Imperial College's SAD algorithm recovers censored poker data, boosting exploit value to 0.815

In poker-like games, opponents who fold hide their private cards, creating a 'showdown censoring' problem: the observed data is missing not at random, so standard estimators converge to a selected distribution and lose statistical coverage over time. Jiaxing Guo at Imperial College London addresses this in a new arXiv paper by proposing a 'floor-safe probe' that deliberately drives a chosen line to showdown, revealing every non-fold continuation and using sequence-form flow to recover the censored fold mass on reveal-certified histories.

The paper introduces Safe Active De-censoring (SAD), which combines this safe observation capacity with public-anomaly routing and robust deployment. The theory provides a safe observation capacity frontier that is concave and piecewise linear, with the origin slope equal to the floor's shadow price. In a fixed-board unbucketed river subgame, a constructed 'public twin' admits a floor-safe response of value V=0.815; the audited public channel certifies only the blueprint floor, while population reveal evidence certifies at least 96% of V. Across a broader synthetic opponent population, public and solved grouped reveal fibers certify median shares of 73% and 91% of the safe-exploitable gap.

This work separates unconditional safety (the floor's guarantee) from the conditional statistical value of active reveal, showing when it's worth paying the cost of driving to showdown. SAD's selective reveal strategy reaches a target with cost N = Θ̃(1/(κρ(I)πε²)) for local censored-fiber directions. While routing across targets remains heuristic, the evidence spans bucketed turn-river endgames and controlled instances, demonstrating a practical path to robust opponent exploitation in imperfect-information games.