Principle-to-Algorithm Decision Loop
Write decision criteria, test them historically, and encode what survives
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 99%
The loop begins before a decision by writing down the criteria expected to make it succeed. After the outcome, the decision maker conducts a postmortem and refines the criteria into an explicit principle. If the rule is clear enough, it can be tested against historical cases, different environments, and adverse periods to determine where it works or fails. Repeatable logic can then be encoded in a computer that processes incoming information consistently, quickly, and without the same emotional pressure as a person. The human retains an essential role: whenever judgment and system output diverge, investigate why. The computer may have processed information the person missed, or the person may recognize a factor that was never encoded. Reconciliation improves both the immediate decision and the underlying system.
Origin
Dalio began by recording the criteria behind every trade and reviewing why it worked or failed. He later converted clear principles into algorithms used across investment, people, and management decisions.
Core principles
- 01Explicit criteria expose reasoning to examination
- 02Historical tests reveal conditions intuition overlooks
- 03Algorithms process information faster and less emotionally
- 04Human-computer disagreement is diagnostic evidence
- 05Decision systems should evolve when either side detects a missing factor
How to run it
- 1
Capture the Criteria
Before or during a decision, write down the conditions and causal reasoning supporting the choice. Make the wording precise enough that another person could apply it.
Pro tip Record expectations before the outcome creates hindsight bias.
Watch out A label such as good opportunity is not a testable criterion.
- 2
Run the Postmortem
After the result becomes visible, compare it with the original expectations. Identify which assumptions held, which failed, and what information was missing.
Pro tip Evaluate the quality of the reasoning separately from whether the outcome happened to be favorable.
Watch out Lucky outcomes can preserve bad rules if results are judged superficially.
- 3
Form a Decision Rule
Turn the lesson into a conditional principle that specifies what to do when similar circumstances recur. Share it with others so they can question and refine it.
Pro tip Use explicit inputs, thresholds, and expected mechanisms where the domain permits them.
Watch out Overfitting a rule to one event makes it fragile.
- 4
Test Across Conditions
Apply the rule retrospectively to relevant historical periods, markets, or cases. Include environments outside your own lifetime and explain any failures.
Pro tip Seek adverse and unusual conditions rather than testing only friendly examples.
Watch out A backtest cannot validate inputs or regimes absent from the historical data.
- 5
Encode Repeatable Logic
Translate stable parts of the rule into equations, algorithms, or structured procedures. Let the system process current information and produce a recommendation.
Pro tip Preserve the readable principle beside the implementation so the rule remains auditable.
Watch out Automation can apply a flawed rule more efficiently.
- 6
Reconcile Human and Machine
Compare the system's output with human judgment and investigate differences. Update either the judgment or the encoded rule after finding the reason for divergence.
Pro tip Treat disagreement as a debugging opportunity rather than automatically overruling either side.
Watch out Blind obedience and casual overrides both destroy the learning value of the system.
In the wild
Dalio wrote down why he entered and exited trades, then discovered that sufficiently clear criteria could be tested across history and markets. He encoded the rules so a computer could process new information in real time. When his judgment differed from the system, he investigated whether the computer had seen more information or whether a missing factor needed to be added.
→ The process produced evolving decision systems that complemented human judgment across Bridgewater.
A company records the evidence behind each hiring decision and reviews performance after six months. Recurring predictors become a scorecard tested against prior hires. The team automates the initial comparison but investigates cases where experienced interviewers disagree with the score.
→ Hiring becomes more consistent while unusual candidates still receive thoughtful human review.
Common mistakes
Writing Rules After the Outcome
Post-outcome criteria are vulnerable to hindsight bias and may falsely describe what actually drove the original decision.
Testing Only Recent History
A rule can appear reliable merely because the test excludes older environments in which its assumptions fail.
Automating Without Reconciliation
Blindly following or routinely overriding the algorithm prevents disagreement from improving the decision system.
Is it for you?
Best for
It is best for frequent decisions with observable inputs, measurable outcomes, and enough historical or simulated cases for testing.
Not ideal for
It is not ideal for unique moral judgments or poorly measured situations where encoding false precision would conceal essential context.
From the transcript
“I've every I got in a habit of every time I would put on a trade, I would write down the criteria for putting on…”
“But I discovered that if I was clear enough, I could see how that decision rule would have worked in the past, throughout history.”
“the computer can do things that the mind can. It can process a lot more information, a lot faster, a lot more precisely, and a…”
From the episode
Ray Dalio on Failure, Meaningful Work and Relationships