Kahneman — Thinking, Fast and Slow
Key question: Where might I be confidently wrong?
- System 1 (fast, intuitive) vs. System 2 (slow, effortful) — confidence is a System 1 feeling, not evidence.
- Overconfidence: experts are often more confident, not more accurate, in ambiguous cases.
- Planning fallacy: you underestimate timelines even knowing the fallacy exists.
- WYSIATI: the mind builds a confident story from whatever's in front of it and stops looking.
Gigerenzer — Risk Savvy
Key question: Is this risk in a form the patient can use, or a form that protects me?
- Natural frequencies beat percentages: "6 of 150," not "4%."
- Absolute vs. relative risk — always anchor to a real denominator.
- Defensive decision-making: some testing/treating protects the physician, not the patient — name it.
Duke — Thinking in Bets
Key question: Was the decision sound given what was knowable at the time?
- Decision quality ≠ outcome quality — the whole thesis.
- "Resulting": don't grade decisions by how they turned out.
- Betting language ("70% confident") is more honest than "I think."
- Truthseeking > being-right — reward updating on evidence.
Maxim
Judge the bet, not the card that fell.
Sob medida — the asymmetry
- The pattern: you kill sunk costs well (MedOps, Clawd, Synthesium, CCF papers) — the blind spot is at the start, where planning-fallacy optimism and interest-driven activation go unscrutinized.
- The counter-move: pre-mortem every new build before the first commit; run the 24-hour rule on any kill/fire/"never again" said while depleted.
- The tell: escalation language clusters with fatigue — log a fatigue level (1–5) with the decision; a verdict logged at 4–5 is provisional until it survives the night.
The practice — decision journal
One entry per meaningful surgical or research decision. Fields 1–4 filled before the outcome is known. File in learning-records/decision-journal/, dated.
- 1. Decision — one sentence, what am I actually deciding.
- 2. Fatigue level (1–5) — gut number now. Kill/fire/"never again" at 3+ is provisional until it clears the 24-hour rule.
- 3. Options considered — including "do nothing" / "wait."
- 4. Expectation + probability — "I expect ___, ~___% confidence." Force a number.
- 5. What would change my mind — the specific finding/event that flips it.
- 6. Outcome (later) — recorded without editing fields 1–5.
- 7. Decision-quality review — sound given what was knowable? What repeats unchanged, what changes in process?
Done means: 4–6 entries with fields 1–5 pre-filled; ≥2 with field 7 completed; at least one high-fatigue verdict actually held 24 hours. Then write the Month 05 formation note (5-heading format, NOTES.md) — a note with no changed behavior is collection, not formation.
Daily protocol integration
- Pre-op Q2 ("what can go wrong?") — attach a rough probability to each risk, not just a list.
- Post-op review — add "was this decision sound given what was knowable," scored independent of outcome, alongside Month 01's dichotomy checklist.
- Clinic — pair every risk quote with a natural-frequency denominator from your own series.