What decision actually exists?
Define scope, horizon and objective.
A badly framed problem can make every later calculation precise but irrelevant.
Side 12
Decision
Science
A study of choosing when outcomes are uncertain, objectives compete and information is incomplete. The aim is not to eliminate judgment, but to make assumptions, trade-offs and updating more inspectable.
state→options→uncertainty→choice→update
05decision stages
06bias patterns
05strategic games
12Side
Separate the decision from the outcome.
A good decision can produce a bad outcome because uncertainty remains after the choice. Judge the process by what was knowable at the time, then use the outcome as new evidence.
What can be done?
Include delay, experiment and no-action.
Option generation should happen before option evaluation to avoid premature narrowing.
What is not known?
States, probabilities and missing information.
Distinguish uncertainty that can be reduced from uncertainty that must simply be carried.
What matters and by how much?
Outcomes need weights.
Money, time, reversibility, learning, downside and strategic position may all matter simultaneously.
What did reality teach?
Outcome ≠ verdict.
Compare result with expectation, revise beliefs and improve the next decision rule.
Decision qualitygood process + uncertainty ≠ guaranteed good outcome
Choices trade outcomes against probabilities.
Expected value is a starting point, not a complete theory of choice. Risk attitude, irreversibility, tail losses and information value can change the preferred action.
Probability-weighted outcome.
Useful when repeated exposure, linear value and manageable downside make averages decision-relevant.
Value need not be linear.
The same monetary gain or loss can matter differently depending on wealth, objectives and risk tolerance.
Can the choice be undone?
Reversible choices often deserve faster experimentation; irreversible choices justify more information and stronger safeguards.
Preserve future choices.
An action can be valuable partly because it keeps multiple future paths open.
Would more knowledge change the choice?
Information is worth acquiring when its expected effect on decisions exceeds its acquisition cost.
Compare against foregone alternatives.
Regret-based reasoning focuses on how painful it would be to learn that another option would have performed better.
Forecasts should be calibrated, not theatrical.
The goal is not certainty. It is probabilities that become meaningful over repeated judgments and can be updated as evidence arrives.
Base rate
Start outside the case.
Ask how often similar situations produce the outcome before focusing on unique details.
Decompose
Break large questions apart.
Forecast component events or drivers when the whole question is too opaque.
Calibrate
70% should happen about 70% of the time.
Track whether stated confidence corresponds to observed frequency across many forecasts.
Update
Move probabilities when evidence changes.
New information should change belief in proportion to its diagnostic value, not its vividness.
Resolve
Define what counts as an outcome.
Ambiguous resolution criteria make later scoring meaningless.
Confidence without resolution is just rhetoric.
Forecasting improves when predictions are recorded, dated and later scored against a clear outcome.
Fast rules are useful — and patterned.
Heuristics reduce cognitive cost. The same shortcuts that enable rapid judgment can also create predictable distortions under the wrong conditions.
Easy to recall feels common.
Vivid or recent examples can overweight perceived frequency.
Initial values pull estimates.
An arbitrary or weakly relevant starting point can influence later judgments.
Losses can loom larger than gains.
Reference points affect how equivalent changes are psychologically valued.
Search becomes asymmetric.
Evidence consistent with an existing belief may receive more attention or less scrutiny.
Past cost contaminates future choice.
Irrecoverable expenditure is allowed to justify continued investment even when future prospects have changed.
Result rewrites process quality.
Known outcomes can make a previously uncertain decision look obviously wise or foolish.
Bias lists are not mind-reading tools.
The useful question is whether a decision process contains conditions known to produce a distortion, not whether another person can be casually diagnosed with a named bias.
Your outcome may depend on someone else’s choice.
Game theory studies interdependent decisions: each actor’s best move can depend on what others do, know and expect.
| Game structure | Central tension | Useful question | Common setting |
|---|---|---|---|
| Prisoner’s dilemma | Individual incentive can undermine joint welfare. | What supports cooperation? | Repeated cooperation, arms races, shared resources. |
| Coordination game | Multiple compatible outcomes may exist. | How do expectations converge? | Standards, conventions, platform adoption. |
| Chicken | Each side wants the other to yield. | How credible is commitment? | Bargaining and escalation. |
| Zero-sum | One side’s gain equals the other’s loss. | Which strategy is robust against an opponent? | Some competitive contests and games. |
| Principal–agent | Delegation + asymmetric information. | How can incentives align behavior? | Employment, contracting, governance. |
Decision lab.
Use compact scenarios to separate beliefs, values and action thresholds.
A reversible project has a 40% chance of success and cheap failure.
Do not ask only whether 40% sounds low. Estimate the upside, downside, learning value, cost of delay and whether a small pilot can cheaply resolve uncertainty.
A vendor offers a discount if you commit today.
Separate the economic value of the discount from the decision-pressure effect of the deadline. Ask whether the offer changes the underlying expected value or merely compresses your review time.
A strategy worked three times in a row.
Do not let outcome streaks replace process evaluation. Ask how many alternative strategies could have produced similar results, what base rate applies and what evidence would falsify the current rule.
You can buy one week of research before a major irreversible decision.
Estimate value of information: how likely is research to change the chosen option, and how large is the difference in value between the options it might switch you between?
Thinking, Fast and SlowDaniel Kahneman · judgment and heuristics
SuperforecastingPhilip Tetlock & Dan Gardner · calibrated forecasting
Decision AnalysisHoward Raiffa · formal decision framing
Thinking in BetsAnnie Duke · process versus outcome