The month you learn to separate a good decision from a good result — and to hand patients risk they can actually understand.
Last month you sat with Brás Cubas, Macbeth, and Dostoevsky and hunted for self-deception. Two quick recalls before we move on.
You already know the feeling. A case goes wrong — a CSF leak, a recurrence you didn't expect this soon, a research read that turns out to be underpowered — and the mind runs one loop: I should have known. I should have done something different. Sometimes that's true. Often it isn't.
Month 04 taught you to catch the ego narrating itself as noble. This month teaches the companion skill: catching the mind conflating a bad outcome with a bad decision. They are not the same thing, and confusing them wrecks two things at once — your honest post-case review, and your willingness to decide boldly under real uncertainty next time. A surgeon who can't tell the difference either becomes reckless (good outcomes reinforce sloppy indications) or paralyzed (one bad outcome triggers permanent over-caution).
The same confusion shows up in research. A well-designed pilot that fails to reach significance is not a bad study. A p<0.05 finding from a badly-designed retrospective series is not a good one. And in the clinic, every risk number you give a patient — recurrence rate, CSF leak rate, hearing preservation odds — either clarifies their decision or quietly frightens them into one they don't understand. This month gives you three tools for exactly this: naming your own overconfidence (Kahneman), communicating risk so patients can actually use it (Gigerenzer), and reviewing decisions on their merits instead of their results (Duke).
It also touches Brazil institution-building directly: a new partnership, hire, or clinic expansion carries real uncertainty either way, and grading the choice by whether the first quarter looks good is a fast way to punish sound long-term bets and reward lucky short-term ones.
This also touches the PhD directly. A defense committee, a reviewer, a co-author disagreeing with your interpretation — none of that is graded on whether the underlying biology cooperated. It's graded on whether the design, the analysis, and the claims were sound given the data you had. The habit of separating decision from outcome is the same habit a rigorous methodologist has already internalized; this month gives you the vocabulary to apply it to your own choices, not just to a paper you're reviewing.
This is not a book to finish cover to cover in a month — it's a reference you return to. Read it the way you'd read an anatomy atlas: for the sections that map onto decisions you actually make. Prioritize the chapters on overconfidence, the planning fallacy, and loss aversion; skim the pure-psychology-lab chapters.
What it's really about for you: your mind runs two systems — a fast, intuitive one (System 1) that is usually right and occasionally confidently, catastrophically wrong, and a slow, effortful one (System 2) that you rarely deploy because it's expensive. The danger isn't System 1 — it's System 1 dressed up as certainty with no System 2 check. In an OR, fast pattern recognition saves lives. In an indication meeting or a manuscript interpretation, it can quietly mislead you and everyone trusts it because you're the expert.
There's a specific application worth naming directly: the affect heuristic — a positive feeling about something makes you rate its benefits as high and its risks as low, in a single undifferentiated judgment, instead of evaluating them separately. This is exactly the mechanism behind a new AI tool feeling both exciting and safe to build the moment it appears interesting. The excitement isn't evidence the risk is actually low; it's a System 1 shortcut that bundles "I like this" with "this is a good use of the next six months." Separating the two — do I like it, and independently, is it a good bet — is a System 2 tax worth paying before the tenth AI project, not just the first.
Key ideas, in your own terms:
A tumor board case: an atypical meningioma with a borderline margin. Everyone in the room, including you, feels confident it will behave indolently based on "how it looked."
A resident hands you an imaging read that opens with "the referring radiologist thought it was probably benign." You haven't looked at the images yet.
A new AI tool idea appears, and within the hour it feels both obviously worth building and low-risk — you're already sketching the architecture.
You estimate a manuscript revision will take two weeks. You've made this exact estimate on the last four manuscripts, and each one took over a month.
Shorter and more practical than Kahneman — read it in a week, then keep the natural-frequencies technique as a permanent clinic tool. Gigerenzer's argument is not "people are irrational" (Kahneman's frame); it's "people reason well when risk is presented in a format the mind actually evolved to use, and badly when it isn't."
What it's really about for you: the gap between a statistically correct number and a number a frightened patient can use. "8% recurrence risk at 5 years" and "12 out of 150 patients like you had this happen within 5 years" are the same fact — but only one of them is usable under stress. Relative-risk framing ("risk doubles") without a base rate is close to a lie, even when the math is right.
Gigerenzer also names an illusion of certainty that cuts both ways: patients want a single confident number, and physicians are trained to supply one, even when the honest answer is a range built on a small local series. The temptation isn't only to mislead the patient — it's to let a false sense of precision quiet your own uncertainty too. A well-calibrated surgeon should be more comfortable saying "our data on this specific scenario is thin, here's the range" than manufacturing a clean percentage that sounds more authoritative than it is.
You're counseling a patient on postoperative CSF leak risk for an endoscopic endonasal approach.
You're deciding whether to greenlight a new AI tool build the same week you're also deciding on a familiar, well-worn approach for a routine skull-base case.
A reviewer or co-author asks for a single hearing-preservation percentage for a rare tumor subtype, and your own series has only 9 comparable cases.
You're deciding whether to order an additional imaging study that's unlikely to change management, mostly because you'd feel exposed without it if something rare happened.
The shortest of the three, and the one that will change your post-complication review most directly. Duke's background is professional poker — a domain where you make a decision with incomplete information, the cards fall however they fall, and a good player is judged on the decision quality, not the hand's result. Medicine and research are the same game with higher stakes.
What it's really about for you: "resulting" — the habit of grading a decision entirely by its outcome. A good indication with an unlucky complication gets graded as a bad decision. A borderline indication that happened to go fine gets graded as a good one. Both gradings are wrong, and both erode your judgment over time if you let them stand uncorrected.
This applies just as much outside the OR. A well-reasoned ad campaign or content bet that doesn't convert this month is not automatically a bad marketing decision, and a lucky viral post from a rushed, under-thought idea is not automatically a good one. If SlideCraft's pricing or a course launch gets graded purely on revenue this quarter, you'll learn the wrong lessons — reinforcing lucky moves and abandoning sound ones. Duke's frame asks you to keep a separate ledger: was the reasoning good, independent of what the market happened to do this month.
A well-indicated, well-executed resection is followed by an unexpected vascular complication. The family and, quietly, part of you wants to know: "what did you do wrong?"
You're about to greenlight a new course, tool, or research collaboration you're genuinely excited about.
An ad campaign built on a sound audience hypothesis and a clean creative test underperforms this month. A rushed, under-researched post from a slow week goes unexpectedly viral.
A team member (Flávia, a resident, a co-founder) flags that your confident prediction about a patient's course or a project's timeline was wrong, with new evidence to back it up.
Here is the part of this month that is about you specifically, not decision theory in general.
Two patterns from the profile live here, and this month asks you to hold both at once.
You have a documented, rare strength: you kill sunk costs. MedOps. Clawd. Synthesium. CCF papers that weren't going anywhere. Most people — most physicians, most founders — cannot do this. They keep feeding a dead project because stopping feels like admitting the original bet was wrong. You've done it four times, with real grief and real follow-through, which means your kill-decisions are, by track record, some of the best-calibrated decisions you make. Duke would call this excellent bet-grading: you can look at a hand that isn't working and fold, instead of chasing the sunk pot.
But the same clarity does not show up symmetrically at the other end of the bet. Kahneman's planning fallacy and the anchoring habit above both describe a bias that hits hardest at the start of a project, not the end: you anchor on the best-case narrative, you underestimate the timeline, you feel the interest-driven activation of a new build as evidence that this one is different. Your exposure isn't in the kills — it's in the starts. The optimism that makes you willing to begin a fifth ambitious thing is the same optimism that made the first four take longer, cost more, and eventually need killing. Sharpening your kill-instinct won't fix this; it's already sharp. What needs the Duke discipline applied before the bet — the pre-mortem above — is the decision to start.
The second pattern this month sharpens is the escalation-and-recovery cycle: a hair-trigger on wasted attention, escalation language that outsizes the event, and — the important part — that escalation clusters with fatigue. A late-night "I'm done with this," a "never again" about a collaborator, a "fire them" after a frustrating call: these read, in the moment, like decisions. Duke's frame says otherwise. A decision made while depleted isn't a bad decision necessarily — it's a decision made with a corrupted instrument, the same way you wouldn't trust a reading from a miscalibrated monitor. You already half-know this; the 24-hour rule makes it operational instead of a vague intention: write the kill or the threat down, and execute it tomorrow, unedited. If it survives the night intact, it was real — do it without guilt, and don't let anyone (including you) call it impulsive. If it doesn't survive, it wasn't a decision about the target at all. It was a fatigue signal, and the real item on the list is sleep, not termination.
It's 11:40pm. A call with a collaborator went badly. You're drafting a message ending the relationship, and it feels completely clear-headed.
Two moments, same week: (A) you finally decide to shut down a research thread that's gone nowhere for a year; (B) you get excited about a new tool idea and want to start building tonight.
There's a structural counter-move already in place, and it's worth naming so you keep using it deliberately rather than by accident: the Pact with Samilly is a designed team, not just a marriage running on goodwill. A decision made solo at 11pm carries none of the built-in 24-hour buffer that a decision surfaced to her the next morning does. The same is true for starts — a new build proposed to a co-founder, a co-author, or a spouse before it's committed to gets a second, less-invested set of eyes on the pre-mortem, which is exactly the check your own optimism can't run on itself. Using the marriage as a decision-quality instrument, not just an emotional one, is the practical form of "truthseeking over being-right" this month asks for.
Concretely: the PhD defense date (Feb 2027) is a good anchor for testing this in real time. You will not need help deciding whether a stalled arm of the research should be killed — your track record says you'll see that clearly, and probably later than you should have started worrying but not later than the moment truly requires action. Where the same rigor is missing is at the front: when a new build, a new course module, a new AI tool idea appears interesting this week, the interest-driven activation from Month 01 and Month 03 conspires with the planning-fallacy optimism from Kahneman to make the start feel low-risk when it is, in fact, the highest-leverage moment to apply scrutiny — because it's the only moment before the sunk cost exists.
You're excited about starting a new marketing experiment (a new content series, a new ad angle) the same week Samilly is reviewing whether a slow-moving clinic initiative should finally be shelved.
Three weeks from a hard PhD deadline, exhausted, you're convinced a whole analysis approach was wrong from the start and want to scrap months of work tonight.
A registry project has quietly stalled for eight months. No one has said "kill it" out loud yet, but the signs match your prior kills exactly.
Put simply: keep judging the bet, not the card that fell — and remember that a good bettor scrutinizes the opening raise as hard as the fold.
These are the load-bearing passages of the month — where each idea actually lives in the books, so you can go straight to the page instead of re-reading whole chapters. All of these texts are in copyright, so what follows is the location and the argument in this lesson's own words, not the authors'. Read the originals; use this as the map.
Near the end of the overconfidence section, Kahneman hands the floor to his friendly adversary Gary Klein and endorses one procedure without reservation. Before a decision is finalized — not after — the team imagines it is a year later and the plan has failed completely, then each person independently writes the history of that failure. The genius is social, not analytical: once a group has converged on a plan, expressing doubt reads as disloyalty, so doubt goes silent exactly when it's most valuable. The premortem, in Kahneman's words, "legitimizes doubts" — it makes the skeptic's job an assignment instead of an act of courage.
Em termos simples: before greenlighting a new build, hire, or course, make "write down why this failed" a required step — so the doubt gets spoken while it can still change the plan, not at the post-mortem.
Kahneman tells the story on himself. His team is writing a decision-making curriculum; everyone privately estimates about two more years. Then he asks the one member who has watched many such teams: how long did comparable teams take? The answer — many never finished at all, and none finished in under seven years — describes their own project, and the team hears it, shrugs, and keeps working to the two-year story. (It took eight.) The chapter's lesson: the inside view (this case's narrative) feels like knowledge; the outside view (the track record of the reference class) is knowledge, and it loses the argument anyway unless you force it to go first.
Em termos simples: when estimating any manuscript, project, or launch, ask "how long did my last four actually take" before asking "how long does this one feel like" — and believe the first number.
Gigerenzer's signature demonstration. Give physicians a screening problem in percentages — prevalence ~1%, sensitivity ~90%, false-positive rate ~9% — and ask what a positive mammogram means; most answer that the woman very likely has cancer, some say 90%. Recast the same facts as counts and the fog lifts: out of 1,000 women, about 10 have cancer and 9 of them test positive; of the 990 without cancer, about 89 also test positive. So a positive result puts her among ~98 women, of whom ~9 are sick — roughly one in ten. Same mathematics, different format; the format is the difference between a counseled patient and a terrified one. The deeper claim: the confusion isn't patient innumeracy, it's how we present the numbers.
Em termos simples: quote every risk to a patient as counts of real people ("of 100 like you, about X"), because that is the only format anyone — including you — reliably reasons with under fear.
Duke opens with the most second-guessed play call in Super Bowl history: Seattle passes on the one-yard line, the pass is intercepted, and by morning the decision is universally "the worst call ever." Duke walks through the actual decision structure — clock, timeouts, interception odds on that throw historically tiny — and shows the call was defensible, arguably good; only the outcome was terrible. She names the error "resulting": reading the quality of a decision off the quality of its result. Her poker table taught her the corollary that matters most for you: resulting corrupts learning — it teaches you to repeat lucky sloppiness and abandon sound process.
Em termos simples: in every complication review, grade the indication against what was knowable pre-op, in a separate column from what actually happened — or the review will train you wrong.
Tetlock dissects how the measurably best forecasters in his tournaments attack a question — say, whether a particular family owns a pet. The amateur starts from the vivid particulars of the family; the superforecaster starts from the base rate (what fraction of comparable households own pets) and only then adjusts for specifics. Anchor on the reference class, adjust with the case; never the reverse — because the particulars are seductive and the base rate is boring, and boring is what keeps you calibrated. The other habit worth stealing from the same chapters: granular probabilities, written down, and scored later. Most experts, he shows, never once check their own calibration.
Em termos simples: when judging a tumor's likely behavior or a project's odds, start from "what usually happens in cases like this" before "what I sense about this one" — and write the probability down so future-you can grade it.
Taleb's central move is a missing word: we have "fragile" (breaks under volatility) and "robust" (resists it), but no everyday word for what gains from stressors — so the category stays invisible and we build lives and institutions that merely resist shocks instead of profiting from them. His practical instrument is the barbell: extreme safety on one side, many small aggressive bets on the other, nothing in the fragile middle. Keep the downside capped and the upside open, and volatility becomes your ally. Your portfolio already has the shape — stable surgical practice on one end, cheap killable experiments on the other; the discipline is refusing the middle-sized irreversible commitment that can actually hurt you.
Em termos simples: structure every new bet so the worst case is survivable and pre-priced — small, reversible, killable — and never stake the clinic-sized middle on a single uncertain thing.
One older voice belongs beside these five, because it is the root of the whole month and it is out of copyright, so it can speak verbatim:
Twenty centuries before "resulting" had a name, this is the same cut: the event and your judgment of the event are two different objects, and only the second one is yours. Em termos simples: the complication is a fact; "I decided badly" is a judgment — and this month's entire toolkit exists to make that judgment on evidence instead of on pain.
Loci: Kahneman, Thinking, Fast and Slow, chs. 23–24 · Gigerenzer, Risk Savvy, screening chapter · Duke, Thinking in Bets, ch. 1 · Tetlock & Gardner, Superforecasting, ch. 5 · Taleb, Antifragile, Book II. All paraphrases; page numbers vary by edition.
This is the concrete deliverable for the month — not a book report, a running habit. One entry per meaningful surgical or research decision: an indication call, a research design choice, a manuscript interpretation, a hire, a strategic bet on the course or the AI tools. Not every minor decision — reserve it for the ones with real stakes or real uncertainty.
1. Decision: what am I actually deciding, in one sentence?
2. Fatigue level right now (1–5): a honest gut number, no explanation needed. If this is a kill, a "never again," or a threat, and the number is 3 or higher — this entry is provisional until the 24-hour rule clears it.
3. Options considered: list them — including "do nothing" or "wait."
4. What I expect + probability: "I expect ___, with about ___% confidence." Force a number.
5. What would change my mind: name the specific finding, test result, or event that would flip the decision.
6. Outcome (filled in later): what actually happened, recorded without editing the earlier fields.
7. Decision-quality review (separate from outcome): was the decision sound given what was knowable at the time? What would I decide again, unchanged? What would I change in the process, not just this case?
Where it goes: one file per decision (or a running log) in learning-records/decision-journal/, dated. Fields 1, 2, 3, 4, and 5 are filled in before the outcome is known — that's the whole point; you cannot honestly grade decision quality if you fill in the expectation after seeing the result. The fatigue field is new this month specifically for you: it turns "read escalation as fatigue signal, not information about the target" from an idea into a number you actually write down.
What "done" means — the artifact and the behavior, not either alone: by month's end, at least 4–6 entries with fields 1–5 written in advance of the outcome, and at least 2 with field 7 (the review) completed once the outcome landed. And done also means: at least one high-fatigue kill or threat entry actually held for 24 hours before being acted on, or actually executed the next day once it survived the night. A journal full of entries with no changed behavior at the moment of deciding is a note about decision-making, not decision-making itself — collection, not formation. This will feel awkward at first — writing down a probability, or a fatigue number at 11pm, feels like exposing yourself. That exposure is the training.
1. Decision: whether to proceed with the endonasal approach or convert to an open approach given intraoperative bleeding at the tumor-vessel interface.
2. Fatigue (1–5): 2 — third hour of a planned five-hour case, controlled conditions.
3. Options: (a) continue endonasal with vessel control, (b) convert to open now, (c) pack and stage a second-look procedure.
4. Expectation + probability: "I expect (a) to succeed with vessel control, ~75% confidence, based on visualization and bleeding pattern."
5. What would change my mind: loss of visualization for >3 minutes, or bleeding rate exceeding what suction can manage.
6/7. Outcome + review (filled next day): vessel controlled, case completed endonasally. Review: decision was sound given the visualization at the time — would decide the same again. Nothing to change in the process.
A second, quieter way it gets faked: writing entries only for decisions you're confident about, and skipping the genuinely uncertain ones because a written 50% feels more exposing than a written 90%. The uncertain entries are the ones actually worth the exercise — a confident call you were right about teaches you nothing about your calibration.
This month sharpens moments already in your daily protocol rather than adding new ones.
In clinic, Gigerenzer's natural-frequencies habit becomes a permanent script: whenever you quote a risk number to a patient, pair it with a real denominator from your own case series, not a bare percentage from a paper they'll never read.
In the PhD's methods sections and grant language, the same script applies to your own reported statistics: prefer a natural-frequency phrasing and an honestly small denominator over a smoothed percentage borrowed from a bigger, less comparable series.
A Sunday review surfaces a recurring pattern: three of the week's decision-journal entries were logged with a fatigue level of 4 or higher, all late at night.
One more attachment point, specific to this month: any time a "kill/fire/never-again" declaration surfaces during the daily protocol's evening review, it now has a home — the fatigue field in the decision journal, not a same-night message. And any time a new build feels irresistibly interesting during the morning study block, the pre-mortem question ("imagine this failed in a year — why") gets asked before it gets a line item on the week's plan, not after the first commit.
None of this adds a new ritual. It attaches new vocabulary and one new field to slots the daily protocol already has.
Judge the bet, not the card that fell.
A decision made in fatigue is not yet a decision. Sleep on it, then execute without flinching.
One closing note on the two maxims together: the first is about grading decisions once they've been made; the second is about knowing when you're not actually in a position to make one yet. You already live the first reasonably well. The second is this month's real edge — for both the kills you already do cleanly and the starts you don't yet scrutinize.
Carry both into next month unfinished — that is the point.
Bring it back to yourself. Open a fresh file in learning-records/decision-journal/ and log one real decision you're facing this week — fields 1–4 only, before you know how it turns out. Then write your Month 05 formation note (the 5-heading format in NOTES.md) once you've finished the three texts.
This month isn't "done" when you finish reading — it's done when the journal has entries and one has come back around for a decision-quality review, independent of outcome.
Texts: Kahneman, Thinking, Fast and Slow · Gigerenzer, Risk Savvy · Duke, Thinking in Bets. Full rationale in SYLLABUS.md; note format in NOTES.md.