90-Day Probabilistic Thinking Protocol: Build Bayesian Intuition

How this programme adapts

Once a week, the player runs a short check-in and reads the week you actually logged — not the week the plan assumed. Every check-in lands on one of the adjustments below. These are the engine's own rules, rendered word for word.

Sparse week
Not enough check-ins to assess this week. Carry your forecast-log habit forward unchanged and aim for daily logging next week.
Expand your forecast log
Logging more probability estimates each week gives your calibration score more signal to work with. Aim for at least five logged forecasts next week.
Calibration needs attention
Your self-rated calibration has been low this week. This week slows down and focuses on one domain: look up base rates before every estimate rather than generating them from scratch.
Strong calibration week
Your logging frequency and self-rated calibration are both healthy. This week adds one harder domain to your forecast log: a question type you have not estimated before.
On track
Your calibration practice is on pace. This week layers in the next scheduled drill from your plan.

Pattern insights are gated behind your own record: the first unlocks after 3 logged days, and every insight states the evidence it stands on. Milestone reviews at day 30, day 60, day 90 read your whole record and choose their coaching from it.

The 90-day programme map

Every day of the programme, in full — the same steps the interactive player walks you through. Expand a week to read its days.

Month 1 — Foundation · Days 1–30

Build your probability-estimation baseline and core calibration habits

Week 1Calibration AuditDays 1–7

Day 1 · Start Your Forecast Log 15 min

Your probability-estimation practice starts today with a single habit: the forecast log. Every morning, before the day fills your attention, you pick one question with an uncertain outcome, assign it a number from 0 to 100, and write it down. The act of committing to a number is what trains calibration over time, not analysing afterward.

Each morning, write one forecast: pick any question with an uncertain outcome, assign a probability from 0 to 100, and log it in a notebook or app
Write three domains where you regularly face uncertainty: work outcomes, personal plans, or current events
Note one forecast you already made informally today before this programme started and give it a number in your log

Day 2 · What Your Number Reveals 10 min

A logged probability is useful only if you can trace why you chose it. Today you examine the first number you wrote yesterday: what evidence drove that estimate? This is not about accuracy yet. It is about building the habit of connecting your confidence to something specific rather than leaving it as a vague feeling.

Open your forecast log and write a one-sentence reason for the probability you assigned yesterday
Pick one question from today where you feel very certain; log it with a probability and note the single strongest piece of evidence for that confidence

Day 3 · Finding the Base Rate 15 min

Before shaping an estimate from your own experience, there is a prior question: what typically happens in situations like this one? Base rates are the statistical frequency of outcomes in a defined reference class. Today you add the habit of looking up this frequency before forming your own view, which is the single most consistent anchor for well-calibrated estimates.

Before making any estimate today, look up what typically happens in similar situations (the base rate) and write it down before adding your own view
Find one published base rate relevant to a current question in your forecast log: project completion rates, acceptance rates, or typical timelines for a plan you hold
Write how your intuitive estimate compares to the base rate you found

Day 4 · What 70 Per Cent Feels Like 10 min

A well-calibrated person who says they are 70 per cent confident is right about 70 per cent of the time across many such statements. Most people are overconfident at every level of certainty, including high-confidence levels where the pattern is strongest. Today you build a concrete sense of what your own 70 per cent actually means in practice.

Pick 5 factual questions you feel roughly 70 per cent sure about; log each with a probability and check the answers
Note in your log whether your 70 per cent set produced roughly 3 or 4 correct answers, or whether you were right more or less often than that

Day 5 · Easy and Hard Estimates 10 min

Not all uncertain questions feel the same. Some produce a clear number quickly; others feel genuinely uncomfortable to pin down. Both types belong in your forecast log. This week your task is observation: which domains feel easiest to estimate, and which feel hardest? The answer maps where your practice will do most work.

Log 2 forecasts that feel easy to assign a number to and 2 that feel genuinely hard; assign probabilities to all four
Write one sentence on what makes the hard estimates feel different from the easy ones

Day 6 · Expert Domains and Confidence 10 min

Research on calibration shows that people who know more about a domain are not always better calibrated about the limits of that knowledge. Familiarity can produce overconfidence rather than accuracy. Today you observe your own confident territory, not to critique it but to mark it clearly in your log.

Pick one domain where you consider yourself knowledgeable; log 3 estimates in that domain with explicit probability numbers
Note whether your intervals feel narrower in this domain than in unfamiliar ones, and write one sentence on why

Day 7 · Week 1 Audit 15 min

Seven days of forecast-logging gives you a first sample to work with. This audit is not a performance review. It is a map of the territory: which domains appeared most often, where your confidence ran highest, and where the logging habit felt easiest to maintain. These observations set the baseline for everything that follows.

Count the total forecasts you logged in week 1 and group them by domain: work, personal plans, current events, or other
Flag one domain where your stated confidence felt systematically high and write one sentence on why

Week 2Base-Rate GroundingDays 8–14

Day 8 · Reference Class for Projects 15 min

The reference class is the group of situations most similar to yours. Looking at outcomes for that group, before adding any inside-view knowledge, is where the outside view begins. Today you install this habit specifically for personal projects, where the planning fallacy is most common and where a reference class is most readily available.

Before estimating the timeline or outcome of a personal project, ask: what is the reference class? How did similar projects turn out for others?
Pick one current project; define the most appropriate reference class; find outcome data or typical timelines for that class
Compare your instinctive timeline or outcome estimate to the reference class data; write the gap

Day 9 · The Prior Probability 10 min

In Bayesian terms, the base rate is your prior: the starting probability before you add case-specific information. Starting from the prior rather than from intuition is a simple procedural shift that anchors estimates more reliably. Today you practice the sequence deliberately: prior first, specific information second.

Log one forecast starting from the base rate: write the base rate, then your adjustment, then the final estimate in sequence
Write one sentence on what specific information justifies any departure from the base rate you found

Day 10 · Inside Knowledge vs Outside Data 10 min

The inside view focuses on the specific features of your current plan. It feels more relevant because it is about you. The outside view asks how often plans like this one actually succeed. Most decisions call for both, but the outside view is far more often skipped, which is exactly where the planning fallacy lives.

Pick any current plan; write your gut estimate first, then find a base rate or reference class; compare the two numbers
Note whether you underweighted or overweighted the base rate in your gut estimate and by roughly how much

Day 11 · Base Rate for Social Outcomes 10 min

Base rates apply to social predictions as much as to timelines and project outcomes. When you predict how someone will respond to a request, there is often a relevant frequency to consult. We tend to treat social situations as uniquely personal, which is precisely when base-rate neglect becomes strongest.

Pick one prediction about another person's response or behavior; find a relevant base rate such as acceptance rates, response rates, or follow-through rates
Log the base rate alongside your final estimate and write the gap between them

Day 12 · When Your Situation Differs 10 min

Base rates are starting points, not verdicts. The key discipline is distinguishing between a genuine reason your situation differs from the reference class and a motivated story you are telling to justify an outcome you prefer. Today you practice that distinction explicitly for one of your open forecasts.

Log one forecast where you believe your specific information justifies departing from the base rate; write the case for that departure in one sentence
Rate your confidence in the departure from 1 to 5; write one piece of evidence that would change your mind

Day 13 · Neglected Frequencies 10 min

Base-rate neglect is most common in the domains where we feel most personally involved: our own plans, our own teams, our own ideas. Today you find one recurring area where you have been estimating from intuition without consulting frequency data, which is the first step to correcting the pattern.

Identify one recurring decision type in your life where you have never used a base rate; look up a relevant frequency or outcome rate
Log the finding and note whether it changes how you view that decision type going forward

Day 14 · Week 2 Calibration Check 15 min

Two weeks of forecast-logging gives you enough entries to see your base-rate habits in practice. Today you review not for accuracy but for process: how often did you look up a frequency before forming an estimate, and how often did you generate from instinct alone? Process review is the diagnostic step before changing habits.

Count how many of this week's forecasts started with a base rate lookup vs instinct; write both numbers
Write one sentence on your natural tendency: do you seek or avoid base rates when they might contradict your intuition?

Week 3Inside vs Outside ViewDays 15–21

Day 15 · State Estimates as Ranges 15 min

A single-point estimate implies a precision you almost never have. A range is more honest and more informative because it makes your uncertainty visible. From today, every forecast you log includes three numbers: low, best-guess, and high. The width of the range is data, not weakness.

State all estimates as a range, not a single number: write a low-end, a best-guess, and a high-end for each forecast you log
Go back to one week 1 forecast and rewrite it as a three-number range
Note what the width of your range reveals about your honest uncertainty level in that domain

Day 16 · The Outside View Defined 10 min

The outside view asks what happened when other people tried the same thing. It produces a different estimate than the inside view almost every time. Today you practice the transition between the two: start with your gut timeline, find the reference class, and note the difference without defending either number.

Pick one personal project with a deadline; write the inside-view estimate first, then find how long similar projects took for others
Write the outside-view timeline as a range; note the difference between the two views in your log

Day 17 · Planning Fallacy in Your Own Life 10 min

The planning fallacy is the consistent tendency to underestimate the time, cost, and risk of future plans while overestimating the benefits. You have your own personal history with it. Today you look at that history directly, not to criticise past decisions but to find the pattern that informs future estimates.

Recall 3 recent projects or plans that ran over time or budget; note what the inside view missed in each case
Write one sentence on any pattern connecting all three overruns

Day 18 · Reference Classes for Major Decisions 10 min

Reference classes are not just for project timelines. They apply to any decision with an uncertain outcome: career moves, financial choices, personal commitments. The discipline is defining the reference class before adding your specific read of the situation, not selecting a convenient class after the fact.

Pick one significant decision you are currently facing; define the most relevant reference class; look up outcome data for that class
Note how the reference class outcome distribution changes your view of the options in front of you

Day 19 · Outside View Under Pressure 10 min

Time pressure and emotional stakes are precisely when the inside view dominates. When a decision feels urgent, your instinctive first number is driven by case-specific details, not frequency data. Today you practice slowing down enough to consult the base rate even when speed feels essential.

Log one forecast made under time pressure; write the first number that came to you, then look up the base rate for that type of question
Note which number felt more persuasive to act on and write one sentence on why

Day 20 · Merging Inside and Outside Views 10 min

Neither the inside view nor the outside view is sufficient alone. The inside view contains genuinely relevant specific information. The outside view provides a base rate your intuition tends to ignore. A useful forecast uses both, with the outside view as the starting anchor and the inside view as the adjustment.

Take one open forecast; write one specific reason your situation differs from the reference class, if any
Merge the inside and outside views into a single probability range; write one sentence on the process you used

Day 21 · Week 3 Arc Review 15 min

Three weeks in, you carry three habits: the daily forecast, the base-rate lookup, and the range estimate. Today you review the arc so far, not for accuracy but for coverage: which tools have become routine and which still feel effortful? This is the last review before Bayesian updating enters the picture next week.

Count how many forecasts from weeks 1 to 3 used a reference class or outside-view data
Write one type of decision you estimate regularly where you have never used a reference class before

Week 4Bayesian Updating FoundationsDays 22–30

Day 22 · Update Your First Forecast 15 min

A forecast is not final when you write it. New information arrives, circumstances change, and a well-calibrated reasoner updates their probability explicitly rather than just swapping one view for another. This week you add the weekly update habit: revisiting one old forecast, noting new evidence, and writing a revised number alongside the original.

Each week, revisit one forecast you logged 7+ days ago: note what new information arrived and update your probability explicitly, writing both the old and new number
Find a forecast you logged in week 1; write what new information you have received since; update the probability explicitly with a new number
Note the direction and size of your update: did you go up or down in confidence, and by roughly how much?

Day 23 · Priors and Likelihoods 10 min

Bayesian reasoning starts with a prior: your estimate before new evidence arrives. When evidence comes in, you ask how likely that evidence would be if your forecast were true, compared to if it were false. Today you practice naming your prior and identifying one specific piece of evidence to watch for, which sets up the update when it arrives. For example: you predict a meeting will end in agreement. You name your starting probability and your reason for it. You then identify one signal to watch for — whether both parties confirm the agenda beforehand — and note how much that signal would shift your probability if it arrives.

Pick one open question in your log; write your prior probability and a one-sentence reason for it
Identify one specific piece of evidence you could watch for; estimate how much it would shift your probability if it arrived

Day 24 · Updating on Weak Evidence 10 min

Not all evidence warrants a large update. Weak or ambiguous information should produce a small shift. Strong, diagnostic evidence justifies a larger one. The common error is treating all new information as equally compelling, which produces both over-updating and under-updating depending on whether the evidence confirms or challenges your prior.

Log one small piece of evidence about an open forecast; record your probability before and after updating
Write one sentence on whether the evidence was genuinely informative or primarily confirming what you already believed

Day 25 · Updating on Disconfirming Evidence 10 min

Research on judgment consistently shows that people update less readily on evidence that contradicts their prior belief than on evidence that supports it. This asymmetry is one of the most durable patterns in the forecasting literature. Today you look for it in your own log rather than reading about it in the abstract.

Scan your forecast log for any entry where evidence you received contradicted your estimate; note whether you updated and by how much
Write one sentence on what makes disconfirming evidence harder to act on than confirming evidence

Day 26 · Calibrating the Update Size 10 min

One way to calibrate your update is to think through what you would expect to observe if your forecast were true, and what you would expect if it were false. The more the actual evidence matches the true scenario, the larger the update. Today you practice this diagnostic thinking for one open forecast.

For one open forecast, write one scenario that would count as strong evidence in its favor and one that would count as strong evidence against it
Assign rough probabilities to each scenario; note how this exercise changes your current estimate

Day 27 · Anchoring and the First Number 10 min

When you commit to a number early in an estimation process, it exerts a pull on all subsequent adjustments. The anchoring effect occurs even when the anchor is arbitrary or clearly irrelevant. Today you find one estimate where you may have anchored on your gut figure before consulting a relevant base rate.

Find one estimate in your log where you committed to a number before looking up a base rate; find a relevant frequency now and note it
Write whether the base rate changes your number and by how much, and one sentence on what the early anchor hid

Day 28 · Base Rate as Prior 10 min

The most principled starting point for a Bayesian update is the base rate as your prior. You begin with the reference class frequency, then update for the specific features of your case. Today you practice this sequence explicitly and log the full chain, which makes the process reviewable and auditable later.

For one open forecast, use the base rate as your starting prior; add one piece of specific case information; update the probability
Log the full chain in sequence: base rate, specific evidence, updated probability

Day 29 · Reviewing Your Update Pattern 10 min

After a week of explicit updating, you have enough entries to see a pattern. Some people update too much on every new piece of information; others barely move. Neither is well-calibrated. The goal is updates proportional to the actual diagnostic value of the evidence, which requires first knowing your own default behaviour.

Review the forecast you updated on day 22; note whether your update size felt proportional to the strength of the evidence
Write one sentence on your dominant updating pattern: do you tend to move too much, too little, or does it depend on the domain?

Day 30 · Month 1 Review 15 min

Thirty days marks the end of your foundation phase. You have established the core habits that carry forward through the remaining 60 days: the daily forecast, the base-rate lookup, the range estimate, and the weekly update. Today you review what you have logged and write a calibration baseline to measure future progress against.

Review all forecasts logged in month 1; count how many you updated at least once
Write a short summary of your calibration baseline: which domains feel most uncertain and which feel most overconfident

Month 2 — Optimisation · Days 31–60

Sharpen your estimation toolkit with structured drills and reference-class methods

Week 5Estimation DrillsDays 31–37

Day 31 · Weekly Forecast Tally 15 min

The research on forecasting skill shows that feedback is the essential ingredient in calibration training. Tallying resolved forecasts and labeling each as overconfident, underconfident, or well-calibrated gives you the feedback signal that trains accuracy over time. This weekly tally habit runs on your chosen weekly review day from today forward.

On your chosen weekly review day, tally your resolved forecasts from the past week: for each one, note whether you were right, overconfident, or underconfident
Tally this week's resolved forecasts now; label each overconfident, underconfident, or well-calibrated
Note the domain where you were most overconfident and write one word describing the error pattern

Day 32 · Fermi Estimation Basics 10 min

Fermi estimation is the practice of approximating large, unknown quantities by breaking them into smaller sub-quantities that are easier to estimate. The method builds calibration because it forces you to surface your assumptions rather than generate a number from vague intuition. When you combine sub-estimates, you also see which components carry the most uncertainty.

Pick one large unknown quantity such as how many decisions you make in a typical day or how many steps you take in a week; estimate it by breaking it into smaller sub-quantities
Log all sub-estimates and the combined total; circle the sub-estimate you are least confident about

Day 33 · Decomposing Uncertainty 10 min

Complex forecasts contain multiple sources of uncertainty, each with its own range. When you estimate the combined outcome without decomposing the components, you routinely underestimate total uncertainty. Breaking a forecast into parts and combining the probabilities produces a more honest overall estimate than treating the question as a single unit.

Take one current forecast; break it into 3 sub-questions, each with its own probability
Combine the sub-probabilities into a single estimate; compare to your original gut estimate and note the difference

Day 34 · Confidence Interval Practice 10 min

A 90 per cent confidence interval means you are genuinely willing to say the true answer falls within your stated range 90 per cent of the time. Most people set intervals that are too narrow: their stated 90 per cent ranges contain the truth far less often. Today you practice setting an interval that is honestly wide enough, which often feels uncomfortable.

For one numerical forecast today, write a 90 per cent confidence interval: the low end and high end you are 90 per cent sure contains the true answer
Note whether your interval feels comfortably wide or narrower than it probably should be; write one reason you might be tempted to narrow it

Day 35 · Calibrating on Your Track Record 10 min

Your personal track record in a domain is direct evidence about your calibration in that domain. If you have been consistently overconfident on project timelines, that history is a data point about how to set future estimates. Today you find one domain where you can look backward at your accuracy and compare it to your current confidence level.

Find or estimate your accuracy rate in one domain where you can look backward: past project timelines, hiring outcomes, predictions about events
Note whether your current confidence level in that domain matches your historical accuracy; write what any discrepancy suggests

Day 36 · Estimation Under Time Pressure 10 min

Time pressure pushes estimates toward your fastest accessible number, which is almost always an inside-view, instinct-driven figure. Practicing fast estimation and then checking accuracy afterward builds awareness of how much your precision changes under speed, and in which direction.

Give yourself 60 seconds to estimate 5 different uncertain quantities; write all 5 numbers quickly, then check them
Note whether your speed estimates were better or worse calibrated than your deliberate estimates earlier this week

Day 37 · Estimation Drills Review 15 min

Week 5 was a sequence of estimation drills designed to expose your calibration habits under different conditions. Today you review the full set, not for accuracy scores but for patterns: which conditions pushed your estimates widest, which narrowest, and where your confidence felt least grounded in actual evidence.

Review this week's estimation drills; count how many produced a range vs a single point estimate
Write one sentence on the condition or domain that consistently produced your widest confidence intervals

Week 6Reference-Class ForecastingDays 38–44

Day 38 · Finding Reference Classes Systematically 10 min

A good reference class is not the first one that comes to mind. The discipline is generating multiple candidates from broad to narrow, then selecting the most informative one. Too broad and the base rate is generic; too narrow and you have sample size problems. Today you practice this selection process explicitly.

Pick one current forecast; write 3 candidate reference classes from broad to narrow; select the most informative one
Note the base rate for the chosen class and use it as the starting point for your forecast range

Day 39 · Project Reference Classes 10 min

Projects are where the planning fallacy is most studied and most consistent. The reference class for your current project exists whether you consult it or not. The only question is whether you let the base rate inform your estimate or rely entirely on the optimistic inside view.

Name one ongoing project; find the reference class for that type of project; write the typical timeline and success rate data you can locate
Compare to your inside-view estimate; note the gap and whether it surprises you

Day 40 · Reference Classes Beyond Work 10 min

Reference-class forecasting applies outside professional projects: job searches, home renovations, personal decisions, financial plans. The personal planning fallacy is just as strong as the professional one. Today you apply the outside view to a personal-life forecast that you might normally estimate entirely from intuition.

Pick one personal-life forecast such as how long a purchase will stay useful or how long a decision process will take; find a reference class
Log the reference class data and update your estimate to reflect the outside view

Day 41 · Narrow vs Broad Reference Classes 10 min

The choice between a broad and a narrow reference class involves a trade-off between sample size and relevance. A very broad class has lots of data but may not match your situation closely. A very narrow class may match closely but have too few examples to trust. Today you explore both ends of that trade-off for one open forecast.

For one open forecast, write both a broad reference class and the narrowest class that still has data; note the base rate for each
Write one sentence on which produces the more useful forecast and why

Day 42 · Reference Class and Motivated Reasoning 10 min

When the reference class produces a worse outcome than you want, the temptation is to argue your case is special. Sometimes it is. The test is whether you can articulate a specific, verifiable feature of your situation that genuinely changes the base rate, rather than a general feeling of being different from the group.

Pick one forecast where the reference class outcome is worse than your gut estimate; write one sentence on whether your situation genuinely differs from the reference class
Rate the strength of your exception case from 1 to 5; note whether it is strong enough to justify departing from the base rate

Day 43 · Tracking Reference-Class Accuracy 10 min

By now you have several forecasts anchored on reference classes from weeks 2, 3, and 6. Some of those forecasts may have resolved. Today you check those resolutions against the reference class prediction to see whether the outside view gave you a better anchor than your inside-view estimate.

Review forecasts you anchored on a reference class in weeks 2 and 3; note how the outcomes compared to the reference class prediction
Write one sentence on whether the outside view produced more accurate forecasts than your inside-view estimates in the cases you can check

Day 44 · Outside View in Groups 10 min

Group decisions are especially vulnerable to inside-view bias because groups often generate shared enthusiasm that overrides outside-view data. The person who introduces a reference class in a group setting contributes accuracy. Today you think about how to bring this practice into a collaborative setting.

Recall one recent group decision; compare the group's estimate to the reference class outcome for that type of decision
Write one sentence on how you could introduce a reference class at the next group decision point without disrupting the process

Week 7Confidence IntervalsDays 45–51

Day 45 · Pre-Mortem Habit 15 min

A pre-mortem is a structured way to use the outside view at the planning stage. Before committing to a plan, you imagine it has already failed and ask: what is the single most likely reason? This exercise surfaces the reference class failure mode before it happens rather than after, and it takes less than a minute.

Before committing to any plan with more than two steps, write a one-sentence pre-mortem: what is the single most likely reason this does not go as expected?
Apply the pre-mortem to one current plan; write the single most likely failure reason in one sentence
Note whether identifying that reason changes any step in the plan

Day 46 · Interval Width by Domain 10 min

Your confidence interval width should vary with your actual uncertainty. In domains where you have deep knowledge and reliable data, narrower intervals may be honest. In unfamiliar domains, wide intervals are the appropriate response. Today you test whether your intervals track your actual familiarity or stay artificially consistent.

Write 90 per cent confidence intervals for 3 questions in domains with different familiarity levels: one familiar, one moderate, one unfamiliar
Note whether your intervals widen appropriately as familiarity decreases

Day 47 · When Intervals Are Too Narrow 10 min

The most common calibration error is setting intervals that are too narrow. People feel uncomfortable with very wide ranges because they imply uncertainty, which can feel like ignorance. But an honest 90 per cent interval should be wide enough to contain the true answer nine times in ten. Today you check recent entries for this pattern.

Look back at 3 recent forecasts you logged as ranges; estimate whether each interval was wide enough to contain the true outcome
Note the proportion that missed; write one cause of the under-coverage you observe

Day 48 · Calibration on Known Quantities 10 min

Calibration training works best when you can check answers promptly. Questions with verifiable answers let you close the feedback loop the same day. Today you run a short calibration exercise using factual questions where you can check your accuracy immediately, which is the tightest feedback signal available.

Take 10 questions with verifiable answers; assign a probability to each; check your accuracy
Compare your accuracy in the high-confidence set vs the low-confidence set; note whether the two numbers track each other

Day 49 · Structured Intervals for Planning 10 min

Planning under uncertainty becomes more honest when you replace a single target with a range of outcomes. Best case, expected case, and worst case cover the reference class distribution and force you to think about the tails, not just the middle. The worst case is the scenario most often omitted in unstructured planning.

For one upcoming plan, write a best case, expected case, and worst case; assign rough probabilities to each scenario
Note which case is most often omitted when you plan without this structure

Day 50 · Interval Practice in Your Log 10 min

The range estimate habit was introduced in week 3, but under pressure it is easy to slip back to single-number forecasts. Today you audit this week's entries and convert any single-point estimates back into ranges. The width of the range is the data; a single point estimate discards information.

Review this week's forecasts in your log; count how many used a range and how many used a single number
Convert any single-number forecasts from this week into three-point ranges: low, best-guess, high

Day 51 · Resolved Forecast Review 10 min

Feedback requires resolution. A forecast that never resolves never trains you. Today you find entries that have reached their outcome date and close them with a verdict. Each resolution adds a data point to your calibration record and closes the loop that makes the practice compound over time.

Find 5 forecasts in your log that have now resolved; compare your stated confidence to each outcome
Write one verdict per resolved forecast: well-calibrated, overconfident, or underconfident

Week 8Forecasting PracticeDays 52–60

Day 52 · Bayesian Update Routine 10 min

You are now into month two of forecast-logging. Some of your month 1 entries are still open, and evidence has arrived since you logged them. Today you do a systematic review of all open forecasts and update at least two with current information, following the chain: prior, evidence, posterior.

Review all open forecasts in your log; note which ones have received new evidence since your last update
Update at least 2 forecasts with new probability numbers; write the specific evidence that drove each update

Day 53 · Tracking Prior vs Posterior 10 min

A logged update chain shows you how your thinking evolved: what you believed before the evidence, what evidence arrived, and what you concluded. This chain is the record that lets you evaluate not just outcomes but the quality of your reasoning process at each step, which is harder to reconstruct from memory alone.

For one open forecast, write the full update chain: your prior, the evidence received, and your current posterior estimate
Note whether you were surprised by how much or how little the evidence moved your number

Day 54 · Direction of Updates 10 min

Updates should move in the direction the evidence points. If new information makes your forecast more likely to be true, your probability should rise; if it makes it less likely, it should fall. A simple check of update direction catches motivated reasoning before it becomes a pattern in your log.

Look through this week's evidence entries; for each piece of evidence, confirm whether your update went in the correct direction
Note any case where you moved opposite to the evidence and write one sentence on why

Day 55 · Resistant Beliefs 10 min

Some forecasts resist updating no matter how much evidence arrives. This can be appropriate when the evidence is genuinely weak. It becomes a problem when the resistance reflects motivated reasoning: you are not updating because you do not want to be wrong. Today you identify one candidate and examine it honestly.

Identify one forecast where you have received multiple pieces of evidence but your probability has barely moved; write what would need to be true to shift you significantly
Note whether your resistance reflects genuine evidence weakness or a preference for your current belief

Day 56 · New Forecasting Domains 10 min

Calibration practice transfers across domains, but it builds fastest where you can get feedback. Today you expand your forecast log into one domain you have not yet tried, where outcomes are observable and can be checked within a reasonable time frame. Stretching into unfamiliar territory is where much of the learning happens.

List 5 question types you have not yet forecasted; pick one from a domain where outcomes will be checkable soon
Log a forecast in that new domain; note what base rate or reference class you used to anchor the probability

Day 57 · Compounding Uncertainty 10 min

Multi-step forecasts, where several events must all occur for the outcome to happen, compound uncertainty at each step. People routinely underestimate how unlikely multi-step outcomes are because they focus on each step in isolation rather than the joint probability. Today you calculate one chain to see the result.

Identify one multi-step forecast where several things must all go right; write a separate probability for each step in the chain
Multiply all the step probabilities to get a joint probability; note whether the result is lower or higher than your intuitive estimate

Day 58 · Reviewing Update Sizes 10 min

Update size is a calibration variable on its own. Consistent under-updaters hold beliefs too rigidly; consistent over-updaters chase every new piece of information with a large swing. Your update history now gives you data on your own pattern across multiple domains and evidence types.

Look at the last 4 times you updated a forecast; note the approximate size of each update: small, medium, or large
Write one sentence on whether your average update size feels proportional to the evidence received or tends to be systematically too small

Day 59 · Month 2 Calibration Assessment 10 min

Month 2 is nearly complete. Today you run a preliminary calibration check on this month's entries: how are your resolved forecasts breaking down across error types, and is there a consistent pattern worth noting before tomorrow's full review?

Review month 2 forecasts; count overconfident, underconfident, and well-calibrated entries
Note the most common error type; write one word describing the pattern

Day 60 · Month 2 Review 15 min

Month 2 deepened your estimation toolkit: Fermi decomposition, confidence intervals, reference-class forecasting, and systematic Bayesian updating. Today you write a short comparison between your calibration in month 2 and month 1, using your logs as evidence rather than impressions. Comparison across months is what reveals the trend.

Write a short comparison of your calibration across month 1 and month 2; use specific numbers or observations from your log entries
Identify the domain where your accuracy is strongest and the domain that needs the most work going into month 3

Month 3 — Mastery · Days 61–90

Integrate calibrated reasoning into durable everyday decision-making

Week 9Calibration ReviewDays 61–67

Day 61 · Monthly Forecast Resolution 15 min

The monthly resolution habit closes out the previous months' open forecasts, converting them from predictions into data. Each resolved forecast adds a point to your calibration record, which is the foundation of the accuracy tracking that makes this practice compound over time. Today you resolve everything from months 1 and 2 that has a known outcome.

On the first day of each new month, resolve all forecasts from the previous month: log the outcome and mark each as well-calibrated, overconfident, or underconfident
Resolve all still-open forecasts from months 1 and 2 where the outcome is now known; mark each entry
Count the totals and write your overall calibration ratio for the first two months

Day 62 · Domain-Level Calibration 10 min

Calibration error is not uniform across domains. You may be well-calibrated in one area and systematically overconfident in another. Grouping your resolved forecasts by domain is the first step to understanding where your practice needs the most targeted work in month 3.

Group your resolved forecasts by domain: work, personal plans, current events, or other categories that fit your log
Identify the domain with the worst calibration; write one hypothesis for why that domain is harder to forecast accurately

Day 63 · Systematic Bias Check 10 min

Most forecasters have a dominant error direction: systematically overconfident, systematically underconfident, or a domain-specific mix. Knowing your direction is a prerequisite for designing the correct procedural counter. Today you check yours across all resolved forecasts rather than from memory.

Review error direction across all your resolved forecasts; note whether you skew overconfident or underconfident overall
Write one procedural change to counteract your dominant error direction for the remaining weeks

Day 64 · Hard vs Easy Accuracy 10 min

Research on calibration shows a consistent asymmetry: people tend to be overconfident on hard questions and underconfident on easy ones. The relevant question for your log is whether this pattern matches your own data, and whether your difficulty ratings at forecast time actually predicted your eventual accuracy.

Sort your resolved forecasts into questions you found easy vs hard to estimate at forecast time; compare accuracy in each category
Note whether your difficulty rating at forecast time was a good predictor of your actual accuracy

Day 65 · Full Three-Check Sequence 10 min

By week 9 you carry a full procedure: base-rate lookup, outside-view reference class, and Bayesian update. Today you run all three checks in sequence on a single new forecast. This is the closest thing to the full calibrated reasoning routine you have been building toward over the past 64 days.

For one new forecast today, run the full sequence: base rate first, reference class second, Bayesian update third; write each step explicitly
Note which step changed your estimate most and write one sentence on why

Day 66 · Feedback Loop Audit 10 min

A forecast log with no resolution dates is a wishlist, not a calibration tool. The feedback signal that trains accuracy requires both a prediction and a known outcome. Today you check how many of your open forecasts have a defined resolution criterion, and fix any that do not.

Review your open forecasts; count how many have a clear resolution criterion vs how many are vague or open-ended
Write a resolution criterion for any open forecast that currently lacks one

Day 67 · Month 3 Launch Prep 15 min

Month 3 is the integration phase. The habits are all running. The skills are built. What remains is weaving them into a coherent practice that runs on real decisions, not just exercises. Today you write one concrete goal for each of the three remaining weeks and confirm your habit stack is intact.

Write one specific goal for each remaining week: a calibration target for week 10, an updating target for week 11, and an integration target for week 12
Review your active habit stack and confirm each of the 8 habits is still running; flag any that have drifted

Week 10Advanced UpdatingDays 68–74

Day 68 · Base-Rate Departures 10 min

Every time you departed from the base rate to make an estimate, you made an implicit claim: your case is genuinely different from the reference class. Week 10 is the time to audit that claim for the departures you logged in months 1 and 2 and see how they held up against actual outcomes.

Find one forecast where you departed significantly from the base rate; write the evidence that justified the departure
Score the evidence quality as weak, moderate, or strong; note whether the departure was warranted in retrospect

Day 69 · Long-Horizon Updating 10 min

Some of the forecasts you logged in week 1 have now aged by more than 60 days. They have absorbed multiple rounds of evidence. Doing a comprehensive update on one of these long-horizon entries is qualitatively different from a single-step update: you are synthesising an evidence trail rather than responding to one piece of news.

Find a forecast logged in week 1 that has not yet resolved; write a comprehensive update using all evidence received over the past two months
Note what drove the single largest shift in probability and whether you updated in real time or only now

Day 70 · Outside View for Life Decisions 10 min

The most consequential decisions in your life are also the ones where the inside view feels most compelling. The stakes feel personal, the details feel unique, and the reference class can feel irrelevant. It rarely is. Today you apply the outside view to a decision at life scale, where the procedure matters most.

Pick one significant decision you expect to face in the next 12 months; define the most appropriate reference class
Write the base rate outcome for that reference class and a one-sentence note on how your situation differs, if at all

Day 71 · Multi-Stage Bayesian Chains 10 min

Multi-stage forecasts require chaining individual probabilities together. Each link in the chain introduces uncertainty. The joint probability of a multi-step outcome is often surprisingly low, which reflects what the research on probability compounding and multi-step chain effects shows. Today you build one full chain and compare it to your intuition.

Take one complex forecast that requires multiple sequential events; write a separate probability for each event in the chain
Multiply all the probabilities for a joint probability; note whether the result is lower or higher than your initial intuitive estimate

Day 72 · Evidence Quality Scoring 10 min

Not all evidence is equal, and sizing an update requires an honest assessment of how diagnostic the evidence actually is. Anecdotes and single data points warrant smaller updates than systematic data. Today you score the evidence quality behind your recent updates and check whether the update sizes were proportional.

Review your last 5 updates; rate the quality of the evidence for each as weak, moderate, or strong
Note whether your update sizes were proportional to the evidence quality ratings you just assigned

Day 73 · Changed Situations vs Updated Beliefs 10 min

There is a difference between updating your probability because new evidence arrived and restarting your forecast because the situation itself changed. Conflating the two produces confusion about whether your current estimate is an update or a fresh prior. Today you find one example of each type in your log and name the distinction.

Identify one forecast where the situation itself changed rather than just your view; write a new forecast from scratch for the new situation
Write one sentence on the difference between an update and a restart and why it matters for interpreting your log

Day 74 · Outside View on Your Own Calibration 10 min

You have now accumulated enough personal data to compare yourself to the research baseline. The general finding is systematic overconfidence on hard questions and underconfidence on easy ones, with overconfidence being the dominant pattern. How does your own log match that picture?

Compare your calibration pattern to the published finding: overconfident on hard questions, underconfident on easy ones; note where your pattern matches
Write one sentence on where your personal pattern diverges from the general finding, if it does

Week 11Systems IntegrationDays 75–81

Day 75 · Your Personal Decision Checklist 15 min

By now you have built a full reasoning toolkit: base-rate lookup, reference class selection, range estimate, pre-mortem, and Bayesian update. A personal checklist crystallises this toolkit into a form you can run before any consequential decision. Today you write that checklist and put it somewhere accessible.

Write a 3-to-5 step decision checklist capturing your core calibration procedure: base rate, range, outside view, pre-mortem, and update
Write the checklist in your forecast log where you can find it before your next consequential decision

Day 76 · Applying the Checklist 10 min

A checklist is useful only when you use it. Today you apply it to a real decision, not a practice exercise. The goal is to see which steps feel natural, which feel effortful, and whether you are tempted to skip any of them under real conditions rather than practice ones.

Apply your decision checklist to one real decision you face today; log the output at each step
Note any step you were tempted to skip and write one sentence on why

Day 77 · Forecasting New Domains 10 min

Calibration skill is most valuable in the domains that matter most to you personally. By week 11, your practice has probably clustered around a few familiar areas. Today you expand deliberately into one area of your life where you have not yet kept a systematic forecast log.

Pick one area of your life where you have not yet forecasted systematically; log 3 new forecasts in that domain
Note the base rate or reference class you used for each forecast, and flag where you had to estimate from scratch with no external anchor

Day 78 · High-Stakes Calibration Practice 15 min

The full procedure is most valuable on consequential decisions: the ones where being well-calibrated changes what you decide or how you prepare. Today you apply every tool you have built to one such decision, running the complete sequence from outside view through pre-mortem in a single sitting.

Identify one consequential decision coming up in the next 30 days; apply the full procedure: outside view, base rate, range estimate, pre-mortem
Log the complete procedure output as a single entry with all four steps visible

Day 79 · Base Rates for Social Predictions 10 min

Social predictions, how people will respond, what groups will decide, how relationships will develop, are subject to the same base-rate neglect as other predictions. We focus on specific personalities while underweighting how often similar situations resolve in typical ways. Today you find one social base rate and check it against your prediction.

Pick one prediction about how another person or group will respond to a situation; find a base rate for that type of response
Compare your prediction to the base rate and note any gap; write one sentence on what drives the difference

Day 80 · Calibration Drill: Verifiable Questions 10 min

With two months of calibration data in your log, today's drill lets you compare your current accuracy to where you started. The aim is not a high score but a trend: are your intervals getting more honest, and are your confidence levels tracking closer to your accuracy rates than they did in week 1?

Take 10 factual questions with verifiable answers; assign a probability to each; check and score your accuracy
Compare your overconfident, underconfident, and well-calibrated counts to your scores from month 1; note any shift in the pattern

Day 81 · Full Habit Stack Review 15 min

Eight habits are now running in parallel. Some will be strong and automatic; others may have drifted or shortened over time. Today you review the full stack, which is a routine you can carry forward beyond the programme if you choose to sustain the practice.

Go through all 8 active habits in sequence and confirm each is running; note any that have drifted or shortened over time
Write the one habit running strongest and the one that needs deliberate re-activation

Week 12Mastery and SustainingDays 82–90

Day 82 · Long-Range Forecast Audit 10 min

Week 12 begins with closing the feedback loop on your earliest entries. The forecasts from weeks 1 to 4 were made when the habits were newest and least practiced. Resolving them now completes the record and gives you a comparison point between your earliest and most recent calibration.

Resolve all forecasts from weeks 1 to 4 that have reached their outcome dates; mark each as well-calibrated, overconfident, or underconfident
Note the total resolution count for the programme and whether your earliest entries were more or less accurate than your recent ones

Day 83 · Final Base-Rate Check 10 min

The base-rate habit was introduced on day 3. Ten weeks later, you can observe whether it has become automatic or still requires deliberate effort. Today you note honestly where you stand with this specific habit, not as a self-assessment but as a data point for designing your sustaining routine.

Pick one recurring decision type in your life; look up the base rate for that type of outcome one final time
Write whether you now anchor on the base rate before every estimate of that type, or whether you still generate from instinct first

Day 84 · Outside View in Review 10 min

The outside view was introduced in week 3. Today you review all the forecasts in your log where you used a reference class, checking how often the reference class prediction turned out to be a better anchor than your inside-view estimate. This is the clearest test of whether the habit delivered value.

Review all forecasts in your log that used a reference class; note what proportion of resolved entries landed within the reference class range
Write one sentence on the value the outside view added to your forecasting across this programme

Day 85 · Bayesian Updating Final Drill 10 min

Week 12's Bayesian drill targets your most uncertain open forecast: the one where you have the least clarity and the most unprocessed evidence waiting to be synthesised. Running a thorough update on this entry tests whether the updating routine has become a natural part of how you approach uncertain questions.

Take the open forecast you are most uncertain about; write a full evidence review; update the probability; set a clear resolution criterion
Note whether the evidence review changed your number and by how much

Day 86 · Final Calibration Score 10 min

Your calibration record now spans close to 90 days. Today you compute your final summary score: across all resolved forecasts, what proportion were you right about at each confidence level? The result is not a grade. It is a map of your calibration profile going forward.

Resolve all remaining forecasts where outcomes are now known; calculate your final calibration ratio: correct answers in the high-confidence set vs the low-confidence set
Write your verdict: systematically overconfident, underconfident, or roughly calibrated, with one supporting observation from your log

Day 87 · Designing the Sustaining System 15 min

A 90-day programme with no sustaining plan ends when the programme ends. The research on forecasting skill shows that practice must continue after training for accuracy gains to hold. Today you design the minimum viable sustaining routine that fits your actual life, not an idealised version of it.

Write a post-programme routine: how often to log new forecasts, how often to review resolved ones, and when to run a monthly calibration check
Write the routine as a one-paragraph note in your forecast log where you can find it in month 4

Day 88 · Integration Test 15 min

Day 88 is a live test of the full system. You face a real decision today, one that matters to you, and you run the complete procedure without any scaffolding: base rate, range estimate, outside view, pre-mortem, and Bayesian update. The question is which steps feel automatic and which still require deliberate attention.

Face one real decision today and apply the full procedure without any prompting: base rate, range, outside view, pre-mortem, and explicit update
Log the full procedure output and note which steps felt automatic and which still required deliberate attention

Day 89 · Final Habit Stack Confirmation 10 min

Tomorrow is the final day. Today you review your complete 8-habit stack for the last time under programme conditions. The question is not whether all 8 are perfect. It is which ones have become durable enough to carry forward without a structured programme driving them.

Review your complete 8-habit stack one final time; note which habits are now automatic and which still require deliberate attention to maintain
Write one habit from the programme that you will carry forward indefinitely

Day 90 · Programme Complete 15 min

Ninety days of forecast-logging, base-rate lookups, reference class forecasting, range estimates, pre-mortems, and Bayesian updates. The habits are built. The record is complete. Today you write the closing entry: not a summary of what you did, but a note on how you think about uncertainty differently than you did on day 1.

Resolve all remaining forecasts where outcomes are known; write a brief note on your calibration trajectory from month 1 to month 3
Describe one specific decision you approached differently because of this programme, and what the procedure told you that instinct alone would have missed

The evidence

The frameworks this programme is built on — each anchored to peer-reviewed research, cited with its DOI and verified against the Crossref registry at build time. Where the evidence has honest limits, the framework says so.

Calibration

Lichtenstein & Fischhoff · Organizational Behavior and Human Performance · 1977

The foundational study of calibration shows that stated confidence levels consistently exceed actual accuracy rates. Lichtenstein and Fischhoff found that even knowledgeable people are often poorly calibrated: someone who says they are 90 per cent sure is correct far less than 90 per cent of the time. The key insight is that more knowledge does not automatically produce better calibration about the limits of that knowledge. Building calibration requires deliberate practice with feedback, not simply accumulating expertise.

DOI: 10.1016/0030-5073(77)90001-0

Moore & Healy, 2008 — Distinguishes three forms of overconfidence: overestimation of absolute performance, overplacement relative to others, and overprecision about the accuracy of beliefs. Shows the pattern varies by task difficulty and domain. doi:10.1037/0033-295X.115.2.502

Base-Rate Sensitivity

Kahneman & Tversky · Psychological Review · 1973

Kahneman and Tversky's research on prediction showed that people systematically underweight statistical prior probabilities when case-specific information is available. When a description fits a stereotype, people base predictions almost entirely on the description and largely ignore how common the outcome actually is. This tendency to neglect the base rate is one of the most replicated findings in judgment research and applies across domains from medicine to project planning.

DOI: 10.1037/h0034747

Bar-Hillel, 1980 — Provides a systematic account of when and why people neglect base rates, showing that causal relevance and specificity of case information drive the degree of neglect. doi:10.1016/0001-6918(80)90046-3

Outside View

Kahneman & Lovallo · Management Science · 1993

Kahneman and Lovallo contrast the inside view, which focuses on the specific details of a current plan and produces optimistic forecasts, with the outside view, which asks how similar plans have fared historically. Their research on planning decisions shows that the inside view consistently generates overly bold forecasts while the outside view produces more realistic ones. Switching to the outside view by identifying a reference class and looking up its outcome distribution is one of the most robust debiasing strategies available.

DOI: 10.1287/mnsc.39.1.17

Flyvbjerg, 2006 — Applies reference class forecasting to major infrastructure projects, showing that inside-view estimates are systematically optimistic and that outside-view anchoring produces more accurate cost and schedule predictions. doi:10.1177/875697280603700302

Forecasting Training

Mellers et al. · Perspectives on Psychological Science · 2015

Research on forecasting tournaments identified a subset of consistently accurate forecasters and studied what distinguished them. The superforecaster research found that accurate forecasters share identifiable cognitive habits: they think in probabilities, update frequently, seek out disconfirming evidence, and maintain calibrated uncertainty rather than strong fixed positions. Critically, this research also showed that structured training in probabilistic reasoning produces measurable accuracy gains, suggesting that forecasting skill is trainable rather than fixed.

DOI: 10.1177/1745691615577794

Chang et al., 2016 — Demonstrates that training and practice both contribute independently to forecasting accuracy in geopolitical prediction tournaments, with training effects persisting beyond the initial training period. doi:10.1017/s1930297500004599

Bayesian Updating

Tversky & Kahneman · Science · 1974

Tversky and Kahneman's landmark paper on judgment under uncertainty documented the systematic heuristics people use when assessing probabilities. The anchoring-and-adjustment heuristic is particularly relevant to Bayesian updating, producing a pattern consistent with under-revision of posterior beliefs: starting from a prior estimate, people tend to adjust too little when new evidence arrives. New evidence produces smaller revisions than the evidence warrants, a pattern that persists across domains and expertise levels. Explicit tracking of prior estimates, incoming evidence, and revised posteriors is the procedural correction.

DOI: 10.1126/science.185.4157.1124

Gigerenzer & Edwards, 2003 — Demonstrates that natural frequency formats and visual representations dramatically improve Bayesian reasoning performance, with implications for how probability information should be presented and consumed. doi:10.1136/bmj.327.7417.741

Safety & fit

This 90-day protocol is a structured, educational programme in probability estimation and calibrated reasoning. It is not medical advice and is not a substitute for professional guidance. Every framework it draws on is referenced to peer-reviewed literature for educational context only.

Questions, answered

How much time does this take?

The Probabilistic Thinking Protocol runs 90 days across 12 themed weeks. Each day asks for between 10 and 15 minutes of deliberate practice — the daily tasks are listed in the programme map on this page, so you can read the whole arc before you start.

How does the programme adapt to my week?

A weekly check-in reads the week you actually logged — not the week the plan assumed. Depending on the week you actually had, it lands on one of 5 adjustments: Sparse week; Expand your forecast log; Calibration needs attention; Strong calibration week; On track. Milestone reviews at day 30, day 60, day 90 read your whole record so far.

Where is my progress stored?

Entirely in this browser. Your daily log, notes, and check-ins live in local storage on this device — no account, no server, and nothing you type is transmitted. You can export your full log as a CSV file or delete everything, at any time, from the My log tab.

Is this medical advice?

No. This is a structured, educational programme built on peer-reviewed research — it is not medical advice, therapy, or a substitute for professional guidance. The full disclaimer, and every framework the programme draws on with its DOI, is on this page.

What happens if I miss days?

The programme is built to absorb real weeks. The weekly check-in works from whatever you logged: a thin week adjusts the plan rather than failing you, and no adjustment ever deletes your record. The adaptation rules above describe exactly how each kind of week is handled.

Where does the evidence come from?

Every phase of the programme is grounded in 5 evidence frameworks — 10 peer-reviewed references in total, each cited with its DOI in the evidence section on this page and verified against the Crossref registry at build time. Where the evidence has honest limits, the framework says so.

Go deeper

Hand-picked companions to this programme from across the estate — the baseline tool, the research deep dives, and the terms worth two minutes each.

assessment The Probabilistic Thinking Assessment A baseline reading before day 1 — retake it after the programme and compare.
deep-dive The Overconfidence Effect Why experts are wrong more often than they think — the calibration research.
deep-dive The Placebo Effect What belief does to outcomes, and what that means for reading evidence.
glossary Second-Order Thinking The consequences-of-consequences discipline.

Programme record

A living programme keeps a public record. Any future change to the daily tasks, the check-in rules, or their thresholds bumps the version and lands here.

v1.0 · July 2026Initial published programme: 90 daily practices across three phases, weekly adaptive check-ins, and milestone reviews at days 30, 60 and 90.