The Lifelong Practice + Final Audit

Phase 10 · Week 80 · ~20 hours · THE M11 MILESTONE (CORE COMPETENCE: ALL ELEVEN AT BAR) · Mastery gate ≥85%

Where it sits. An analyst who stops reading companies decays exactly like a language you stop speaking. This is the last module of the program. Everything before it built a capability; this one builds the engine that keeps the capability alive and compounding for the decades after week 80, and then runs the final honest audit that certifies core competence. M10.01 and M10.02 proved you can run the whole system unaided at institutional grade on a US and an Indian company you had never seen; M10.03 taught you to communicate the result Minto-style. Those closed the program. They did not close the practice, because there is no such thing as closing the practice. So this module does two jobs. First, it stands up the lifelong system: a public track record that makes you honest, an annual practice cadence (roughly fifty teardowns, eight-to-ten deep dives, and four quarterly review letters a year), a high-signal information diet that keeps you current without drowning, a deliberate program for widening your circle of competence, the journal-to-checklist loop that turns every mistake into a permanent upgrade, and, if you want it, a clear-eyed map of the professional paths and the ethics of managing other people's money. Second, it runs the final competency audit: a structured, honest self-assessment against all eleven competencies of the master map's Section 1, competency by competency, bar by bar, evidence by evidence, plus the ritual that gives the whole eighteen months its emotional close. That ritual is breaking the seal on the naive one-pagers you wrote in week 3 (M0.05) and re-grading them with the eyes you now have. Passing the audit records the M11 milestone: core competence, all eleven at bar. It is the moment the program hands the work back to you for good.

Suggested hour budget. Core content (the compounding thesis, the public record, the annual cadence and the January ritual, the information diet, widening the circle, the professional paths, the audit protocol, and the re-grading ritual): ~7h · worked examples (a model published thesis with its later scorecard; the January ritual run end-to-end; a worked audit of three competencies): ~2h · practice set (self-assessment and design exercises, plus the ritual arithmetic and reading-the-news drills): ~3h · the applied mini-project: the full eleven-competency final audit + the sealed one-pager re-grading + the "how far" note: ~6h · flashcards + mastery check + the closing journal entry: ~2h. Unlike every prior module, the bulk of this one's value is not on the page. It is in the honest audit you run on yourself and the system you commit to running for years. Do not rush the audit to "finish the program"; the audit is the program's final exam.

App link. This module unlocks the portfolio dashboard's "mastery audit" view: the app renders your skill-tree heatmap, your FSRS review health, your calibration score, and your challenge-accuracy history as an eleven-row scorecard mapped onto the Section 1 competencies, so the audit you run by hand here has a data-backed mirror. It also flips on the lifelong-cadence scheduler: the spaced-skill engine (M8.05's "you haven't valued a cyclical in 40 days") now runs indefinitely, and a new quarterly-letter prompt and January-ritual reminder fire on the calendar. The decision journal and calibration dashboard (installed M0.04, fed since M6.05 and Phase 9) become the permanent record the app keeps scoring (reasoning over outcome) for as long as you use it. Nothing in the app "completes" here; that is the entire point. The heatmap you unlock is an instrument you return to, not a trophy you collect.

Learning objectives

By the end you can:

1. Explain why the program ends but the practice does not: state the compounding thesis in your own words (why an analyst's real edge is built in the years after formal study, from reps against reality recorded in a journal), and why stopping causes measurable skill decay.
2. Stand up a public, dated, unedited track record: choose a form (a blog/Substack, an X/Twitter thread cadence, a peer group, or a paper-portfolio log), publish your first thesis, and explain precisely why a public dated record creates accountability and compounds reputation in a way a private one cannot.
3. Run the lifelong practice regime as a life system: operate the annual cadence (≈50 teardowns → watchlist/simulator decisions, 8–10 deep dives, 4 quarterly review letters), the quarterly master-text re-read, and the January ritual (re-pull ERP / risk-free / country-risk data from Damodaran, re-run your valuation defaults, and redo your behavioral self-audit).
4. Build a high-signal information diet and read the news like an analyst: assemble a small stack of the highest-signal free sources (Berkshire letters, Oaktree memos, Mauboussin/Counterpoint, Nomad letters, the best podcasts; India: Bakshi's BFBV, Safal Niveshak), and read any headline as rate-of-change versus expectations, discarding the noise.
5. Expand your circle of competence deliberately and run the mistake loop: pick and learn a new business model or sector to a defined bar, and turn every logged mistake into a permanent checklist item via the journal → checklist loop.
6. Map the professional paths honestly: compare buy-side, sell-side, independent/RIA, and managing your own capital; state what a track record must show to be credible; and articulate the core ethics of managing other people's money (fiduciary duty, suitability, honest reporting).
7. Run the final competency audit: assess yourself against all eleven master-map competencies, for each naming the observable bar, the concrete evidence you should now hold, and the honest "am I there?" test. Then grade it without flinching.
8. Re-grade your week-3 sealed one-pagers and define "done": measure how far you have come, articulate what "core competence" means and why mastery keeps going, and commit the lifelong system to writing.

Prerequisites & connections

Builds on. All seventy-nine modules before it, but most directly the machinery this one makes permanent. The rep cadence, the miss taxonomy/error log, and periodic self-recertification come straight from M8.05 (§4.8 of that module is the seed this one grows into a life system). The knowledge system (atomic notes, company files, the checklist, the decision journal) was installed in M0.04 (Section 5.3 of the master map) and has been fed by every module since. The sealed one-pagers you re-grade were written in M0.05. The personal checklist and the behavioral self-audit are from M6.05; the base-rate/outside-view and circle-of-competence-in-writing discipline is from M6.03; the bias catalog is from M6.04. The January ritual's re-pull of ERP, the currency-matched risk-free rate, and the country-risk premium runs the exact method you built in M3.02 and put into a WACC in M3.03; the reverse-DCF you keep sharp is M3.05. The quarterly review letters, calibration, and the process loop are M9.04; position sizing is M9.01. The communication craft for your public record and letters is M10.03 (Minto, conclusion-first). The two capstone deep dives you just completed (M10.01–M10.02) are the template for the lifelong deep dives. The information diet and reading-the-news-as-rate-of-change discipline extend Phase 7 and the macro dashboard artifact.

Feeds forward. Nothing in the curriculum. This is the terminus. It feeds your life: the years after week 80, in which the master map is explicit that the distance between the analyst you are now and the world-class analyst you are becoming is "roughly 150 more teardowns, 30 more deep dives, and a few full market seasons lived through your own journal" (Section 0.2). Every artifact in the corpus (the two playbooks, the four checklists, the valuation workflow, the sector cheat-sheets, the macro dashboard, and the master reading list) was designed to be used without this program, without a tutor, and without any AI, forever. This module is the instruction manual for that forever.

The one-sentence version of this module. The program was scaffolding; this module removes the last of it, hands you a system for compounding the skill on your own for decades, and makes you prove, honestly and against all eleven bars, that the scaffolding can now come down.

Core teaching content

4.1 Why the program ends but the practice doesn't: the compounding thesis

Here is the uncomfortable, liberating truth the master map has been telling you since Section 0.2: week 80 is not the finish line; it is the starting line with the training wheels finally off. The eighteen months bought you something specific and valuable: the ability to run every core analytical task by hand, correctly, on a company you have never seen, and to know when the numbers are lying to you. That is core competence, and it is genuinely rare. But core competence is not mastery, and the gap between them is not closed by more modules. It is closed by reps against reality over years, and the difference is the difference between a pilot who has passed their check-ride and a pilot with ten thousand hours: the same license and a completely different level of judgment.

Only one thing separates them: time in the seat, logged and reviewed.

Why the skill decays if you stop. Analytical skill is procedural, not declarative (the automaticity thesis of M8.05 §4.1). Procedural skills decay without use, and they decay silently. You do not notice that your instinct for mid-cycle normalization has gone soft until you misjudge a cyclical with real money on it. Three things rot fastest: the plumbing (the mechanical fluency of a hand-spread or a WACC build, which slows without reps), the calibration (your sense of what a fair multiple or a plausible growth rate is, which drifts as markets re-rate and your memory anchors on stale levels), and the frontier (your circle of competence, which shrinks relative to the world as new business models appear and you do not learn them). The lifelong system in this module is engineered specifically to attack all three: the weekly teardown keeps the plumbing fast, the January ritual and the quarterly letters keep the calibration fresh, and the deliberate circle-widening keeps the frontier growing.

Why the edge compounds, and why it compounds only if recorded. An analyst's durable edge is not raw intelligence; it is a large, honestly-graded library of pattern recognition: "I have seen this shape of business, this kind of governance smell, this stage of a capital cycle before, and here is how it played out." That library is built one rep at a time, and it compounds like capital: the hundredth teardown is faster and sharper than the tenth because it stands on ninety-nine stored patterns. But (and this is the load-bearing point) it only compounds if the reps are recorded and reviewed. A rep you did and forgot teaches almost nothing; a rep you journaled, with a dated thesis and a "what would change my mind," and then re-read a year later against what actually happened, teaches enormously, because it lets you grade your own reasoning against reality and update your base rates (working-agreement 6, and the calibration spine of M9.04). This is why the decision journal is "non-negotiable" and why "your edge in year five compounds from what you record in year one."

The compounding thesis, in one line: reps build the library, the journal makes the library honest, and honesty is the edge.

What "the practice" actually is. Strip away the romance and the lifelong practice is five concrete, repeatable activities, each of which this module operationalizes: (1) you publish a dated record so you cannot lie to yourself; (2) you rep on a fixed cadence so the skill stays fast and the library grows; (3) you stay current on a tiny high-signal diet so your inputs are good and your attention is not shredded; (4) you widen the circle deliberately so the frontier grows; and (5) you run the mistake loop so every error becomes a permanent upgrade instead of a repeated tax. The rest of Section 4 is those five, plus the professional-paths map for those who want it and the audit that closes the program.

4.2 Building a public track record: the honesty engine

The single highest-leverage habit you can start the week you finish this program is to make your analysis public, dated, and unedited. Not because anyone is waiting to read it (at first no one will) but because of what publishing does to you. It is the honesty engine, and here is the mechanism.

Why public beats private (the accountability mechanism). A private thesis is infinitely editable by your own memory. Six months later, when the stock has fallen, you will remember having been "more cautious than that," and when it has risen you will remember "calling it". This is hindsight bias (M6.04) operating on an unfalsifiable record, and it quietly destroys calibration. A public, dated, timestamped thesis is falsifiable. It says, in writing that you cannot alter, "on 12 March 2026 I judged this a good business at a demanding price, and here is exactly why, and here is what would change my mind." When the future arrives, you cannot squirm: the record grades you. That external, immutable timestamp is the thing a private journal cannot fully provide (you can always secretly edit a private file), and it is why the discipline is public and unedited: the point is not the audience, it is the irreversibility. You are borrowing the crowd's memory to defeat your own hindsight.

Why it compounds reputation. Over years, a public record of dated, honestly-reasoned theses (including the ones that were wrong, left standing and annotated with what you learned) becomes the single most credible thing you can show anyone: a prospective employer, a co-investor, a future client, or a peer group. It is credible precisely because it includes the misses: anyone can cherry-pick winners after the fact, but a continuous dated record that you never edited proves process, not luck (the whole ethos of working-agreement 8). A five-year public record of forty theses, honestly scored, is worth more than any CV line, because it is the one thing that cannot be faked. This is the asset the "going professional" paths in §4.6 are gated on, and it takes years to build, which is exactly why you start it in week 81, not the day you decide to go pro.

The forms: pick one, start small. You do not need an audience or a platform; you need a dated public commitment. In rough order of increasing exposure:

The peer group and the paper-portfolio log are the recommended starting forms because they are low-friction and high-honesty; a blog or Substack is the recommended primary form for anyone who might go professional, because it produces the durable, linkable portfolio. Many analysts run two: a paper-portfolio log for calls and a blog for the long-form theses behind the biggest ones.

How to start (this is the whole trick): publish before you feel ready. The universal failure here is waiting: for a better platform, a more impressive first piece, a bigger edge. The fix is a rule: your first published piece is the next teardown you were going to do anyway, posted as-is, dated, with a "what would change my mind" line and a note that you will score it honestly later. Three disciplines make the record trustworthy: (1) date everything and never back-date; (2) never delete or silently edit: if you were wrong, append a dated correction or a scorecard, leaving the original visible (the "strikethrough, don't delete" rule); (3) always include your falsifier, the specific thing that would prove you wrong, so the later grade is objective, not a vibe. And two guardrails, both from the program's spine: everything you publish is analysis, not advice (working-agreement 9: a company is a teaching specimen and a thesis is your reasoning made public, not a recommendation to anyone), and every number in it must be traceable to a primary filing (working-agreement 7: a public record with a made-up number is worse than no record).

Start this week. The record you will value most in 2031 is the ugly, honest one you started in 2026.

4.3 The ongoing practice regime: the annual cadence, the reading regime, and the January ritual

M8.05 §4.8 gave you the seed cadence; here it becomes a full-year operating system with three layers: a rep layer, a reading layer, and an annual-maintenance layer.

Layer 1: the rep cadence (the year's analytical volume). The company-of-the-week does not stop at week 80; it changes shape one last time and then runs forever (master map Section 5.2, weeks 70–80+): every teardown ends in a watchlist decision (own / watch / pass) and, where you run the simulator or real capital, a journaled action with a dated thesis and sell criteria. The target volume, which is also the master map's arithmetic for the road from core competence to mastery, is:

That is ~50 teardowns and ~10 deep dives a year, so in the ~3 years from week 80 the master map's "150 more teardowns, 30 more deep dives" is simply this cadence, sustained. The teardowns keep the plumbing automatic and grow the pattern library; the deep dives produce the theses you actually act on and publish; the quarterly letter is the connective tissue. Modeled on Buffett's and the great fund letters, written to an imagined partner (even if the partner is only your future self), it forces you to summarize what you did, own what you got wrong, and state what you learned. Writing four honest letters a year is, over a decade, one of the most powerful calibration instruments in existence, because it makes you narrate your own reasoning to an audience four times a year and confront the gap between the story and the results.

Layer 2: the reading regime (keeping the foundations from fading). Reps keep the skill fast; re-reading keeps the foundations deep. The regime is one master text re-read per quarter, four a year, the ones whose ideas you most want to stay resident:

• Q1: a valuation/quality text (Damodaran's The Little Book of Valuation or selected chapters of Investment Valuation; or Mauboussin's Expectations Investing) to keep the reverse-DCF and value-driver instincts sharp.
• Q2: a philosophy/temperament text (Graham's The Intelligent Investor chapters 8 and 20, on Mr. Market and margin of safety, reward re-reading annually; or the latest Berkshire letter and a batch of Oaktree memos).
• Q3: a strategy/moat text (Greenwald's Competition Demystified, Helmer's 7 Powers, or Mauboussin's Measuring the Moat) to keep the four-lens moat protocol from collapsing into adjective-matching.
• Q4: a macro/cycles text or a behavioral text (Marks's Mastering the Market Cycle, or Kahneman revisited, or a Dalio re-watch of How the Economic Machine Works) to keep the weather-report and the self-knowledge current.

The point of re-reading rather than always reading something new is that the master texts are dense enough that you extract different things at different stages of your own development. Graham at year one and Graham at year five are different books, because you are different. New reading happens too (that is the information diet, §4.4, and the circle-widening, §4.5), but the annual re-read is the spaced repetition of your foundations.

Layer 3: the January ritual (the annual recalibration). Once a year you run a fixed maintenance ritual that resets your quantitative and behavioral baselines. January is the natural date, because Damodaran refreshes his data library in early January and most companies' fiscal-year data has landed. Skipping it is how analysts end up running five-year-old assumptions without noticing. Three tasks:

1. Re-pull the market inputs and re-run your valuation defaults. Go to Damodaran's data page (pages.stern.nyu.edu/~adamodar) and pull the fresh values: the implied US equity risk premium (he updates it monthly; the January value is the natural anchor), the risk-free rates (the US 10-year T-bond yield; and for India the local-currency G-sec yield and the Damodaran "clean" adjustment for the sovereign default spread), the country risk premium for India (his default-spread × relative-equity-volatility method), and the industry data you lean on (unlevered betas, and the sector cost-of-capital and margin/ROIC benchmarks). Then re-run the WACCs of your two or three core watchlist names with the fresh inputs and note how the default changed. The point is not any one company; it is that your standing assumptions about "what a fair discount rate is" and "what a normal margin is" get re-anchored to the current world every year, defeating the calibration drift of §4.1. (As of early-2026, illustrative anchors to expect roughly: US 10-year in the ~4–4.5% area, implied US ERP in the ~4.5% area, India country-risk premium in the ~2.5–3% area. All move, all must be re-pulled, none is permanent: check each January.)
2. Redo the behavioral self-audit. Re-run the scored self-audit from M6.05. Which two or three biases are currently most live in your decisions? Your dominant biases change with your circumstances (a run of wins breeds overconfidence; a drawdown breeds loss aversion and inaction). Compare this year's audit to last year's and note the movement; update the environment fixes and checklist items that counter your current top biases, not the ones you had a year ago.
3. Review the year's error log and update the checklist. Read the miss taxonomy (M8.05 §4.2) and the year's journaled mistakes as a batch, find the pattern, and promote the recurring errors into new checklist items (the mistake loop, §4.5). This is where the checklist grows deliberately, once a year, from your own logged evidence, rather than sprawling.

The January ritual takes a focused day. It is the analytical equivalent of a pilot's annual recurrency training: cheap, boring, and the thing that prevents the expensive, exciting failure.

4.4 Staying current without drowning: the information diet

The modern analyst's problem is not too little information; it is a firehose engineered to capture attention and sell it, most of which is noise that degrades judgment. The skill is not consuming more. It is a deliberately small diet of high-signal sources plus a way of reading anything that separates signal from noise. Two parts: the diet, and the reading method.

The diet: a small stack of the highest-signal free sources. The organizing principle: prefer sources that think in years and reason in public over sources that react in minutes. The durable free stack, most of it the "ongoing free letters" the master reading list catalogues:

• Berkshire Hathaway annual letters (Warren Buffett). The clearest writing on business quality, capital allocation, and temperament in existence; decades of them are free. Read the new one each February; re-read old ones by theme. [Free]
• Oaktree memos (Howard Marks). The best free writing on cycles, risk, and sentiment: where are we in the pendulum, what is being priced. A new memo is a drop-everything read. [Free]
• Michael Mauboussin (Counterpoint Global / Morgan Stanley). The most rigorous free research on ROIC, moats, base rates, and expectations: Measuring the Moat, The Base Rate Book, and the periodic pieces. [Free]
• The Nomad Investment Partnership letters (Nick Sleep). A masterclass in long-horizon thinking and "scale economies shared," free as an archived collection; re-read annually. [Free]
• India: Sanjay Bakshi (Fundoo Professor / BFBV) and Safal Niveshak. Bakshi's Behavioural Finance & Business Valuation notes and blog are the best India-specific treatment of temperament, valuation, and the promoter/governance lens; Safal Niveshak (Vishal Khandelwal) is a durable, thoughtful India value-investing resource. [Free]
• A very short podcast rotation. Invest Like the Best, Business Breakdowns, and Acquired for business teardowns done aloud by professionals (use them for calibration: pause, form your own read, then compare); Odd Lots for macro and market plumbing. Two or three episodes a week, chosen by topic, is plenty. [Free]

That is the whole diet: under a dozen sources, almost all free, none of them a real-time feed. Notice what is not on it: no minute-by-minute market news, no financial television, no anonymous tip accounts, no "hot stock" newsletters. Those are not information; they are attention extraction, and they push you toward the recency, herding, and overtrading errors of M6.04. Guard the diet's smallness as fiercely as its quality.

The goal is a diet, not a buffet.

The reading method: rate-of-change versus expectations. When you do read the news (an earnings print, a policy decision, a data release), read it the way the macro dashboard taught you (Phase 7 and the macro-dashboard artifact): not as a level, but as a rate-of-change measured against what was already expected. Three questions turn any headline into signal:

1. What was expected? Markets price in the consensus expectation, not the level. "Company grows revenue 12%" is meaningless until you know whether 12% beat or missed the ~15% the price implied. "RBI holds the repo rate" is a non-event if a hold was expected and a large event if a cut was. The number is only information relative to the prior expectation.
2. What is the rate of change and is it accelerating or decelerating? A metric improving but decelerating (revenue up 12% after four quarters of 20%) often matters more than the level, because markets price the second derivative. The macro dashboard's whole discipline is reading every indicator as a direction and an acceleration, not a snapshot.
3. Does it change the thesis, or is it noise? For 95% of headlines about a company you own, the honest answer is "it does not change the multi-year thesis": a single quarter's beat/miss, a broker upgrade, a macro scare. The analyst's edge is inaction on noise (the discipline of M9.03 and the temperament of M6.05): you already wrote, in your dated thesis, the specific things that would change your mind; if the news is not one of them, it is noise, and you note it and move on. The rare headline that is on your falsifier list gets your full attention. This filter ("is this on my pre-written list of thesis-breakers?") is what lets you follow the world closely without being jerked around by it.

The synthesis: a small diet of sources that reason in years, read as rate-of-change against expectations, filtered by your own pre-written falsifiers. That is how you stay genuinely current while everyone around you drowns.

4.5 Continuous learning: widening the circle and the mistake loop

Two engines keep an analyst improving after the plumbing is automatic: widening the circle of competence (learning things you cannot yet analyze) and the mistake loop (turning errors into permanent upgrades). The first grows the frontier; the second raises the floor.

Widening the circle of competence: deliberately, not accidentally. Your circle of competence is the set of businesses you can genuinely analyze: you understand the model, the KPIs, the drivers, and the failure modes well enough to have an edge (M6.03; you wrote yours down there). The two errors are opposite: never widening it (so the world outgrows you as new business models appear) and fooling yourself about its edges (the far more dangerous one: believing you understand something you do not, which is how smart people lose money in businesses they had no business owning). The discipline is to widen it on purpose, one domain at a time, to a defined bar, before you need it. A concrete protocol for adding a new sector or model to the circle:

1. Pick the domain and the bar. Choose one new business model or sector that keeps appearing on your watchlist or in the world (say, insurance if you have avoided it, or a new model like usage-based/consumption software, or a frontier area like a new energy value chain). The bar is the program's own sector-playbook standard: you can state the business model, the specific KPIs with healthy/stress thresholds, the right valuation lens, and the red flags, which is the exact shape of a Phase 5 module.
2. Learn it the program's way. Read the two or three anchor sources (the master reading list's sector section, plus the relevant chapter of the sector cheat-sheets artifact), then immediately tear down one clear exemplar and one troubled one in the domain: theory landing on a real filing within 48 hours (working-agreement 4), one Indian and one US where both exist.
3. Certify it to yourself. Run a cold teardown of a third, never-studied name in the domain and score it against the teardown rubric (M8.05). If it clears the bar, the domain is inside the circle; if not, you have found exactly what you still do not understand, and you drill that. This is the same self-certification honesty as M8.05. A domain you feel you understand but cannot tear down cold is not in the circle yet.

Two or three deliberate widenings a year, each certified, and in five years your circle has grown to cover most of the investable world, while the analyst who "keeps up" by osmosis has a circle that quietly shrank. And the corollary discipline, equally important: maintain a written list of what is outside your circle and stay off it, the businesses whose economics you have honestly concluded you cannot judge (certain frontier tech, some financials, anything whose disclosure you cannot trust). Knowing the edge of the circle, and respecting it, is worth more than any single widening.

The mistake loop: journal → pattern → checklist. Every serious analyst has a checklist, and the only legitimate way it grows is from mistakes, yours and the masters' (M6.05; the program's checklist "grows only from logged mistakes"). The loop, run continuously and formalized once a quarter in the letter and once a year in the January ritual:

• Log every mistake in the journal, honestly, with the cause named. Not "the stock went down" (that is an outcome, and a good decision can produce a bad outcome; working-agreement 8) but "I anchored on management's narrative and skipped the reverse-DCF, so I overpaid" (that is a process error, and it is fixable). The mistake taxonomy from M8.05 §4.2 gives you the categories: coverage, calibration, speed, false-precision, sourcing, tripwire.
• Find the pattern across mistakes. One error is an incident; the same error three times is a systematic flaw and your highest-value fix. Reading the log as a batch (quarterly, annually) surfaces the pattern that any single entry hides: "I keep under-scanning governance on founder-led growth stories," say.
• Promote the pattern into a checklist item. The fix is a permanent, boring, always-run checklist line that makes the error mechanically hard to repeat: "For any founder-controlled company, run the full related-party + pledge + auditor-tenure scan before forming a price view — no exceptions." That is a mistake converted into a permanent upgrade. Do this and your checklist becomes a compressed autobiography of everything you have ever gotten wrong, which is exactly what a great checklist is.

The two engines together: widening raises your ceiling (more of the world you can analyze), the mistake loop raises your floor (fewer of your known errors repeated). An analyst who runs both for a decade is a genuinely different animal from one who ran neither, even with identical starting talent.

4.7 The final competency audit: the protocol

This is the milestone the entire program was backward-designed to deliver: an honest, structured self-assessment against all eleven competencies of the master map's Section 1, run competency by competency, to certify core competence, M11. It is not a quiz; it is a portfolio review of yourself, and its only value is its honesty (the same integrity rule as M8.05's self-certification: a generous audit certifies nothing except a false confidence that surfaces later, expensively).

How to run it. For each of the eleven competencies you assess three things: (1) the observable bar, the specific, demonstrable thing the master map says "done" looks like (not a feeling but an action you can perform); (2) the evidence, the concrete artifacts you should now hold that prove you can clear the bar (a graded capstone, a passed gate, a built dashboard, a filled journal); and (3) the honest "am I there?" test, a single hard question that you answer yes or not yet, with the evidence in front of you, as a hostile examiner would. You score each competency 2 (at bar: you can demonstrate it cold, with evidence), 1 (partial: you can do it with support or on easy cases but not cold), or 0 (not yet). The protocol below is the instrument; the applied mini-project (§8) is where you run it in full and where the pass standard lives.

The eleven bars, the evidence, and the honest test:

Three cross-cutting threads are audited inside these rows rather than as separate lines (master map Section 1): ethics & governance lives inside C2 and C7 (can I judge honesty and spot governance rot?); data & tools fluency lives inside the evidence for C1, C5, C6, and C8 (can I pull the filing, drive the screener, build the model?); and communication is C11. And the knowledge system (M0.04) is audited by a simple existence test: is it alive: atomic notes accumulating, a company file for everything studied, a checklist grown from logged mistakes, a decision journal with dated pre-outcome entries? A dead knowledge system is itself a failing grade on the practice.

4.8 The re-grading ritual and what "done" means

Breaking the seal: the "how far have I come" reflection. In week 3 (M0.05) you wrote your first naive one-pagers on a full Indian annual report and a full US 10-K, and sealed them, to be reopened now. This is the program's emotional close, and it is also a genuine assessment instrument, so do it deliberately: reopen them, read them without wincing, and re-grade them with the eyes you now have. Mark, specifically, (1) every place you now see that you missed the real profit engine, mistook reported profit for real profit, called a moat that was not there or missed one that was, ignored the balance sheet, or had no idea what the price was implying; (2) every concept you now command that you did not have a word for then (NOPAT, incremental ROIC, reverse-DCF, cash conversion, the tripwire scan, mid-cycle normalization); and (3) the one thing your naive self got right by instinct, because there is usually one, and it tells you something about your native judgment. Then write the "how far" note: a page comparing the analyst who wrote those one-pagers to the analyst reading them now. This is not sentiment for its own sake. It calibrates you against your own baseline, which is the truest measure of growth there is, and it is a document you will treasure and re-read in years to come. (Then re-seal the originals and the note, to reopen at some future milestone; the ritual repeats.)

What "done" means, and why it isn't. Passing the final audit records core competence: every one of the eleven competencies demonstrated at its stated bar, with evidence, honestly. That is a real, rare, and worthy achievement: the "core professional competence" the program promised, reached at about eighteen months on the immersive pace. But "done" here means done with the scaffolding, not done improving, and the distinction is the whole philosophy of this module. The master map is explicit (Section 0.2): the difference between the analyst you are at week 80 and the world-class analyst you are becoming is "roughly 150 more teardowns, 30 more deep dives, and a few full market seasons lived through your own journal." Core competence is the license; mastery is the career.

Why mastery continues without end: the domain itself moves (new business models, new sectors, new accounting, new market regimes; the frontier of §4.5 never stops receding); calibration is never finished (every market season teaches something the last one could not); and the pattern library compounds without limit (the ten-thousandth teardown still teaches, because it adds a pattern the previous nine-thousand-nine-hundred-ninety-nine did not have). The greatest investors read companies into their nineties not from obligation but because the game is genuinely inexhaustible and they never stopped being students. That is the destination this program was really pointing at — not a finish line, but a practice you love enough to run for life. You now have the skill, the system, and the honesty to do exactly that.

The audit below is the last thing the program asks of you; everything after it, you do for yourself.

Common mistakes & how experts think differently

These are the mindset errors that decide whether the skill you built keeps compounding for decades or quietly rots. They are the failure modes of the practice itself, not of any single analysis.

1. Treating week 80 as a finish line. The novice frames graduation as arrival and eases off; six months later the plumbing is slow and the calibration is stale. The expert frames it as the start of the real education, the point at which reps against reality begin, and knows that a skill unused decays silently (§4.1). The tell of the mistake is relief ("finally done"); the tell of the right mindset is appetite ("now I get to do this for real, for years").

2. Waiting to publish until you are "ready." The novice hoards analysis privately, polishing toward a perfect first piece that never ships, and so never gets the honesty benefit of a dated public record. The expert publishes the next teardown they were doing anyway, warts and all, because the value of the record is its irreversibility, not its polish (§4.2). Ready is a trap; dated is the point.

3. Editing or curating the record. The novice deletes the losers, back-dates the winners, and quietly rewrites the theses that aged badly, producing a record that proves luck, not process, and destroying their own calibration in the bargain. The expert leaves every miss standing, annotated with what it taught, because the misses are the credibility (§4.2, §4.6): a record you would edit is a record no one, including you, should trust.

4. Confusing consumption with improvement (information addiction). The novice mistakes reading twenty market-news items a day for staying sharp, and feels productive while their judgment is shredded by recency and noise. The expert keeps a deliberately tiny high-signal diet and reads it as rate-of-change against expectations, treating the firehose as attention-extraction to be avoided (§4.4). More inputs is not more edge; better inputs, filtered by your own falsifiers, is.

5. Producing nothing (consuming instead of repping). The cousin of #4: the novice reads about investing endlessly and analyzes almost nothing, because consuming is comfortable and producing an output is hard. The expert protects the output cadence, fifty teardowns and eight-to-ten deep dives a year, because the skill lives in the produced output (the twelve answered questions and the verdict), never in the reading (M8.05, common mistake 2). If a year passed with much reading and few teardowns, the year built comfort, not skill.

6. Skipping the January ritual and running frozen assumptions. The novice sets a discount rate and a "normal margin" once and uses them for years, drifting further from reality every quarter without noticing. The expert re-pulls the ERP, the risk-free rate, and the country-risk premium every January, re-runs the defaults, and re-anchors their sense of "fair" to the current world (§4.3), because a five-year-old WACC is a silent, compounding error. Calibration is a maintenance task, not a one-time setting.

7. Circle-of-competence self-deception. The novice believes they understand a business they cannot actually tear down cold, and finds out the hard way with real money. The expert widens the circle deliberately and certifies each widening (a cold, rubric-scored teardown of a never-studied name in the domain), and keeps an explicit written list of what is outside the circle to stay off it (§4.5). The dangerous edge of the circle is not the part you know you don't know — it's the part you wrongly think you do.

8. The mistake tax: repeating errors instead of promoting them. The novice makes the same process error repeatedly (anchors on the narrative, skips the governance scan) and treats each instance as bad luck. The expert runs the mistake loop: logs the error by cause, finds the pattern across the log, and promotes it into a permanent checklist item that makes the error mechanically hard to repeat (§4.5). A mistake you don't convert into a checklist line, you will pay for again; a mistake you do, you never pay for twice.

9. Believing "going professional" is the only validation. The novice thinks the skill only "counts" if it becomes a finance job, and undervalues the enormous returns to understanding businesses as an operator, board member, entrepreneur, or steward of one's own capital. The expert knows the capability is valuable on all four professional paths and off all of them (§4.6), and chooses the path (including "none") that fits their temperament and life — not the one with the most prestige. The skill is the asset; the job is one optional way to package it.

10. Grading the final audit generously. The novice wants the M11 milestone badly enough to score themselves "at bar" on competencies they can only perform with support, turning the audit into a self-esteem instrument. The expert grades as a hostile examiner with the evidence in front of them, scores "not yet" without shame where it is honest, and stays in the relevant practice until the evidence is real (§4.7), because the audit exists to tell you the truth, and the only person a generous audit fools is the one whose money is later on the line. Effort is not a criterion; demonstrable, cold, evidenced capability is.

11. Ethics as an afterthought once real money appears. The novice treats the fiduciary duties of managing others' money as paperwork to be minimized. The expert treats them as first-order (§4.6, Section 8.1): client interest strictly first, honest reporting including the bad quarters, conflicts disclosed, competence bounded, regulation respected. The analyst's entire long-run value rests on trust, and trust is destroyed once and permanently. The same honesty that makes your private record useful is a duty the moment the capital is not yours.

12. Letting the knowledge system die. The novice stops feeding the notes, the company files, and the journal once the modules stop enforcing it, and loses the compounding asset that was the whole point. The expert keeps the system alive because it is the pattern library and the calibration record: a company file for every name studied, a checklist grown from logged mistakes, a journal of dated pre-outcome theses (§4.7, M0.04). A dead knowledge system means the reps stopped compounding; a living one is the machinery of the edge itself.

Worked examples

Three examples in the spirit of a module whose subject is your own practice: a model public thesis with the later scorecard that makes it honest; the January ritual run end-to-end on a real pair of companies; and a worked audit of three competencies, including one honest "not yet." Company figures are round approximations labeled illustrative, modeled on ~FY24/FY25 scale: pull the actual filing and recompute; the companies are teaching specimens, not recommendations, and every valuation is a range.

WE1: A model public thesis and its honest scorecard

The point of this example is to show what a publishable, dated, unedited thesis looks like (§4.2) and, more importantly, what the honest scorecard appended a year later looks like, because the scorecard is where the credibility and the calibration live.

The dated thesis (as published, 12 March 2026, illustrative).

Pidilite Industries (NSE: PIDILITIND). Own/watch/pass: WATCH. Analysis, not advice; a teaching specimen. Business (1 line): the branded adhesives and construction-chemicals leader (Fevicol, Dr. Fixit), selling low-ticket, high-frequency products through a vast distribution network to carpenters, contractors, and households. Why it's a good business: brand + distribution moat confirmed in the numbers: ROCE ~25–28% vs a ~12% hurdle, sustained; pricing power through commodity-input cycles; asset-light, cash-generative (illustrative, ~FY24/FY25 scale). Variant question: the market prices ~50× earnings, implying category leadership and pricing power roughly forever. My variant is whether input-cost normalization (VAM) plus premiumization can hold EBITDA-margin gains, or whether they reverse as competition and raw materials mean-revert. The 3 things that must go right: (a) volume growth stays mid-to-high single digit; (b) margins hold near the elevated band, not revert to the historical ~20–22%; (c) the moat resists D2C/regional fragmentation in construction chemicals. Valuation & the price verdict: reverse-DCF on ~1.3% FCF yield at ~12% required return implies ~10.7% perpetual growth, roughly India's nominal GDP, i.e., the price already assumes durable category leadership. Verdict: demanding; a great business at a price that leaves little margin of safety. What would change my mind (my falsifiers): two consecutive quarters of volume growth <3%; A&P cut to defend margin; a step-down in ROCE below ~20%; a credible national competitor taking share in adhesives. Sell/act criteria: would move WATCH→OWN only on a de-rating toward the high-30s P/E with the moat intact, or a demonstrated new growth vector (e.g., international/adjacencies scaling).

The scorecard, appended one year later (12 March 2027; illustrative, do not edit the original above it).

Scorecard. Kept WATCH all year; did not buy. Over the year (illustrative): volume growth held ~6–7% ✓; margins stayed near the elevated band ✓; no national competitor took material share ✓; the stock de-rated modestly but never to my high-30s trigger, so no action. Process grade: sound. The thesis was right on the business and disciplined on price; I correctly did not pay up. Calibration note: my falsifiers were well-specified and none tripped, which is why WATCH was the right standing call. What I'd do differently: my "would change my mind" list was all downside; I had no upside trigger beyond a de-rate, so I had no plan for "great business keeps compounding and never gets cheap", a real gap, because demanding-but-wonderful businesses sometimes deserve a starter position anyway (the "compounder you never buy because it's never cheap" error, M9.03). Checklist upgrade promoted: for a wonderful business at a demanding price, pre-commit an upside trigger and a starter-position rule, not only a de-rate trigger.

Why this is the model. The thesis is dated, falsifiable, and traceable to filings; the price verdict is a reverse-DCF read, not a target; the scorecard grades the reasoning (sound process), not the P&L; and it ends in a checklist upgrade promoted from a real gap in the analysis: the mistake loop (§4.5) running in public. This one artifact, repeated forty times over five years, is the credible track record of §4.6.

WE2: The January ritual, run end-to-end

The point here is to show the ritual (§4.3) as a concrete annual task: re-pull the inputs, re-run the default, note the delta. We do a US name (clean) and flag the India adjustment. All inputs are illustrative: as of early-2026, verify each January from Damodaran's data page.

Step 1: re-pull the market inputs (from pages.stern.nyu.edu/~adamodar, January refresh).

Step 2: re-run a US WACC default (illustrative US software compounder, "NovaCloud Inc"). Using the M3.03 build. Take a relevered equity beta of ~1.15 (unlevered 1.05, light debt). Cost of equity = risk-free + β × ERP = 4.3% + 1.15 × 4.5% = 4.3% + 5.18% = 9.5%. With a near-all-equity structure, WACC ≈ 9.5%. Last January the same build gave cost of equity = 3.9% + 1.20 × 4.6% = 3.9% + 5.52% = 9.4%. Delta: essentially flat (~+0.1%). The higher risk-free almost exactly offset the slightly lower ERP and beta. The lesson is not the tiny net change; it is that you verified it rather than assuming, and you now know your ~9.5% default is current, not a three-year-old habit.

Step 3: the India adjustment (illustrative Indian software exporter, "Bharat Systems"). For an INR valuation you rebuild the cost of equity in rupees (M3.02): cost of equity (INR) = INR-clean risk-free + β × (mature-market ERP + India CRP). Illustratively, ≈ 5.0% + 1.10 × (4.5% + 2.7%) = 5.0% + 1.10 × 7.2% = 5.0% + 7.92% = 12.9%. Note the country-risk premium fell (3.0% → 2.7%), which nudged the INR cost of equity down year-on-year, a real change you would have missed by running last year's number. (The INR-clean risk-free applies Damodaran's method of stripping the sovereign default spread from the local G-sec; the exact mechanics are in M3.02, and here the point is the ritual, not re-deriving the method.)

Step 4: the behavioral half. Re-run the M6.05 self-audit. Suppose last year's dominant biases were overconfidence (after a good run) and this year's are loss aversion and inaction (after a flat, choppy year), a completely different pair, requiring different environment fixes (last year: force a steelman on every buy; this year: a pre-committed rule to act on your falsifiers rather than freeze). Then batch-read the year's error log, find the pattern, promote one checklist item. Total time: a focused day. Output: current WACC defaults for both currencies, a refreshed sense of "fair," an updated bias profile, and one new checklist line. That is the ritual: cheap, boring, and the thing that keeps a decade of analysis from drifting quietly out of calibration.

WE3: A worked audit of three competencies (including an honest "not yet")

The point is to demonstrate the §4.7 protocol (bar, evidence, honest test, score) with the honesty that makes it worth anything. We audit C1, C5, and C10 for an illustrative learner at week 80.

C1: Read the language of business. Bar: spread any filing by hand, explain every line and footnote, rebuild the three-statement linkage from memory. Evidence held: M1.10 hand-spread of a full 10-K and an Indian AR, graded 92% vs key; M1.05 linkage exam 94%; app statement-puzzle branch at mastery. Honest test: "Handed the Reliance AR and a Deere 10-K cold last month, did I spread both and rebuild the linkage without a template?" Yes, did both in the capstones. Score: 2 (at bar).

C5: Value a business three ways. Bar: intrinsic, relative, and expectations valuations with internally consistent assumptions; know which to trust when. Evidence held: M3.10 three-way valuation of one Indian + one US company, assumptions inside the defensible bands; app sandbox challenges cleared; the M10.01–M10.02 capstones each carried a full three-way valuation. Honest test: "On the capstone, did my DCF, comps, and reverse-DCF use consistent assumptions (g = RR × ROIC held; TV growth ≤ economy), and did I say which to trust for that specific company and why?" Yes, and I correctly leaned on the reverse-DCF for the high-multiple name and on comps for the cyclical. Score: 2 (at bar).

C10: The signature skill. Bar: teardown in 1–2 hours and deep dive in 2–3 days on a novel company, institutional-grade thesis. Evidence held: M8.05 certification passed (two timed teardowns ≥22/28, deep dive ≥26/32); M10.01–M10.02 capstones passed the institutional rubric. Honest test: "Cold, on a never-seen company, on the clock — can I do it?" Yes, certified and then done twice more at higher stakes. Score: 2 (at bar).

And an honest "not yet" (illustrative): C8, Understand the macro machine. Suppose the same learner, auditing honestly, finds: Evidence held: M7.07 dashboard built and mechanism exam passed at 86%; regime memo graded acceptable. Honest test: "Can I explain, mechanically, how a 50bp RBI hike transmits to the specific NBFC I own — and did I actually maintain the dashboard since week 63, or did it go stale?" The mechanism I can do; but I let the dashboard lapse for two months and my regime read is rusty. Score: 1 (partial). The honest consequence: C8 is not yet at bar, so the audit is not a clean pass, and the remediation is specific and small (revive the dashboard, run four weeks of rate-of-change reads, re-take the regime-memo exercise), not a vague "study more macro." This is exactly what the audit is for: not a grade to feel good about, but a precise map of the one thing still to close before core competence is genuine. A learner who scored C8 a generous "2" to finish the program would have learned nothing; the one who scored it an honest "1" knows exactly what to do this month.

WE4: Scoring a year of forecasts: the calibration scorecard (US $)

The §4.2 scorecard and the §4.3 quarterly letter both promise to grade your reasoning, not your P&L. But "graded reasoning" is only honest when it is measured. The standard instrument is the Brier score: the mean squared error of your probability forecasts (the calibration spine of M9.04, made quantitative). This runs a year of it. All figures illustrative, synthetic.

The setup. Harbor Lane Research (synthetic) keeps a public paper-portfolio log and a $100,000 notional book. Over the year it published 20 dated, probabilistic calls, each a falsifiable claim carrying a stated confidence ("70%: this name holds net revenue retention above 115% through FY"). Per the module's spine, the year-end letter is scored on the calibration table below, not on the book's return.

Step 1: tabulate the calls by stated confidence and score each bucket. For a call given probability p, the Brier contribution is (p − 1)² if it came true and (p − 0)² if it did not (outcome 1 or 0); the Brier score is the average of those over all calls.

The 90% row, for instance, is 6×(0.9−1)² + 2×0.9² = 0.06 + 1.62 = 1.68.

Step 2: the headline numbers. Brier score = 4.60 ÷ 20 = 0.23. Overall hit-rate = 13 ÷ 20 = 65%. Average stated confidence = (0.9×8 + 0.7×8 + 0.5×4) ÷ 20 = 14.8 ÷ 20 = 74%.

Step 3: read the score honestly. A 65% hit-rate sounds like an edge. But the Brier of 0.23 is barely better than the 0.25 you would score by mechanically saying "50%" to every call, and that near-tie is the whole finding. The calibration table shows why: the analyst is overconfident, claiming 74% on average while being right only 65% of the time, and the damage concentrates in the 90% bucket (said 90%, realized 75%, with two calls wrong at a punishing 0.81 each). Under-confidence would have been the safer error; this is the "a run of wins breeds overconfidence" drift the January ritual (§4.3) exists to catch. The mistake-loop output (§4.5): promote a checklist line: "cap stated confidence at 80% unless the base rate and the evidence jointly justify more; no naked 90% calls." The scorecard did its job: it converted a comfortable-looking hit-rate into a specific, fixable calibration flaw, which is precisely what "graded on reasoning, not P&L" means once you put a number on it.

WE5: The cadence that builds the record: coverage refresh and the compounding library (India ₹)

The annual cadence of §4.3 is not an arbitrary teardown count; it is sized to keep a coverage universe from going stale, and sustained over years it is the engine that compounds both the pattern library and the track record. This does that arithmetic for a synthetic Indian independent analyst. All figures illustrative, synthetic.

Step 1: the cadence is sized to the coverage refresh. The analyst carries a watchlist of 24 names and wants each re-torn-down at least every 6 months, so no read drifts out of date. Teardowns required = 24 × (12 ÷ 6) = 24 × 2 = 48 per year, essentially the ~50-teardown cadence. "About fifty a year" is not folklore; it is what a two-dozen-name watchlist refreshed twice a year actually costs.

Step 2: what happens when the watchlist grows but the cadence doesn't. Success widens coverage: say the watchlist reaches 40 names while the cadence stays at 50. Refreshes per name = 50 ÷ 40 = 1.25 per year, so the average name's read is now 12 ÷ 1.25 = 9.6 months old: silent calibration drift (§4.1) creeping in through coverage sprawl. The fix is a deliberate choice, not a shrug: trim the watchlist back toward two dozen, or lift the cadence, the same "guard the smallness" discipline as the information diet (§4.4).

Step 3: the library compounds, and so does the record. Sustain the cadence and the arithmetic is the road to mastery: at ~50 teardowns and ~10 deep dives a year, three years produces 3 × 50 = 150 teardowns and 3 × 10 = 30 deep dives, exactly the master map's "150 more teardowns, 30 more deep dives" (§4.1, §4.8). The public ₹ record built alongside it: a paper book of ₹10,00,000 grown to ₹18,00,000 over five years is a CAGR of (18 ÷ 10)^(1/5) − 1 = 12.5%, roughly 1.5 points above an ~11% benchmark. But the credential (§4.6) is not the 12.5%. It is the process-visible, dated, unedited log of ~40 own-calls across a full bull-and-bear cycle, on which the hit-rate of 26 ÷ 40 = 65% clears the ~50% base rate and is spread across 40 names and both regimes, which is what makes it skill rather than one lucky bet.

The outcome is the by-product; the recorded, honest process is the asset.

Practice set

Twelve problems, ramped guided → independent → timed. Because this module's subject is your own practice, several are design and self-assessment exercises with a model answer to grade yourself against, and several are the ritual arithmetic and reading-the-news drills that keep the quantitative skills sharp. Produce your full answer before reading the Solution (working-agreement 3).

P1 (guided: why public and unedited). A peer argues: "I keep a detailed private investment journal, so making my theses public would add nothing but ego risk." Give the two mechanisms by which a public, dated, unedited record does something a private journal cannot.

Solution. (1) Irreversibility defeats hindsight bias. A private file is silently editable by your own memory and your own keyboard. A year later you will "remember" having been more cautious or more prescient than you were, and you can even quietly change the file. A public, timestamped record cannot be altered without everyone seeing, so when the future arrives it grades you honestly; you are borrowing the crowd's memory to defeat your own hindsight (§4.2, M6.04). (2) The unedited record proves process, not luck. A record that includes the misses, left standing and annotated, is credible precisely because anyone can cherry-pick winners after the fact; a continuous dated record you never edited is the one thing that cannot be faked, and it is the asset every professional path is gated on (§4.6). The peer is right that a private journal captures reasoning; they are missing that externally-verified irreversibility and credibility to others are separate goods only the public record provides. (Ego risk is real and is exactly the point: the discomfort is the accountability working.)

P2 (guided: read the news like an analyst). You own an Indian IT-services company. Headline: "Q2 revenue grew 8% YoY, EBIT margin steady, management guides FY to 'high single digits.'" The stock falls 6% on the day. Explain, using the §4.4 method, why a "good" number produced a fall, and how you decide whether to act.

Solution. Read it as rate-of-change versus expectations, not as a level. (1) What was expected? The price implied faster growth than 8%: say the market and prior guidance pointed to ~11–12% constant-currency growth; 8% is a miss versus the embedded expectation, so the number is "good" in the abstract but bad relative to what was priced in, hence the fall. (2) Rate of change and second derivative? If 8% follows several quarters of decelerating growth, the market is pricing the deceleration (the second derivative), which matters more than the level; "high single digit" FY guidance may confirm a demand slowdown (discretionary spend weakening, the IT-services red flag of M5.06). (3) Does it change the thesis? Check your pre-written falsifiers: if your thesis-breakers included "constant-currency growth decelerating below ~9% with book-to-bill <1," this print may be on the list and demands full attention: pull the book-to-bill, TCV, and headcount net-adds before deciding. If your thesis was always "a cash machine at a fair price, growth secondary," and the falsifiers were about capital allocation and margins (both steady here), the print is noise and the 6% move is an opportunity to reassess price, not a reason to act reflexively. The discipline: the day's move is not the signal; the gap between the print and your pre-written falsifiers is.

P3 (guided: the January ritual arithmetic). Last January you valued a US company with cost of equity = 3.8% + 1.10 × 4.7%. This January you re-pull: risk-free 4.4%, implied ERP 4.4%, and your relevered beta is now 1.05. (a) Compute both costs of equity. (b) State the delta and the lesson.

Solution. (a) Last January: 3.8% + 1.10 × 4.7% = 3.8% + 5.17% = 8.97% ≈ 9.0%. This January: 4.4% + 1.05 × 4.4% = 4.4% + 4.62% = 9.02% ≈ 9.0%. (b) Delta ≈ +0.05%, essentially flat. The higher risk-free (+0.6%) was almost exactly offset by the lower ERP (−0.3% × beta) and the lower beta. The lesson is not the trivial net change; it is that three inputs each moved materially in offsetting directions, and only by re-pulling all three did you confirm your ~9% default is still right rather than assume it. Had risk-free risen while ERP held, the default would have moved ~+0.6%, enough to shift a valuation verdict on a long-duration name. The ritual's value is catching the years when the offsets don't cancel (§4.3, WE2).

P4 (independent: design your public record). Design your own public track record: pick a form, define your first piece, and state the three disciplines that keep the record honest. Grade against the model.

Solution (model). Form: a paper-portfolio log (public spreadsheet/blog of dated own/watch/pass calls with thesis + falsifiers + sell criteria attached) as the primary record, plus a peer group of four circulating one thesis a month for the honesty of real critique, the two lowest-friction, highest-honesty starting forms (§4.2); graduate to a Substack for long-form deep-dive theses once a quarter if going pro. First piece: the next teardown I was going to do anyway, published as-is, dated, with a "what would change my mind" line, not a polished debut (defeats the "wait until ready" trap). The three disciplines: (1) date everything, never back-date; (2) never delete or silently edit: strikethrough and append a dated correction/scorecard, leave the original visible; (3) always publish the falsifier (the specific thing that would prove me wrong) so the later grade is objective. Two guardrails: everything is analysis, not advice (working-agreement 9), and every number is traceable to a primary filing (working-agreement 7). Self-grade 2 if you named a specific form, a this-week first piece, all three disciplines, and both guardrails; 1 if you designed a record but hedged the "start now" commitment; 0 if you deferred to "when I'm ready."

P5 (independent: the mistake loop). Your journal shows three losing decisions over a year. Decision A: bought a founder-led growth company; missed that 40% of "revenue" was to a related entity; lost capital when it unwound. Decision B: bought a different founder-led name; didn't scan the pledge (which was 55%); it de-rated on a pledge-driven fall. Decision C: passed on a wonderful compounder because it "was never cheap"; it doubled. Run the loop: find the pattern and promote checklist items.

Solution. Log by cause (not outcome): A and B are both tripwire misses (M8.05 §4.2) on founder/promoter-controlled companies, a pattern, not two incidents: you under-scan governance specifically when a compelling founder narrative is present (the anchoring-plus-tripwire combination). C is a different error: the "compounder you never buy because it's never cheap" calibration/selling error (M9.03). Promote two checklist items: (1) "For any founder/promoter-controlled company, run the full related-party + pledge + auditor-tenure scan before forming any price or quality view — no exceptions, regardless of how good the story is." This makes A and B mechanically hard to repeat by forcing the missed reflex to the front. (2) "For a wonderful business at a demanding price, pre-commit an upside trigger and a starter-position rule, not only a de-rate trigger" (the WE1 upgrade). The meta-lesson: one miss is an incident; the same miss twice (A and B) is a systematic flaw and your highest-value fix, and the fix is a permanent, boring checklist line, not a resolution to "be more careful."

P6 (independent: widen the circle). You have avoided insurance your whole practice. Design the deliberate widening to bring life insurance into your circle of competence, to the program's bar, and state how you certify it.

Solution. The bar (§4.5): you can state the life-insurance business model, the specific KPIs with healthy/stress thresholds, the right valuation lens, and the red flags, which is the shape of a Phase 5 module (here M5.03). Learn it the program's way: read the anchor sources (the master reading list's insurance section; the sector cheat-sheet's insurance page), internalize the toolkit: APE, VNB and VNB margin (25–30%+ strong), EV and RoEV, persistency (13th-month >85%), solvency (IRDAI min 150%), and the lens P/EV, not P/E. Then within 48 hours tear down one clear exemplar and one troubled name, one Indian (e.g., an SBI Life / HDFC Life type) and, where comparable, a US/global life or P&C insurer for contrast (working-agreement 4 and 5). Certify it: run a cold teardown of a third, never-studied life insurer and score it against the M8.05 teardown rubric. Did you reach for P/EV and VNB margin in the first two minutes, spot a persistency slide or a ULIP-tilt margin risk, and read solvency drift toward the 150% floor? If it clears the bar, insurance is inside the circle; if not, you've found precisely what you don't yet understand and drill that. The honesty rule: a domain you feel you get but cannot tear down cold is not in the circle yet.

P7 (independent: professional paths). A friend with your training and a five-year public record asks which professional path to pursue. Their temperament: highly independent, dislikes selling and client management, has modest personal capital, wants to be judged on being right over years. Map the options and name what their record must show.

Solution. Rule out by temperament: sell-side rewards coverage, access, and relationship-selling, not being right; a poor fit for someone who dislikes selling and wants to be judged on correctness. Managing own capital alone fits the temperament perfectly but is capped by "modest personal capital": it compounds their own book but can't be a full livelihood yet. The fit: buy-side (be judged on being right over years; conviction and process rewarded) is the temperamental bullseye, and the five-year public record is exactly the entry key that a buy-side seat usually demands from a non-pedigreed candidate; alternatively independent/RIA if they'd rather own their process and can stomach building distribution and compliance (heavier client-management, which they dislike, so secondary). What the record must show (§4.6): length through a full cycle (a bull and a bear; 5 years qualifies if it spans a drawdown); process visible (dated theses, reasoning, sell decisions, mistakes owned, not just returns); honest, comparable, unedited (right benchmark, net of costs, no cherry-picking, misses left standing); and attributable to skill across names and regimes, not one lucky bet. Recommendation: pursue buy-side, leading with the public record as the credential; keep compounding own capital in parallel as both a hedge and further track-record evidence.

P8 (independent: the ethics of OPM). You've begun managing a friend's money informally. A product you can access pays you a large trailing commission and is "probably fine" for them, though a cheaper index option would suit their goal at least as well. Your own biggest position has fallen and you're tempted to have their account buy it too, to support the price. Name every ethical breach in play and the rule that governs each.

Solution. Three breaches, each against a core fiduciary duty (§4.6, Section 8.1). (1) Steering them to the commission product violates client-interest-first / no self-dealing: you may not choose a product because it pays you more when a cheaper option serves them at least as well; the trailing commission is a conflict that (2) must at minimum be disclosed, and is better avoided entirely. (3) Having their account buy your falling position to support the price is front-running / market manipulation in spirit and self-dealing. Using their capital to serve your book is a gross breach, independent of whether the stock is a good idea, and likely illegal. The governing rules: fiduciary duty means the client's interest is the only interest that governs the decision; suitability means matching the strategy to their goal (the index option, here); conflicts must be disclosed and competence bounded; and the whole thing sits inside the SEBI (RIA/PMS) or SEC/Advisers-Act framework that exists to enforce exactly these duties. The one-line test you failed: would I do this if their interest were the only thing that mattered? No — so don't. The honest move is to put them in the low-cost index option that suits their goal and take no conflicted commission.

P9 (timed: 5 minutes; keep the reverse-DCF sharp). For your published WATCH thesis, the name trades at a 1.6% FCF yield and you require 11.5% (INR). (a) What perpetual FCF growth does the price imply? (b) India's nominal GDP is ~10–11%. State the price verdict and what it means for a margin of safety.

Solution. (a) Fast Gordon inversion: implied g ≈ required return − FCF yield = 11.5% − 1.6% = 9.9% ≈ 10%. (b) ~10% implied perpetual growth sits right at India's nominal GDP, so the price already assumes the company grows in line with the whole economy forever, i.e., durable category leadership priced in as near-certainty. Verdict: demanding. A great business, but the price embeds the greatness, leaving essentially no margin of safety; edge would require the future being better than "keeps pace with GDP forever," which is a high bar for an already-large company (spine question 5; M8.05). This is why the honest call was WATCH, not OWN — the business is not the question, the price is.

P10 (independent: run one audit row honestly). Pick the one competency (C1–C11) you are genuinely weakest on. Write its audit row: the observable bar, the evidence you actually hold, the honest test, and your score (0/1/2); and, if not a 2, the specific, small remediation.

Solution (model, using C7 as an example weak spot). C7. Evaluate management & capital allocation. Bar: read incentives from proxies/board reports, judge the reinvestment/M&A/buyback record over a decade, spot misalignment and governance rot. Evidence I hold: the M3.07 capital-allocation dossier (passed); the M8.04 proxy teardown (passed). Honest test: "Handed an unfamiliar DEF 14A and ten years of capital allocation, can I judge alignment cold — or do I lean on the module's scaffolding?" I can read a proxy, but I've only done it on the two companies the modules assigned; I have not done it cold on a novel name since. Score: 1 (partial). I can do it with support, not yet cold. Remediation (specific and small): run the proxy/Board's-report + decade-of-capital-allocation read on three never-studied names over the next month (one US DEF 14A, one Indian Board's Report with RPT and pledge, one with a messy M&A record), scored against the M3.07 rubric, until a cold read clears the bar. The point of the exercise: an honest "1" with a one-month, three-rep fix is worth infinitely more than a generous "2", because the "1" tells you exactly what to do and the "2" tells you nothing (§4.7).

P11 (timed: 6 minutes; signal, noise, or falsifier). You own a US industrial (order-driven, cyclical). Classify each headline as signal to reassess, noise to note-and-ignore, or on my falsifier list, full attention, and say why. (a) "Fed holds rates, as expected." (b) "Company's book-to-bill fell to 0.85 for the second straight quarter; backlog coverage shrinking." (c) "Analyst at Bank X downgrades to Hold, cuts target 8%." (d) "Company wins largest-ever order; management raises FY revenue guide."

Solution. (a) Noise: a hold that was expected is a non-event (rate-of-change vs expectations: zero surprise); note and ignore. (b) On the falsifier list, full attention: book-to-bill <1 for several quarters with shrinking backlog coverage is precisely the industrials red flag (M5.08) and almost certainly one of your pre-written thesis-breakers; this is the ~5% of headlines that demands you pull the order-inflow, backlog quality, and working-capital trend now. (c) Noise: a broker rating change is opinion, not new fundamental information; unless the downgrade contains a fact on your falsifier list, note and ignore (and beware anchoring on their target). (d) Signal to reassess, with discipline: a record order and a guide-raise is genuinely good, but read it as rate-of-change vs expectations (was a raise already priced?) and against the cycle (a peak-cycle order surge into rising leverage is the cyclical trap, not an unalloyed positive); reassess the thesis, don't just cheer. The meta-skill: only (b) touches the pre-written falsifiers, so it gets the deep work; the discipline of inaction on (a) and (c) is what lets you follow the name closely without being jerked around (§4.4).

P12 (independent: the cadence arithmetic). The master map says the gap from week-80 core competence to world-class is "~150 more teardowns, 30 more deep dives, and a few full market seasons." At the §4.3 annual cadence, how long is that, and what does the arithmetic tell you about the nature of mastery?

Solution. At ~50 teardowns and ~10 deep dives a year, 150 teardowns and 30 deep dives is ~3 years of sustained cadence, and "a few full market seasons" (a bull, a bear, a shock) is also a multi-year, calendar-bound requirement you cannot rush. So mastery from here is roughly a 3–5 year proposition, gated by reps and by lived time, not by more study. The lesson the arithmetic delivers: mastery is not a body of knowledge you could cram. It is a quantity of honestly-recorded reps against reality plus exposure to real market regimes, both of which take years by construction (§4.1, §4.8). This is why the program "ends" at core competence and hands you a cadence rather than more modules: the remaining distance is only walkable one week, one teardown, one journaled decision at a time. It also reframes the daily loop and the weekly teardown as not "extra practice" but the actual mechanism of becoming world-class — the highest-leverage hours you have.

P13 (guided: score your own calibration). Over a year your public log carried 12 dated probabilistic calls: at 80% confidence, 6 calls of which 5 came true; at 60%, 4 calls of which 2 came true; at 40%, 2 calls of which 1 came true. (a) Compute your Brier score (mean of (p − outcome)², outcome 1 if it happened else 0). (b) Your overall hit-rate and your average stated confidence. (c) Are you well-calibrated, over-, or under-confident, and how does the Brier compare to the "always say 50%" baseline?

Solution. (a) Bucket squared-errors: 80% → 5×(0.8−1)² + 1×0.8² = 0.20 + 0.64 = 0.84; 60% → 2×(0.6−1)² + 2×0.6² = 0.32 + 0.72 = 1.04; 40% → 1×(0.4−1)² + 1×0.4² = 0.36 + 0.16 = 0.52. Total = 2.40, so Brier = 2.40 ÷ 12 = 0.20. (b) Hit-rate = (5+2+1) ÷ 12 = 8 ÷ 12 = 66.7%; average stated confidence = (0.8×6 + 0.6×4 + 0.4×2) ÷ 12 = 8.0 ÷ 12 = 66.7%. (c) Well-calibrated overall: your average confidence (66.7%) equals your realized rate (66.7%), and bucket-by-bucket the realized rates (83%, 50%, 50%) track the stated ones within small-sample noise. The Brier of 0.20 beats the 0.25 naive baseline, confirming real (if modest) forecasting signal. Contrast WE4, where an identical-looking 65% hit-rate hid overconfidence because the confidences themselves were inflated: the hit-rate tells you how often you were right; the Brier tells you whether your probabilities were honest, which is the thing the scorecard actually grades.

P14 (guided: size your cadence to your coverage). (a) You want every name on a 20-name watchlist re-torn-down every 4 months; how many teardowns a year is that? (b) You can realistically sustain only 50 a year; at that ceiling, how stale does the average name's read get? (c) Two legitimate fixes, and the principle behind the choice.

Solution. (a) 20 × (12 ÷ 4) = 20 × 3 = 60 teardowns/year. (b) At 50/year across 20 names, refreshes per name = 50 ÷ 20 = 2.5/year, so the interval is 12 ÷ 2.5 = 4.8 months. The average read runs ~4.8 months old rather than the 4 you wanted; not a crisis, but drifting. (c) Either trim the watchlist (fewer names, each fresher; e.g., 16 names × 3/yr = 48, within budget) or lift the cadence toward 60. The principle: coverage is not free. Every name you add either raises the required cadence or lengthens the staleness of every other name's read, so a watchlist is actively pruned, exactly like the information diet's guarded smallness (§4.4). Uncontrolled coverage sprawl is how the calibration on your "core" names silently rots (§4.1).

P15 (independent: is this a credible track record?). A candidate shows you a public record: 3 years long, all in a rising market; a headline +140% on one position that alone is 60% of the stated return; the losers deleted "to keep the page clean"; and a 72% hit-rate on the 11 calls that remain visible. Score it against the §4.6 credibility properties and give your verdict.

Solution. It fails on every property that matters. (1) Length through a full cycle: three years all in a bull market has never been tested by a real drawdown, so it proves nothing about risk-taking; credible records span a bull and a bear, with 5+ years the practical floor. (2) Process visible and unedited: the losers were deleted, the cardinal sin (§4.2). It destroys credibility (the misses are the credibility) and makes the 72% a survivorship-biased number computed only on the survivors, not a real hit-rate. (3) Attributable to skill, not one bet or regime: 60% of the return from a single position is an anecdote: one home run in one favorable regime, not a process. Verdict: not credible. The honest version would be five-plus years with every call left standing and dated, losers annotated with what they taught, returns stated net against the right benchmark, and the result spread across many names and both regimes. The tell that you are looking at a curated highlight reel rather than a record is precisely that it is clean — a real record never is.

P16 (independent: design your three-layer annual regime). Design your own lifelong practice regime with all three layers of §4.3 made concrete (specific numbers and dates), plus one clause on what each layer defends against. Grade against the model.

Solution (model). Layer 1, reps (defends the plumbing and grows the library): ~50 timed teardowns/year (Saturday is the protected day), each ending in an own/watch/pass and a dated public call; 8–10 deep dives/year (settling toward one a quarter) on names whose variant question is live; 4 quarterly review letters (last Sunday of each quarter), scored on reasoning. Layer 2, reading (defends the foundations from fading): one master-text re-read per quarter: Q1 a valuation/quality text, Q2 a temperament text (Graham ch. 8 & 20, or the year's Berkshire letter), Q3 a moat text, Q4 a macro/behavioral text. Layer 3, annual maintenance (defends calibration): the January ritual on a fixed dated day: re-pull ERP / risk-free / country-risk from Damodaran and re-run the WACC defaults, redo the M6.05 behavioral self-audit, and batch-read the error log to promote one or two checklist items. Self-grade 2 if every layer carries specific numbers and a protected date/day and names what it defends against; 1 if you listed activities without dates or without the "what it defends" logic; 0 if it reads as aspiration rather than a schedule you could run next week.

P17 (independent: turn a calibration finding into a fix). Your January review of last year's scorecard shows a sharp pattern: on founder-led growth stories you stated 90% confidence and were right only 60% of the time, while on everything else you were well-calibrated. Diagnose the error by cause and prescribe the fix across the mistake loop, the checklist, and position sizing.

Solution. Cause, not outcome (§4.5): a calibration miss localized to a category. A compelling founder narrative is pulling your stated confidence far above your real accuracy (anchoring on the story; §4.2, M6.04), and it is systematic, not bad luck, because it recurs across the whole founder-led bucket. The fixes: (1) Checklist line: "for founder-led growth names, run the full governance / related-party / pledge scan and cap stated confidence before assigning any probability; treat the narrative as a claim to be falsified, not as evidence." (2) Sizing (M9.01): position size should track calibrated conviction, not felt conviction, so size founder-led growth positions off the realized 60%, not the stated 90%, i.e., materially smaller until the scorecard shows the gap has closed. (3) Re-measure: next January's scorecard is the test: if the founder-led bucket's realized rate has moved toward its stated rate, the checklist line stays; if not, the fix was wrong and you iterate. The meta-point: a calibration finding is worthless until it changes a behavior (a checklist line and a sizing rule), and the following year's Brier score is how you know it did.

P18 (timed: 7 minutes; keep the lifelong quantitative reflexes sharp). (i) Brier score of three calls: you said 70%, 90%, 40%; the outcomes were happened, happened, did-not (±0.01). (ii) A 15-name watchlist refreshed every 3 months needs how many teardowns/year? (iii) 18 of 30 dated calls played out: hit-rate? (iv) A name trades at a 2.0% FCF yield and you require 10%: implied perpetual growth (±0.2)? (v) You can run C6 (sector analysis) cold on the sectors the modules assigned but not on a novel one: what audit score, and why?

Solution. (i) [(0.7−1)² + (0.9−1)² + (0.4−0)²] ÷ 3 = (0.09 + 0.01 + 0.16) ÷ 3 = 0.26 ÷ 3 = 0.087. (ii) 15 × (12 ÷ 3) = 15 × 4 = 60/year. (iii) 18 ÷ 30 = 60%. Remember it is credible as skill only if it beats the base rate across many names and a full cycle (§4.6). (iv) fast Gordon inversion: g ≈ required − FCF yield = 10% − 2.0% = 8.0%. (v) 1 (partial): "with support or on easy cases but not cold on a novel case" is the definition of a 1 (§4.7); the honest remediation is three cold sector reads on never-studied names scored against the M8.05 rubric. Scoring: 5/5 in time = the reflexes are lifelong-ready; ≤3 = re-run WE4–WE5 and P13–P14 before you rely on them in the field.

P19 (independent — convert your own memo into the four professional formats). Take the best one-page memo you have written in this program, on a company you still hold or still watch. Without doing any new analysis, produce four artefacts from it and log how long each took. (a) The morning-meeting call: ninety seconds of speech, which is roughly 200 words, ending in what a client should do. (b) The investment-committee memo: the same one page plus two additions the committee will not proceed without. (c) The initiation skeleton: a section list with one action title per section, in the order a reader meets them, and the two exhibits that carry the argument. (d) The update note: assume the company has just reported a quarter with revenue 4% below your forecast and margin 60 basis points above it, and write the four-part note. Then answer the question the exercise exists for: which of the four was hardest, and what does that tell you about which seat suits you?

Solution (model shape, not a key). (a) Lead with the recommendation and the horizon, one sentence of what the business is, the variant perception with a number, one support, the falsifier, and the action. If it runs past 200 words the memo's governing thought was not sharp enough (M10.03 §4.2), because compression is a test of the argument rather than of the writing. (b) The two additions are a proposed position size with its reasoning, which is M9.01's conviction and downside-floor argument written out, and pre-committed sell criteria, which is M9.03's five reasons made specific to this name with numbers attached. Nothing else changes; a committee memo that adds pages instead of adding those two things has misread the room. (c) The initiation skeleton is the report structure of M10.03 §4.6 with action titles substituted for topic labels, and the flip test applies: read only the titles and the argument must stand. Two exhibits, not nine. (d) The four parts are what changed, what it does to the numbers, whether the thesis moves, and what we do. The trap is the third: a 4% revenue miss with a 60-basis-point margin beat may leave your earnings forecast almost unmoved and your thesis untouched, and saying so plainly is a better note than manufacturing a revision. On the closing question, the honest reading is usually about tempo rather than skill. If (a) was hardest you are uncomfortable compressing, which the sell side punishes daily. If (d) was hardest you are uncomfortable committing to an action on partial information, which the buy side punishes weekly. If (b) was hardest you have a thesis but not a position, which is the most common gap in this program's graduates and the one §4.6's paths make expensive.

Applied mini-project: THE FINAL COMPETENCY AUDIT (the M11 milestone)

This is not a mini-project; it is the final gate of the entire program: the honest, evidenced self-assessment against all eleven competencies that records M11: core competence, all eleven at bar, plus the re-grading ritual that closes the eighteen months. Budget the bulk of this module's remaining hours here, and run it with the integrity the whole program has demanded: the audit's only value is its honesty (§4.7).

Part 1: Assemble the evidence. Before you score anything, gather the artifacts. For each competency you should be able to put a concrete, gradeable object on the table: a graded capstone, a passed gate at its threshold, a built dashboard, a filled journal, an app mastery reading. If you cannot find the evidence for a competency, that is itself a finding: you may have passed a gate but not retained the capability, which scores at most a 1. Lay all eleven competencies' evidence out at once, so the audit is a portfolio review, not eleven disconnected quizzes.

Part 2: Score all eleven rows. Using the §4.7 table (bar · evidence · honest test), score each competency 2 / 1 / 0 against the rubric below, as a hostile examiner with the evidence in front of you. Write, for each, one sentence of justification tied to the specific evidence and one sentence answering the honest test. Effort is not a criterion; demonstrable, cold, evidenced capability is (Common mistake 10).

The final-audit rubric (score each of the eleven competencies)

The pass standard (the M11 gate). Core competence, the M11 milestone, requires all eleven competencies at 2 (at bar). This is deliberately strict: the master map's definition of done is "every competency in Section 1 demonstrated at its stated bar" (Section 0.2), and the program's whole value proposition is that all eleven are real, not that most are. A single competency at 1 means the audit does not pass yet, and that is not a failure of the program but its integrity working: you have found, precisely, the one capability still to close, and the milestone checkpoints (master map Section 4.3) are explicit that "if you fail a 'prove it,' stay in that phase." Any competency scored 0 or 1 gets a specific, small, dated remediation (the WE3 / P10 model: not "study more" but "run three cold proxy reads this month, scored against the M3.07 rubric") and a return to the relevant practice until the evidence is genuine. The gates do not bend; the calendar does.

Part 3: The re-grading ritual (break the seal). Reopen the naive one-pagers you sealed in week 3 (M0.05): the Indian AR and the US 10-K. Re-grade them with today's eyes: mark every missed profit engine, every reported-vs-real-profit confusion, every absent or false moat call, every ignored balance sheet, every "no idea what the price implied"; list every concept you now command that you lacked a word for then; and note the one thing your naive self got right by instinct. Then write the "how far" note, one page comparing the analyst who wrote those pages to the one reading them now (§4.8). This is a graded deliverable of the milestone and the program's emotional close; do it slowly. Re-seal the originals and the note together to reopen at a future milestone.

Part 4: Commit the lifelong system to writing. Produce a one-page lifelong-practice plan: your chosen public-record form and where it lives; your annual cadence targets (teardowns, deep dives, quarterly letters) and the weekday you protect for them; your January-ritual date and checklist; your information diet (the specific sources); your next two deliberate circle-widenings; and, if relevant, your professional-path intention and what your record must still show. This plan is the artifact the module produces; it is the instruction manual for your forever.

Scoring rubric for the mini-project deliverable (criteria × 0/1/2)

Pass bar for the mini-project: ≥10/12, with the "'Not yet' handled with integrity" criterion ≥1 mandatory, because an inflated audit is worse than a failed one. The M11 milestone itself is recorded only when all eleven competencies score 2; a mini-project can pass (you ran the audit with integrity) while the milestone waits on a remediation, and that is the honest, intended outcome for anyone with a genuine gap. What passing unlocks is not another module; it is the rest of your practice, run on the system you just committed to writing.

Reading & resources

The definitive resource for this module, and for the decades after it, is the program's own master reading/resource list: artifacts/master-reading-list.md, the comprehensive, categorized, phase-cross-referenced catalogue of every book, free course, tool, dataset, primary-source portal, podcast, and ongoing free letter the program relies on, marked [Free]/[Paid] and [Beginner]/[Intermediate]/[Advanced]. It is built to be the "use these forever, without AI" master reference; this module is where you formally adopt it. The specific resources this module leans on:

On practice, mastery, and the compounding of skill.

• Gautam Baid, The Joys of Compounding (2020). The Indian synthesis of lifelong learning, temperament, the personal knowledge system, and continuous self-improvement; the single best companion to this module's ethos. The Phase 10 reading-spine text. [Paid] [Beginner]
• Anders Ericsson & Robert Pool, Peak (2016). Deliberate practice, why comfortable reps plateau, and why the mistake loop and feedback matter; the science under §4.1 and §4.5 (from M8.05). [Paid] [Beginner]
• Robert Greene, Mastery (2012), or Josh Waitzkin, The Art of Learning (2007). Either on the long arc from competence to mastery and the discipline of staying a student; optional, motivational. [Paid] [Beginner]

On the public record, calibration, and the mistake loop.

• Annie Duke, Thinking in Bets (2018) and Philip Tetlock & Dan Gardner, Superforecasting (2015). Process-over-outcome, calibration, and the discipline of grading reasoning not results; the backbone of the public-record scorecard (§4.2) and the quarterly letter (§4.3). [Paid] [Intermediate]
• Atul Gawande, The Checklist Manifesto (2009). Why a boring, always-run checklist beats expert memory; the theory behind the mistake loop's "promote the pattern into a checklist item" (§4.5). [Paid] [Beginner]

The ongoing free letters and sources (your syllabus for life, §4.4).

• Berkshire Hathaway annual letters (Buffett). Business quality, capital allocation, temperament; decades free at berkshirehathaway.com. [Free] [Beginner]
• Oaktree memos (Howard Marks). The best free writing on cycles, risk, and sentiment; free at oaktreecapital.com. [Free] [Intermediate]
• Michael Mauboussin (Counterpoint Global / Morgan Stanley): Measuring the Moat, The Base Rate Book, and the periodic ROIC/expectations pieces; free. [Free] [Advanced]
• The Nomad Investment Partnership letters (Nick Sleep). Long-horizon thinking and scale-economies-shared; free archived collection. [Free] [Intermediate]
• India: Sanjay Bakshi (Fundoo Professor / BFBV) lecture notes and blog, and Safal Niveshak (Vishal Khandelwal). The best free India-specific writing on temperament, valuation, and the promoter/governance lens. [Free] [Intermediate]
• Aswath Damodaran. His blog (Musings on Markets) and the data library at pages.stern.nyu.edu/~adamodar, refreshed each January: the ERP, risk-free, country-risk, beta, and sector benchmarks the January ritual re-pulls (§4.3). [Free] [Intermediate–Advanced]
• Podcast rotation: Invest Like the Best, Business Breakdowns, Acquired (business teardowns aloud, for calibration), Odd Lots (macro/plumbing). Two or three episodes a week, chosen by topic. [Free] [Beginner]

On going professional and the ethics of OPM.

• CFA Institute, Code of Ethics and Standards of Professional Conduct (freely summarized). The profession's fiduciary standard; the concrete form of §4.6's duties. [Free] [Intermediate]
• SEBI (Investment Adviser, PMS, and AIF regulations) and the US SEC / Investment Advisers Act overviews: the regulatory frameworks that enforce the OPM duties, India and US. Read the plain-language investor/adviser guides on sebi.gov.in and sec.gov. [Free] [Intermediate]

Primary sources (free, forever, and the only ones you truly need). SEC EDGAR (10-K/DEF 14A/transcripts); BSE/NSE company pages, screener.in, and Tijori (Indian annual reports, shareholding/pledge, ten-year panels); RBI, MOSPI, and FRED (macro); company investor-relations pages (decks, call transcripts). Every teardown, deep dive, and January ritual for the rest of your life is pulled from here — no secondary summary, and no AI, required (working-agreement 7). [Free] [Beginner–Advanced]

Going deeper (this cycle's additions).

• Going deeper: Good Judgment Open (gjopen.com) [Free] [Intermediate]. A live forecasting platform where you post dated, probabilistic predictions and are Brier-scored against a crowd over time; the hands-on way to build and measure the calibration the WE4 scorecard formalizes, before your own money is the test (§4.2, M9.04).
• Going deeper: Michael Mauboussin, The Success Equation (2012) [Paid] [Intermediate]. The definitive treatment of untangling skill from luck in outcomes: base rates, the paradox of skill, and why a track record needs breadth and a full cycle to signal skill (the §4.6 / WE5 / P15 point) rather than one fortunate bet.

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. That is 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; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "The lifelong system — a public dated unedited record, the annual cadence of ~50 teardowns / 8-10 deep dives / 4 quarterly letters, the January ritual, a small high-signal information diet read as rate-of-change against expectations, deliberate circle-widening with cold certification, the journal-to-checklist mistake loop, and the honest eleven-competency final audit",
  "use": "This is the last module in the program, so be exact about where a tool belongs in a practice you will run for decades. It belongs on the CLERICAL half of every loop and nowhere near the scoring. (1) THE JANUARY RITUAL'S DATA PULL: §4.3 asks you to re-pull the implied ERP, the currency-matched risk-free rates and the India country-risk premium and re-run your WACC defaults. That is a scheduled, deterministic fetch-and-recompute, and once it is a script you actually run it every January instead of the years you remember to — which is the entire difference between a ritual and an intention. (2) THE JOURNAL AS A DATASET: your entries are unstructured text with a date, a probability, a falsifier list and, later, an outcome. Have a tool normalise them into rows so the scorecard is computed rather than assembled by hand, and have it flag entries missing a stated probability or a falsifier, because those are the entries that will be unscoreable in three years. (3) THE INFORMATION-DIET TRIAGE: §4.4's third question — does this change the thesis, or is it noise? — is a matching problem against a list you have already written. Feed it your pre-written falsifiers per position and have it route each headline to 'on the falsifier list' or 'not', with the matching clause quoted, and then read the routed ones yourself. Note what this does: it enforces inaction on the 95%, which is the discipline you are worst at, by making the default 'noise' rather than 'have a look.' (4) THE CIRCLE-WIDENING READING PLAN and the hostile-examiner rehearsal for §4.7's honest test — ask it to attack your claimed evidence for a competency, which is uncomfortable in exactly the way the audit needs. What it must never touch: the SCORE. The 2/1/0 on each of the eleven rows is a judgement about yourself made against evidence, and outsourcing it defeats the only property the audit has, which is honesty. A model asked whether you are at bar will say something agreeable, because that is what it does with an underdetermined question about the person asking.",
  "tools": ["Coding assistants — the January-ritual fetch-and-recompute script, the journal-to-rows normaliser, and the calibration scorecard below", "Chat assistants — the falsifier-matching triage on the news, the circle-widening reading plan, and the hostile-examiner rehearsal of your audit rows", "Document/PDF AI — the deep-dive reading load, and the SPIVA persistence scorecards and fund factsheets when you are pricing the professional paths of §4.6"],
  "prompt": "Here is my decision journal as raw entries: <PASTE>. Normalise it to one row per call with these columns and nothing inferred: date, instrument or question, the claim exactly as I wrote it, my stated probability (blank if I did not state one), my falsifier list, the resolution date, the outcome (1/0/unresolved), and the entry's own text for the outcome. Then give me three lists: entries with no stated probability, entries with no falsifier, and entries whose stated resolution date has passed with no outcome recorded. Do not score me, do not compute a Brier score, and do not tell me whether I was well calibrated — I am doing that, and the three lists are the deliverable, because they are the entries that will be worthless to me in five years.",
  "verify": "Two checks, and the second is the one the whole module turns on. (1) THE RITUAL SCRIPT IS RE-DERIVED, NOT TRUSTED: re-pull one input by hand each January and confirm the script's value against it, because a fetch-and-recompute that silently reads a stale cached page produces exactly the frozen-assumption failure of common mistake 6 while looking like diligence. Read what the script tells you back — a cost of equity you did not type is the number worth staring at. (2) THE SCORECARD'S SAMPLE SIZE IS STATED IN THE SAME SENTENCE AS THE SCORE: the quantitative lens below computes how many dated calls a Brier score needs before it means anything, and the answer is in the hundreds. A calibration verdict on one year of entries is a point estimate with an interval wide enough to contain 'perfectly calibrated' and 'badly overconfident' at once. Report the interval; act on the point estimate only where the action is cheap and reversible, which — as WE4's checklist line happens to be — is often.",
  "diy": "The gate is unaided, and for this module the gate is the rest of your life. You run the January ritual by hand at least once, so you know what the script does; you compute a Brier score and a calibration table with a pen when the year's log is twenty lines; you read a headline against your own falsifier list without a tool routing it; you tear down a never-studied company cold to certify a circle-widening; and you score all eleven audit rows yourself, as a hostile examiner, with the evidence on the table. No tool in this callout is admissible in the audit room, and the honest 'not yet' is the one output nothing can generate for you."
}

Modern Data Analysis

By hand first. You built a scorecard on twenty dated calls and read it bucket by bucket; you computed a Brier score of 0.23 and compared it to the 0.25 that mechanical fifty-percenting scores; you found overconfidence hiding behind a 65% hit-rate and converted it into a checklist line; you sized a teardown cadence to a watchlist and watched it go stale as coverage sprawled; you inverted a FCF yield to an implied growth rate in one line. All of that is pen work and stays pen work — a calibration table you cannot draw by hand is a calibration table you will not trust when it says something you do not want to hear.

Today's workflow. The change is not that a computer scores your calls faster. It is this: a decision journal is a dataset, and once you treat it as one, three questions become answerable that the page cannot settle: what your Brier score is actually made of, how far from meaningless it is at your current sample size, and which of your errors are calibration errors (slow to diagnose, hundreds of observations) versus process errors (diagnosable at n = 1). Concretely:

• A schema, fixed the day you start. Date, question, claim as written, stated probability, falsifier list, resolution date, outcome, and the entry that recorded the outcome. §4.2's three disciplines (date it, never edit it, publish the falsifier) are exactly the columns that make the log scoreable later, which is a second reason for them the module does not state: the falsifier makes the grade objective, and it is also what makes the row machine-resolvable rather than a matter of later opinion.
• The Murphy decomposition, not just the Brier. A Brier score splits exactly into reliability − resolution + uncertainty: how far your stated probabilities sit from the realised rates in their own buckets, how far those bucket rates sit from the overall base rate, and how hard the questions were. Reporting the three separately is the difference between "my score got worse" and "my questions got harder," which are opposite findings that a single number cannot distinguish.
• An interval attached to every score. The lens below computes it. The headline is that a year's log is roughly a quarter of the way to a Brier score you could defend.

Tools & sources (IN + US). pandas for the journal table and the bucket aggregation, numpy for the scoring and the sampling arithmetic, matplotlib for the calibration curve (stated probability on one axis, realised rate on the other, with bucket counts printed on each point so the reader can see how thin the evidence is). Sources: skill persistence, both markets. The primary evidence on whether performance persists is published free, twice a year, for both jurisdictions: S&P Dow Jones Indices' SPIVA Scorecards and (the document that actually answers the persistence question) the accompanying Persistence Scorecards, published separately for the US and for India (spglobal.com/spdji → SPIVA). Read the transition matrices yourself and quote your own numbers with the horizon stated; do not quote a persistence figure from memory, including one you saw in a book, because these tables are re-cut every period and a number without its window and its universe is not a finding. Sources: calibration. Good Judgment Open (gjopen.com, free, already on this module's reading list) is the only place most analysts will ever accumulate a few hundred resolved, Brier-scored, dated forecasts against a crowd, which, per the lens below, is roughly the sample size the instrument needs; the underlying literature is the Good Judgment Project work behind Superforecasting, also on the list. Sources: the January ritual, both currencies. Damodaran's data library (pages.stern.nyu.edu/~adamodar) for the implied ERP, the country-risk premium and the industry betas; FRED (DGS10 for the US risk-free, GDP for the nominal-growth cap) for the US side; CCIL and FBIL for Indian G-sec and benchmark yields, RBI's database and MOSPI for the Indian macro inputs, and NSE/BSE plus screener.in for the panel your defaults are re-run on. Sources: the primary record itself. EDGAR (10-K, DEF 14A) and the NSE/BSE company pages and Indian annual reports, which §4.4 correctly says are the only sources you truly need and which will still be free and still be there in twenty years.

# The decision journal as a dataset: Brier, its exact decomposition, and how big n must be — R3
def score(rows):                      # rows: (stated_p, n_calls, n_that_happened)
    N    = sum(n for _, n, _ in rows)
    obar = sum(k for _, _, k in rows) / N                     # base rate of the QUESTION SET
    B    = sum(k*(p-1)**2 + (n-k)*p**2 for p, n, k in rows) / N
    rel  = sum(n*(p - k/n)**2    for p, n, k in rows) / N      # miscalibration
    res  = sum(n*(k/n - obar)**2 for p, n, k in rows) / N      # discrimination
    return B, rel, res, obar*(1-obar), obar
WE4 = [(0.9, 8, 6), (0.7, 8, 5), (0.5, 4, 2)]                  # the module's own year of 20 calls
B, rel, res, unc, obar = score(WE4)
print(round(B,5), round(rel,5), round(res,5), round(unc,5), round(rel-res+unc,5))
#     0.23      0.01125    0.00875    0.2275   -> 0.23  decomposition rebuilds WE4's own number

# How many calls before the score means anything. Under PERFECT calibration a call stated at p
# contributes a squared error with mean p(1-p) and variance p(1-p)(1-2p)^2.
from math import ceil
var = sum(n*p*(1-p)*(1-2*p)**2 for p, n, _ in WE4) / sum(n for _, n, _ in WE4)   # 0.03648
for hw in (0.05, 0.03, 0.02, 0.01):
    n = ceil(var * (1.96/hw)**2)
    print(f"+/-{hw}: {n:5d} calls = {n/50:4.1f} years at this module's ~50-a-year cadence")
#   +/-0.05:  57 = 1.1 yr | +/-0.03: 156 = 3.1 | +/-0.02: 351 = 7.0 | +/-0.01: 1402 = 28.0

Verify. Prove the engine against the module's own worked figures before it scores a single real entry. WE4: bucket contributions of 1.68 / 1.92 / 1.00, a total of 4.60, a Brier of 0.2300, a hit-rate of 65.00% and an average stated confidence of 74.00%. All reproduced, and the decomposition independently rebuilds the same 0.23 from three numbers WE4 never computes. P13: 0.84 / 1.04 / 0.52, total 2.40, Brier 0.2000, hit-rate and average stated confidence both 66.67%, the equality that makes it "well calibrated overall." P18(i): 0.0867. WE5's cadence arithmetic: 24 × 2 = 48 teardowns, 50 ÷ 40 = 1.25 refreshes a year and therefore a 9.6-month-old average read, 3 × 50 = 150 and 3 × 10 = 30 over three years, and the five-year book's 12.47% CAGR against an 11% benchmark (+1.47 points). P14: 60 teardowns a year and a 4.8-month interval at a 50-a-year ceiling. The January-ritual builds: WE2's 9.475% this January against 9.42% last (a +0.055 point delta) and the INR cost of equity of 12.92%; P3's 8.97% → 9.02%, a +0.05 point delta. The Gordon inversions: WE1's 12.0 − 1.3 = 10.7%, P9's 11.5 − 1.6 = 9.9%, P18(iv)'s 10.0 − 2.0 = 8.0%. Every one reproduces; nothing in this module's arithmetic is wrong. One convention to hold straight when you code the scorecard: the Brier's uncertainty term is a property of your question set, not of you, so two analysts' Brier scores are only comparable if they answered the same questions — which is almost never true and is the single commonest misreading of the number.

Quantitative lens

The module makes the right argument (grade the reasoning, not the P&L, and put a number on it) and then, like almost everyone who reaches this point, it stops one step short of asking what the number can support. Three computations finish it. All of them are run on the module's own figures, so nothing below is a claim about the world; it is a claim about what these instruments can and cannot establish.

One: your Brier score is mostly a property of your questions. Decompose WE4's twenty calls: reliability 0.01125 − resolution 0.00875 + uncertainty 0.22750 = 0.23000, which reconstructs the module's headline exactly. Look at the shares. Uncertainty, the difficulty of the question set, is 98.9% of the score. Miscalibration contributes 0.011 and discrimination claws back 0.009. Two consequences follow immediately. First, comparing Brier scores across different question sets is meaningless, because you are mostly comparing how hard the questions were; the decomposition travels, the score does not. Second, and sharper than the module's own reading: the right constant-forecast baseline is not 0.25. Saying "50%" to everything scores 0.25 regardless of the questions, but saying the question set's own base rate to everything scores exactly the uncertainty term, here 0.2275. Against that baseline WE4's 0.23 is 0.0025 worse, not better. The module's verdict (overconfident, damage concentrated in the 90% bucket) is right and this makes it slightly harsher. And it separates two examples whose scores sit only 0.03 apart but which are different in kind: P13 decomposes to 0.00556 − 0.02778 + 0.22222 = 0.20000, and its resolution is 3.2× WE4's: that forecaster is not merely honest about their probabilities, they actually discriminate between what will and will not happen, and they beat their own base-rate baseline of 0.2222 by 0.0222. Calibration is the cheaper virtue; resolution is the one that means you know something.

Two, and neither example has remotely enough data to say any of that. Take WE4's exact mix of stated confidences and suppose the forecaster is perfectly calibrated. The expected Brier is then 0.1700, and the per-call variance of the squared error is 0.03648, so at n = 20 the standard error is 0.0427 and a 95% band runs 0.086 to 0.254. WE4's observed 0.23 sits comfortably inside it. Go to the bucket the module indicts: 6 of 8 at stated 90% is a realised 75%, and the Wilson 95% interval on 6/8 runs 40.9% to 92.9%, which contains 90%. On eight observations, "badly overconfident" and "perfectly calibrated" are the same data. That is not a criticism of WE4, whose arithmetic is exact and whose point estimate is the best available; it is the missing sentence. And it does not overturn WE4's response, which is worth defending: the prescribed action is a checklist line capping stated confidence at 80%, and capping your confidence is cheap, reversible and asymmetric in payoff, a rational response to weak evidence, because acting on a point estimate and believing it are different things. What the arithmetic forbids is the other use: telling anyone, including yourself, that you have measured your calibration. Here is what measuring it costs, at this module's cadence of about fifty dated calls a year: ±0.05 needs 57 calls (1.1 years); ±0.03 needs 156 (3.1 years); ±0.02 needs 351 (7.0 years); ±0.01 needs 1,402 (28.0 years). Bucket by bucket it is worse, because each bucket holds a fraction of the calls: to establish at 80% power that your stated-90% calls are really 75% takes 40 calls in that bucket alone; to catch a subtler 82% takes 117; to catch 85% takes 270. The honest instruction that follows is not "don't score". It is score every year, report the interval, and do not draw a conclusion from it until year five. Meanwhile the journal's other output is available immediately, and this is why §4.5's mistake loop is the higher-yield half of the same notebook: a calibration error needs hundreds of observations to detect, and a process error needs one. "I anchored on management's narrative and skipped the reverse-DCF" is diagnosable the day you write it down. Spend the journal's first five years on process errors and let the calibration curve accumulate underneath.

Three: the same arithmetic explains why §4.6 is right about track records, for a reason it does not give. §4.6 says five years is the practical floor and that the credential is the process-visible log rather than the return number. Both are correct, and the second is correct because the return number cannot carry the inference at any realistic career length. Detecting a genuine information ratio of 0.5 from annual observations, at 80% power and 5% two-sided, takes 31.4 years; an IR of 0.3 takes 87.2 years; even an IR of 1.0, which very few sustain, takes 7.8 years. Turn it around: five years of annual returns can only establish an IR of about 1.25 or better, which is to say it can only detect skill so large that its absence would have been obvious anyway. So the five-year floor is a floor on the record's content (does it span a drawdown, are the misses standing, is the reasoning visible) and emphatically not on its statistical power. The module's own WE5 sits right on this edge and sits there honestly: 26 of 40 dated calls is a 65% hit-rate, an exact one-sided binomial gives p = 0.040 against a coin, and the Wilson 95% interval runs 49.5% to 77.9%, a lower bound of a coin flip. It clears the bar, barely, at forty calls; which is exactly the module's point that breadth across names and regimes is what makes a record signal rather than anecdote, now with the arithmetic attached. On the related question of whether anyone's skill persists, the discipline of this addendum applies to itself: read the SPIVA Persistence Scorecards for the US and for India, quote the horizon and the universe, and do not accept a persistence figure from anyone, this page included, that arrives without them.

Four, and this is the corpus's closing argument, so it should be stated at its sharpest and no sharper. The standing position of this whole modernization layer (R10) is that the tools are an amplifier. The precise version is that they have collapsed the cost of two things and barely moved the cost of a third. Retrieval is transformed: the note is found, the ten-year panel is assembled, the eight annual reports are read in the time it used to take to order them. Drafting is transformed: the memo, the letter, the exhibit caption. Judgement under uncertainty is not, and judgement under uncertainty is what this entire path has been training: the reverse-DCF's fifth step, the moat that has to be proved in the ROIC series, the governance cluster that is evidence and not a verdict, the position size that tracks calibrated rather than felt conviction. Notice why the argument about this is so loud, because the reason is in the arithmetic above and not in anyone's temperament: retrieval and drafting are verifiable at n = 1 (the citation is in the filing or it is not, the draft exists or it does not) while a change in judgement quality is only observable at n in the hundreds. One side of the debate gets its evidence the same afternoon; the other side's evidence takes a decade to accumulate. That asymmetry guarantees that the confident claims will run far ahead of the measurements, in both directions.

So state the falsifier, which is what this program has asked you to do with every other claim. The amplifier position stops being right the day a tool's dated, pre-outcome probability forecasts beat a trained analyst's on the same question set, over enough resolved calls to clear the intervals computed above, decomposed so that the gain is visible as resolution and not as reliability, and on questions whose answers were not available when the model was built. Every clause is load-bearing. A tool that scores well on questions whose answers it has already read has demonstrated retrieval, which was never in dispute. A tool that improves its Brier by lowering reliability has learned to hedge, which improves the score and reduces the information. Only a rise in resolution is a rise in judgement. A paired design (you and the tool answering the same questions) needs roughly (1 − ρ) of the sample sizes above, where ρ is the correlation between the two forecasters' squared errors; I have not measured ρ and have not seen it published, so I give you the factor and not a number. And then the point worth ending on: the instrument that would settle this is the one this module has just spent twenty hours telling you to build. A decision journal with your dated probability and the tool's dated probability on the same question, resolved on a stated date, scored with the decomposition above. Nobody arguing about this publicly has that dataset. You can have it in five years — which is less time than the argument has already been running, and rather more useful than another opinion about it.

Honest limits, and they are load-bearing because everything above is a statement about inference, which is the easiest thing in this document to overstate. The sampling arithmetic treats calls as independent draws, and they are not: theses on correlated names inside one market regime resolve together, so the effective sample is smaller than the row count and every interval above is optimistically narrow; the true requirement is worse than 351 calls, by a factor nobody can give you without your own dependence structure. The perfect-calibration variance is computed on WE4's particular confidence mix, and a different mix gives a different answer in a direction worth knowing: a forecaster who states 50% on everything has zero variance and a Brier pinned exactly at 0.25 (total precision, no information), so a tighter interval is not by itself a better forecaster. The decomposition is exact for the binned table as given, but the binning is a choice, and fewer, wider buckets flatter reliability by averaging away disagreement inside a bucket; report your bucket counts next to your curve. The information-ratio power calculation assumes normal, independent annual excess returns, which is wrong in both tails and ignores that an IR is itself an estimate with its own interval. None of these figures is a measurement of any real analyst, any real fund, or any real tool: they are what a stated number can and cannot support, which is a narrower and more durable claim than the one people usually want. And the deepest limit is the one that closes the program rather than the section: none of this arithmetic tells you whether the next judgement is right. It tells you how long you must wait, and how honestly you must record, before you are entitled to an opinion about your own judgement at all. That waiting, recorded, is the practice.

Do it in code: fix the journal schema on day one and never let an entry through without a stated probability, a falsifier and a resolution date; compute the Brier and its three components, never the score alone; print the sample size and the interval in the same line as the score; keep the calibration curve with bucket counts on the points; score annually and conclude at year five; run the January ritual as a script you re-derive one input of by hand; and if you want to know what the tools do to judgement, log your call and theirs, dated, and wait. See G1/QM1.01, G1/QM1.02 and G5/AI0.07.

Where this goes next: galaxy cross-links

• G1 · QM1.01 Probability, Distributions & Estimation. The sampling distributions, confidence intervals and small-sample honesty behind every number in the lens above; the module that turns "my Brier is 0.23" into "my Brier is 0.23 ± 0.08, which is not yet a finding."
• G1 · QM1.02 Hypothesis Testing & Regression. Power, sample size and the difference between failing to detect an effect and establishing its absence, which is the exact distinction the five-year track record and the twenty-call scorecard both turn on.
• G2 · QD2.04 Reproducible Research & Deployment. The January ritual as a scheduled job whose every figure regenerates from its sources years later, and the journal as a versioned, append-only store; this is what makes §4.2's "never delete or silently edit" an engineering property rather than a promise.
• G3 · CN1.02 Hypothesis-Driven Problem Solving. The discipline that a dated thesis with a falsifier is a hypothesis, and that the quarterly letter is the write-up of a test you designed in advance rather than a story assembled afterwards.
• G5 · AI0.06 Failure Modes & Guardrails. Where a tool is silently wrong and where it is loudly wrong, and why the failure modes that matter for a lifelong practice are the quiet ones: the stale cached input in your ritual script, the fabricated citation in your deep dive, the agreeable answer to a question about yourself.
• G5 · AI0.07 The AI-Augmented Analyst. The galaxy this addendum's last argument points at: what a working analyst actually delegates, what they keep, and how to hold the line between them as the tools change. It is the right place to go next, because the boundary drawn here is a claim to be tested against your own record, not a conclusion to be adopted.

Teach it back & journal

Feynman prompt. Write a one-page letter to a friend who has just finished a serious course of study in anything (music, medicine, coding) titled "Why finishing is the beginning." In plain language and with no finance jargon, teach them: (1) that a hard-won skill decays if you stop using it, silently, so the graduation is a start line, not a finish; (2) the single most useful habit: keep a public, dated, honest record of your work, including the parts you got wrong, because you cannot lie to a timestamp and the honest record is the only credential that can't be faked; (3) the idea of a small, high-quality information diet read for what's changed versus what everyone expected, instead of drowning in a firehose of noise; and (4) that mastery is a practice you run for life because the field keeps moving and your library of experience keeps compounding. End with the one sentence you would want them to remember. Rule: if you cannot make a non-finance friend want to keep practicing their own craft after reading it, you have not yet understood why this module is the capstone rather than an epilogue.

Journal reflection: the closing entry (the most important of the program). Write this the day you finish the audit, and date it; it is the entry you will re-read at your first anniversary and beyond. Record: (1) your eleven audit scores, competency by competency, and, named without flinching, every row that scored below 2, with its specific, dated remediation; (2) the "how far" comparison between the analyst who sealed the week-3 one-pagers and the one who just re-graded them: the single biggest change you see in your own thinking; (3) your lifelong-practice commitment, made concrete: the public-record form you will start this week and its first piece, the weekday you protect for the teardown, your January-ritual date, your information diet, and your next two deliberate circle-widenings; and (4) the honest answer to the question the whole program was really asking: do you now understand businesses well enough to keep learning on your own, for the rest of your life, without this program and without any AI? If the answer is yes, then that is the thing the eighteen months were for, not any milestone badge. Re-seal this entry with the week-3 one-pagers, to reopen at a future milestone. The program is over; the practice is yours.

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