectures on building an investment checklist from studied mistakes (widely available free on YouTube; e.g., his Google/university talks) and The Dhandho Investor (2007) for the mindset behind it. [Free/Paid] [Intermediate]
• Philip Tetlock & Dan Gardner, Superforecasting (2015). The science of calibration and why it is trainable; the empirical backbone of the calibration section. [Paid] [Intermediate]
• Carl Richards, The Behavior Gap (2012). The plain-language treatment of why investors earn less than their investments. [Paid] [Beginner]
• Charlie Munger, "The Psychology of Human Misjudgment," in Poor Charlie's Almanack. The temperament-and-bias source text; his "no wise pilot fails to use his checklist." [Paid] [Advanced]
• Daniel Kahneman, Thinking, Fast and Slow. Loss aversion (~2×), hindsight bias, overconfidence; the M6.04 anchor, revisited here as the why behind temperament. [Paid] [Advanced]

Free / durable.

• Farnam Street (fs.blog): the decision-journal template and essays on checklists, mental models, and calibration. [Free] [Beginner]
• DALBAR "QAIB" summaries and Morningstar "Mind the Gap" annual studies: the behavior-gap evidence; read a recent summary and note the magnitude debate (as of mid-2026, check the latest figures). [Free] [Intermediate]
• Warren Buffett: shareholder letters and the 1984 "Superinvestors of Graham-and-Doddsville" essay; and Berkshire annual-meeting Q&A on temperament (many transcribed free). [Free] [Beginner]
• Sanjay Bakshi: Behavioural Finance & Business Valuation (BFBV) lecture notes and blog (Fundoo Professor), the India-specific treatment of temperament, checklists, and biases. [Free] [Intermediate]
• Gautam Baid, The Joys of Compounding. The Indian synthesis of philosophy, temperament, and the personal knowledge system; excellent for this phase. [Paid] [Beginner]
• Going deeper: Gary Klein, Performing a Project Premortem (Harvard Business Review, 2007) [Paid] [Intermediate]. The origin of the pre-mortem: imagining a decision has already failed ("prospective hindsight") surfaces risks a normal review misses; the source behind the capstone's pre-mortem and the journal's pre-committed kill criteria (the journal section and Worked Example 5). Klein's talks and free summaries circulate widely if the HBR piece is paywalled.
• Going deeper: Michael Mauboussin, The Base Rate Book (Morgan Stanley / Counterpoint Global, free PDF) [Free] [Intermediate]. The empirical base rates for sales growth, margins, and returns that anchor the outside view you blend against your inside-view model when setting a journal confidence (the journal section; the 30% base rate in Worked Example 5 is a stand-in for exactly this data).

Program companions. The Mental-Model & Moat Checklist artifact (finalized here); the red-flags / quality-of-earnings checklist (Phase 2) and the valuation workflow (Phase 3), which the checklist points to; M6.03–M6.04 (the decision science and bias catalog behind the self-audit); M9.04 (where the loop from checklist to journal to calibration becomes the certified process).

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. That's how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds to the hand method and never replaces it. (R1/R10)

AI-Augment this skill

{
  "skill": "The mistake-to-checklist loop, the decision journal as the core calibration instrument, reading your own scoreboard (calibration table, Brier score), base-rate + pre-mortem blending into a journal confidence, environment design, temperament, and the Phase-6 capstone synthesis",
  "use": "Two narrow jobs, and one absolute ban. (1) CHECKLIST DRAFTING: once YOU have diagnosed the root cause of a real, specific past mistake (§4.3's loop), ask it for 3-5 candidate PHRASINGS of the resulting checklist item, so you can pick the clearest one-line yes/no question rather than keep your first draft — the diagnosis, and the choice of which candidate to keep, stay yours. (2) JOURNAL PIPELINE: ask it to help structure raw journal entries (date, thesis, stated confidence, resolution, outcome) into the clean table a calibration score is computed from — a data-engineering job, never a grading one. THE BAN: never ask it to diagnose why a past decision went wrong, to grade whether a decision was good or bad, or to supply the confidence number for a live journal entry — those are exactly the unaided judgments this module's mastery gate exists to build, and outsourcing them at the moment you are supposed to be building the skill defeats the entire capstone.",
  "tools": ["Chat assistants — Claude, ChatGPT for checklist-phrasing candidates and journal-table structuring", "A spreadsheet or notebook — where the calibration table, Brier score, and its decomposition are actually computed", "A personal decision-journal spreadsheet or app (even a plain dated text file) — the raw data none of this works without"],
  "prompt": "I have diagnosed this root cause from a real past mistake: [state the specific, diagnosed cause]. Draft 4 candidate one-line checklist items that would have caught it, each phrased as a yes/no question answerable in under a minute. Do not diagnose or reinterpret the cause -- I have already done that part.",
  "verify": "Recompute your own calibration table and Brier score by hand from a handful of resolved journal entries before trusting a spreadsheet formula on the full history; test every AI-drafted checklist phrasing against the actual mistake that produced it -- if it would not have caught that specific mistake, rewrite it, don't keep it for being well-phrased.",
  "diy": "Without a tool in the room: turn a specific, diagnosed mistake into a checklist item; build a calibration table and compute a Brier score from a set of resolved predictions by hand; blend a base rate and a pre-mortem into one honest confidence number; and say, from your own resolved journal, where on the confidence scale your overconfidence concentrates."
}

Modern Data Analysis

By hand first. The first worked example builds a calibration table and a 4-item Brier score once, from a given 24-prediction summary. A real journal keeps going. It's a growing table you re-score every quarter as entries resolve.

Today's workflow. Log every journal entry as a row (date, thesis, stated confidence, resolution date, outcome 0/1). As entries resolve, recompute the calibration table and Brier score automatically. Then go one step past the worked example's single number and decompose the Brier score into two parts. Reliability is how far your stated confidences are from your actual hit rates, and you want it low. Resolution is how much your confidences discriminate outcomes at all, and you want it high. Split it this way because a low Brier score on its own doesn't tell you which of the two to fix.

Tools & sources (IN + US). The journal itself doesn't care what tool you use. A spreadsheet works, so does a notebook, even a dated text file. The country-specific sourcing happens one column upstream, in the base rate that the fifth worked example feeds into the journal. That means screener.in reference-class histories for India, and Mauboussin's Base Rate Book or SEC EDGAR companyfacts for the US. They're the same sources the decision-science material builds out in full. Here the base rate is assumed to be in hand already, and you just log it.

# recomputed in-session — R3
def murphy_decomposition(buckets):
    """buckets: list of (stated_p, n, hits). Returns (REL, RES, UNC, BS) --
    Murphy (1973): Brier score = Reliability - Resolution + Uncertainty."""
    N = sum(n for _, n, _ in buckets)
    obar = sum(h for _, _, h in buckets) / N          # overall base rate
    REL = sum((n / N) * (p - h / n) ** 2 for p, n, h in buckets)
    RES = sum((n / N) * ((h / n) - obar) ** 2 for p, n, h in buckets)
    UNC = obar * (1 - obar)
    return REL, RES, UNC, REL - RES + UNC

# WE1's own 24 predictions, grouped exactly as its table -- the verification anchor
buckets = [(0.90, 8, 5), (0.70, 10, 7), (0.50, 6, 3)]   # (stated p, n, hits)
REL, RES, UNC, BS = murphy_decomposition(buckets)
print(f"REL={REL:.5f}  RES={RES:.5f}  UNC={UNC:.5f}  BS={BS:.5f}")

Verify. murphy_decomposition(buckets) on the first worked example's 24-prediction table gives REL=0.02521, RES=0.00625, UNC=0.23438, and BS=REL−RES+UNC=0.25333. That's cross-checked against a direct Brier computed on the same grouped assumption, where every prediction in a bucket is scored at that bucket's stated probability, and it returns the identical 0.25333.

That's a different number from the 0.29 in Step 3 of the worked example, and it's meant to be. Step 3 hand-scores four individually-named predictions (one from each bucket, chosen to make the arithmetic tractable by hand), while the decomposition scores all 24 predictions grouped by bucket. They're two valid, different-sized samples from the same journal, so there's no disagreement to resolve. The decomposition also adds a diagnosis the worked example doesn't reach. This forecaster's full-sample Brier (0.25333) barely beats the "always predict the base rate" benchmark (UNC=0.23438), by an edge of only 0.01896, or 8.1% of UNC, because RES is small. The stated confidences (90/70/50) claim more discrimination than the actual bucket hit rates (62.5/70/50) deliver. The 90% bucket carries almost all the miscalibration, as Step 2 already found by eye, and now there's a number on how much the 90% bucket costs and how little the 70/50 buckets do.

Quantitative lens

Murphy's decomposition, in full. The function above turns the worked example's "eyeball the gap" diagnosis into an accounting identity: Brier = Reliability − Resolution + Uncertainty. Reliability penalizes stated confidences that don't match actual hit rates, which was the 90%-bucket problem. Resolution rewards confidences that discriminate, spreading hit rates apart from the overall base rate, regardless of whether they're well-calibrated. Uncertainty is fixed by the base rate alone, and it's what "always guess 62.5%" would score. On the worked example's 24 predictions, the forecaster's edge over that know-nothing baseline is almost entirely absent (RES=0.00625 of UNC=0.23438). The predictions aren't badly wrong. The trouble is that the confidence labels (90/70/50) don't spread the outcomes as far apart as they claim to. A forecaster who said "70%" for everything that actually resolved true 62.5 to 70% of the time would score worse on reliability. In Brier terms, though, that forecaster is barely distinguishable from one sincerely using three different numbers that happen to cluster. So as the journal grows, track resolution alongside the headline Brier.

Re-deriving the toolkit's calibration power calculation. The Analyst's Toolkit & Knowledge System, back at the start of the course, states that testing a claimed 80%-calibrated confidence bucket against a true 55% rate needs 10 entries for 49.6% power, 20 for 74.8%, and 24 for the conventional 80% power target. Detecting a claimed-80%-really-70% ten-point miss needs 137. Re-deriving with the exact one-sided binomial test (crit_k = the largest rejection count with P(Binom(n, 0.80) ≤ k) ≤ 0.05, power = P(Binom(n, p_true) ≤ crit_k)) reproduces 49.6% at 10 entries and 74.8% at 20 exactly. But an exact-binomial search for the smallest n reaching 80% power gives 21 against the 55%-truth case, where the toolkit says 24, and 116 against the 70%-truth case, where it says 137. Both are several entries lower than the toolkit's figures.

Neither figure is a mistake. Two legitimate methods answer "how many entries" slightly differently. The toolkit's 24 and 137 come out almost exactly (23.14 rounds up to 24, 136.81 to 137) from the standard normal-approximation, two-sided planning formula, n = [z_{α/2}√(p₀(1−p₀)) + z_β√(p₁(1−p₁))]² / (p₀−p₁)². That's the closed-form shortcut people use because the exact binomial test has no closed-form inverse for "smallest n." The exact test is the more precise instrument, and it needs a few fewer entries (21 and 116) than the planning formula's conservative estimate (24 and 137). The gap is 3 entries at the small end and grows to 21 entries by the 137 case, because the normal approximation's conservatism scales with n. If you want a safety margin baked in, journal for the planning formula's numbers. If you want to know how much power a specific n you already have actually delivers, use the exact test.

The limits, plainly. Both the decomposition and the power re-derivation inherit every caveat in the toolkit's own "Honest limits" paragraph. Independence across predictions and a fixed true accuracy are both approximations for a learner whose judgment is improving over the very years the journal spans. That's why a single pooled test is a diagnostic snapshot and can't be a final verdict. The 21-versus-24 and 116-versus-137 gap is a real methodological finding. It doesn't mean either calculation is wrong. Matching the exact test's own two cross-checked cases (n=10, n=20) against the toolkit's cited figures is what shows both computations are sound, before you compare the different follow-on questions they answer.

Where this goes next: galaxy cross-links

• G1 · QM1.01 (Probability, Distributions, Sampling & Estimation): the sampling-distribution and confidence-interval machinery behind every power figure above.
• G1 · QM1.02 (Hypothesis Testing & Regression): the general theory of test power and the exact-vs-approximate distinction applied above to one specific binomial case.
• G2 · QD2.04 (Reproducible Research & Deployment): turning the calibration/decomposition functions above from a one-off script into a versioned pipeline that re-scores itself as the journal grows, which is the upkeep a real multi-year decision journal needs.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "Why the best investors use a checklist and a diary, and why being calm beats being clever." Without jargon, get three things across. First, even brilliant experts fail at the simple, repeatable stuff (the surgeon who forgets a step, the pilot who forgets a lock), and a short checklist fixes that by making you do what you already know. Second, writing down what you believe before you find out what happens is the only way to learn honestly, because memory lies to make you feel smart. Third, when the price of something you own falls by half, what decides whether you get rich or poor is whether you can stay calm and stick to your plan. How clever you are doesn't decide it. If your page needs the words "calibration," "do-confirm," or "loss aversion," you don't own it yet.

Journal prompt. Three entries, and keep them, because they're the first page of the record you'll be graded on later. (1) Your self-audit, honestly. Of the biases in the catalog, which two showed up most in your actual past financial behavior? Name the specific incident, not the tendency. What's the one checklist item and the one environment rule you're adopting this week because of each? (2) Your first real kill criteria. For the company you ran through the Mental-Model & Moat Checklist in the capstone, write the two or three specific, observable events that would tell you the thesis is broken. Write them before you know any outcome. In three years you'll grade your past self by whether you honored them. (3) The gap you're most afraid of. In a single sentence, write the difference between the investor you believe you are and the one your temperament would actually be in a 50% drawdown. Add the one structural change (a rule, an environment fix) most likely to close it. You'll test this for real in the Phase 9 simulator. (4) Your enough number, dated. Write down the annual spending, in today's money, that your portfolio has to fund, and the real withdrawal rate you're assuming. Divide one by the other and write the resulting capital figure down with today's date beside it, the same way the section on enough does it. Then answer two questions in a sentence each. What changes about your allocation and your risk-taking on the day you reach it? And if you later find yourself revising the number upward, what evidence would make that a legitimate change in your needs, as opposed to the goalpost problem wearing a justification?

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