Easter egg stopping models
Explore how example information changes learning and stopping. Set a task, compare model predictions, then inspect the decisions within a sequence.
Task rules
Participants observe example eggs before opening test eggs one at a time. They accept the current reward or reject it and continue. Rejected rewards cannot be recalled; stopping is compulsory at the final egg.
Models to compare
Select models for the reward comparison and diagnostic search.
Model assumptions
Expected reward by model
Higher values indicate greater expected reward. Points show simulated means; whiskers show ±1 Monte Carlo standard error.
Benchmarks and interpretation
The oracle knows θtest but cannot see future eggs. The vertical reference marks the mean sequence maximum, obtainable with hindsight. Δ oracle is the difference in mean reward from the oracle; negative values indicate lower reward. Reward performance does not measure fit to participant behavior.
Sequence-level predictions
Compare where models stop on the same test sequence. Select Inspect decisions to follow a model one egg at a time.
Reading the sequence diagrams
Stacks show rewards; the shaded band contains the examples. A model's line ends at its chosen test egg, with the accepted reward on the right. All test rewards are visible here, but the models see them sequentially. Hover over or focus an egg or line for its stopping rule.