Read, bluff, reveal Place the six-part tray between exactly two players, stand the lid in its keyed moon slot, and read the Caller's inward leaf tip as the forbidden route. The Caller commits one legal Flight and announces a legal route; the Watcher predicts with Reply. Reveal, score one-point Echo or two-point Mimic, rotate the Caller's leaf, and swap roles.
The last perch changes the bluff Each Caller's previous route becomes forbidden for both the actual flight and public claim, leaving exactly two legal routes. That constraint follows the role as it swaps, so the public state changes every exchange. Exact strategy analysis gives the opening Caller a 0.5 value, zero maximum duality gap, positive mixing gain in all 25 score states, and no single pure Caller action.
Echo one. Mimic two. If the Watcher's prediction matches the public claim, the reply is an Echo worth one spark. If prediction differs from claim, it is a Mimic worth two. Prediction matching the hidden actual route decides who scores. This separates confidence from correctness: a bolder reply changes the stake without changing the two legal destinations.
Six parts carry the whole state The tray carries mirrored docks and score vines; the lid closes the case and stands as the intended keyed privacy screen; two selectors record hidden Flight and Reply choices; two leaves record both score and last route. The closed digital envelope measures 98 x 64 x 12.1 mm, and the modeled storage assembly contains six valid occurrences with no clash above the stated digital tolerance.
Replayed before release Digital Playtest retained and replayed 1,600 seeded games across all 16 ordered optimizing, social, exploratory, and adversarial policy pairs. Every trace was legal and terminal within nine scoring exchanges. Exhaustive traversal covered 1,580 reachable states and 7,264 transitions. These are executable digital results, not customer play or enjoyment evidence.
Real slices, physical claims held OrcaSlicer 2.3.2 processed all six exact STLs in six separate generic Ender-3 PLA jobs using a 0.4 mm nozzle, 0.20 mm layers, three walls, 15 percent infill, and supports disabled. Hash-bound G-code and native projects report zero errors, zero out-of-bounds parts, and 0.0 g support material. No physical print or calibrated fit is proved.
Comments