Two Clocks
What happens to a decision while you are waiting to know more?
One project · two changing clocks
A promising project. A real deadline.
You have 12 weeks and $100,000. Building costs $60,000 and takes 8 weeks. The idea has a 50% estimated chance of being viable. Each test costs $5,000 and one week; its signal may be wrong. What would you do?
The same weeks, two clocks
Build can start through week 4Choose an action
What does the model expect from here?
These compare additional value from this moment. Earlier test costs are sunk; a realized outcome is a different quantity.
Replay this project keeps its hidden viability, test draws, and execution draw. You can change your decisions without changing that underlying future.
Waiting moves the decision too.
Evidence may change what you should do, but it cannot return spent time or budget. An unfavorable outcome does not, by itself, make the earlier choice unjustified.
The useful question is whether another test could change your action while that action is still available.
Read the Two Clocks essay →Explore other futures and strategies
These optional comparisons help separate a decision's expected value from one realized outcome.
The same decision, different outcomes
Build now from the starting conditions. Sample independent projects; the choice and 50% estimated chance of execution success stay the same.
Each mark is a random result, not a verdict on the decision.
Compare strategies on matched projects
Build now; test once and build after a positive signal; or choose the highest expected-value action after each result. Each strategy faces the same hidden projects and random draws.
The adaptive policy is optimal only under this synthetic model's assumptions. A batch of 100 can favor another policy by chance.
Adjust assumptions for the next replay
Changing an assumption starts a fresh decision on this same hidden project. Comparisons across different assumptions are not like-for-like.
How this illustrative model works
There are 12 weeks and $100,000. Building costs $60,000, takes 8 weeks, and returns $180,000 if execution succeeds. A preliminary test costs $5,000 and one week. The starting chance that the underlying idea is viable is 50%.
A clearer test gives a supportive signal 80% of the time when the idea is viable, and 20% when it is not. Results update the viability estimate using Bayes’ rule; they do not change the hidden project. Execution succeeds 90% of the time for a viable idea and 10% otherwise. Optional settings change these assumptions for a replay.
The result of each test is randomly sampled from the hidden project. The same-project replay preserves that project and its draws. Expected values count future choices; realized net subtracts all incurred test and build costs. All amounts and probabilities are invented for this explanation, not forecasts or recommendations.