You already forgot why you made that decision
Three months ago, you made a decision you were confident about. Maybe you chose a tech stack, turned down a job offer, or restructured your team. Today, you know how it turned out. And here is the problem: you no longer remember what you actually believed when you made the call.
You think you do. But what you remember is a reconstruction — a narrative your brain assembled after the fact, contaminated by everything you learned between then and now. If the decision worked out, you remember feeling certain. If it failed, you remember having doubts you never actually had. Baruch Fischhoff demonstrated this in 1975 in the first experimental study of what he called creeping determinism: once people know an outcome, they systematically overestimate how predictable that outcome was from the start. Subjects told that a particular result had occurred rated it as far more likely than subjects who hadn't been told — even when explicitly instructed to ignore the outcome information.
This is hindsight bias, and it is not a minor quirk. It is a systematic distortion of your ability to learn from experience. Every decision you make without recording your reasoning at the time is a decision you will misremember — and a lesson you will therefore fail to extract.
A decision journal is the counter-mechanism. It creates a time-stamped, tamper-resistant record of what you believed, why you believed it, and what you expected to happen — written before you could possibly know the outcome. When you review it later, you are comparing your actual past reasoning against actual results, not your reconstructed past reasoning against a story you told yourself.
What a decision journal actually contains
Shane Parrish of Farnam Street formalized the most widely used decision journal template, and its design is deliberate. Each entry captures six elements at the moment of decision:
- Date and time. Decisions cluster around periods of stress, fatigue, or excitement. Patterns only become visible over dozens of entries.
- The decision itself, stated in one clear sentence. Forcing a single sentence prevents the hedging that lets you claim credit for any outcome.
- The reasoning — the specific factors, evidence, and mental models that drove the decision. Not a post-hoc rationalization, but the actual chain of thought as it exists right now.
- The expected outcome, stated concretely enough that future-you can judge whether it happened. "This will go well" is useless. "Revenue from this client will exceed $50K within six months" is assessable.
- Your confidence level, expressed as a probability. This is the element that makes the journal a calibration instrument rather than a diary. If you say 80% and the thing happens 80% of the time across many entries, you are well-calibrated. If your 80% predictions come true 50% of the time, you are systematically overconfident — and now you know it.
- Your mental and physical state. Were you rested or exhausted? Calm or angry? Pressured by a deadline or choosing freely? This field turns your journal into a dataset on the conditions under which your reasoning degrades.
The template works because it captures the decision in amber. It cannot be revised, reinterpreted, or softened once the outcome arrives. That is the entire point.
The review protocol matters more than the record
Writing entries is necessary but not sufficient. The journal's real value emerges during structured review, and the review protocol must be designed to defeat the very biases the journal exists to expose.
Parrish recommends reviewing entries on a rolling schedule — 30 days, 90 days, 6 months — depending on the decision's time horizon. But the sequence of the review matters as much as its timing:
Step 1: Re-read your original reasoning without looking at the outcome. Cover or hide the result. Force yourself back into the epistemic state you occupied at the time of the decision. What did you believe? What were you uncertain about?
Step 2: Predict the outcome again, given only your original reasoning. Does your current prediction match the confidence level you recorded? If it diverges — if you now feel "obviously" this was going to fail — you have caught hindsight bias in the act.
Step 3: Uncover the actual outcome. Now compare three things: your original prediction, your just-now prediction, and reality. The gaps between these three data points contain the actual learning.
This protocol is what Donald Schon called reflection-on-action — deliberately reviewing past practice to surface the tacit knowledge and hidden assumptions that drove your behavior. In The Reflective Practitioner (1983), Schon distinguished this from reflection-in-action (thinking on your feet during the event itself). Both matter, but reflection-on-action is what a decision journal systematizes. You cannot engage in honest reflection-in-action about a decision from three months ago. Your memory is already corrupted. You need the written record.
Separating decision quality from outcome quality
Annie Duke, in Thinking in Bets (2018), identifies the most common error in decision review: resulting — judging the quality of a decision by the quality of its outcome. You took a calculated risk and it paid off, so you conclude it was a good decision. You made a careful, well-reasoned choice and it failed, so you conclude you were wrong.
Both conclusions are dangerous. A good decision can produce a bad outcome because of information you couldn't have had, variance you couldn't control, or simple bad luck. A bad decision can produce a good outcome for the same reasons in reverse. If you only evaluate decisions by their results, you will abandon sound reasoning whenever you get unlucky and reinforce sloppy reasoning whenever you get lucky.
Your decision journal protects you from resulting because it preserves the reasoning independent of the outcome. When you review an entry and the outcome was poor, the question is not "was the outcome bad?" — that is already known. The question is: "Given what I knew at the time, and what was knowable, was the reasoning sound?" If yes, the bad outcome is information about the world, not evidence of flawed thinking. If no — if your reasoning contained gaps you should have caught, if you ignored available evidence, if you let emotional state override analysis — then the reasoning itself needs repair, regardless of whether this particular outcome happened to work out.
This is the fundamental discipline: evaluate the process, not the result. Over dozens of entries, the patterns that emerge will show you exactly where your reasoning reliably breaks down — under time pressure, in specific emotional states, about certain categories of decisions — and that is the kind of self-knowledge no amount of introspection alone can produce.
Calibration: the skill your journal trains
Philip Tetlock's research on forecasting accuracy — culminating in Superforecasting (2015) and the Good Judgment Project — established that calibration is a learnable skill. A well-calibrated person who says "I'm 70% confident" is right roughly 70% of the time. Most people are not calibrated. They are overconfident on hard questions and underconfident on easy ones. Tetlock found that even brief training on probability estimation and cognitive biases produced measurable improvements in forecasting accuracy.
Your decision journal is a calibration training tool. Every entry with a confidence percentage is a forecast. Across 50 or 100 entries, you can plot your calibration curve: for all the decisions where you wrote "80% confident," what percentage actually turned out as predicted? If the answer is 55%, you have a specific, quantified bias to correct. If the answer is 82%, your epistemic self-assessment is accurate in that range, and you can trust it more.
The superforecasters Tetlock identified shared several traits, but the most relevant one here is perpetual beta — they treated their own beliefs as hypotheses to be updated, not positions to be defended. A decision journal operationalizes that stance. Each entry is a hypothesis. Each review is an update. Over time, the journal trains you to hold your beliefs with appropriate confidence — neither more nor less than the evidence warrants.
The AI parallel: experiment tracking and decision audit trails
If you work in software engineering or machine learning, the decision journal pattern will be immediately familiar — because your systems already do this.
MLflow, the open-source experiment tracking platform, records every model training run with its parameters, metrics, code version, input data, and output artifacts. Each run is a time-stamped, immutable record of what was tried and what resulted. When a model in production starts degrading, engineers don't guess what changed — they compare the current run against the historical log. The reasoning (hyperparameters, architecture choices, training data) is preserved alongside the outcome (accuracy, loss, latency) in a structure that makes before-and-after comparison trivially easy.
This is a decision journal for machines. And the principle is identical: you cannot learn from decisions you did not record.
The pattern extends beyond ML. Architecture Decision Records (ADRs) in software engineering capture the context, options considered, and rationale for significant technical decisions — so that when a future team asks "why did we choose PostgreSQL over DynamoDB?" the answer exists in a searchable, reviewable document rather than in the faded memories of engineers who may have left the company. Git commit histories, changelog entries, and incident post-mortems all serve the same function: creating an audit trail that makes past reasoning transparent to future review.
The difference between human decision journals and AI experiment tracking is that AI systems do this by default. Every training run is logged. Every deployment is versioned. Every prediction can be traced back to the model, the data, and the parameters that produced it. Humans, left to their defaults, record nothing — and then wonder why they keep making the same mistakes.
Your decision journal closes that gap. It gives your reasoning the same traceability that well-engineered systems give their outputs.
Building the habit
The journal fails if you only use it for "big" decisions. The most valuable entries are often mundane — a hiring choice, a project prioritization call, a decision to skip a meeting, a bet on which feature to ship first. These are the decisions you make on autopilot, and autopilot is where your worst patterns live.
Start with a minimum viable practice: one entry per day for decisions you are about to make or just made. Keep the template consistent. Write before you know the outcome, every time. Set review reminders at 30-day intervals. When you review, follow the three-step protocol — re-read reasoning, re-predict, then compare.
The first ten entries will feel mechanical. By entry fifty, you will start noticing patterns: the emotional states that correlate with poor reasoning, the categories of decisions where you are systematically overconfident, the situations where your intuition is actually reliable. By entry one hundred, you will have a calibrated, evidence-based understanding of your own decision-making — not the flattering story you tell yourself, but the operational reality.
Pre-commitment (the previous lesson) decides what you will do before the moment arrives. The decision journal decides what you will learn after the moment passes. Together, they form a closed loop: pre-commit to a course of action, record the decision and its reasoning, review the outcome against the prediction, and feed the learning back into your next pre-commitment. Time pressure (the next lesson) adds the third constraint — ensuring you actually make the decision rather than endlessly deferring it.
What the journal reveals
You are not as rational as you think. Neither am I. Neither is anyone. But rationality is not a fixed trait — it is a practice, and like any practice, it improves with accurate feedback. The decision journal is the feedback mechanism. Without it, you are navigating by a map that rewrites itself every time you look at it. With it, you have a fixed record that lets you see clearly where your map was right, where it was wrong, and — most importantly — why.
The question is not whether your decisions would benefit from this kind of scrutiny. They would. The question is whether you are willing to create a record honest enough to prove yourself wrong.