Arena blog · updated daily

GLYPH as a combinatorial game: perfect information, no chance

GLYPH is a two-player, perfect-information, deterministic, zero-sum game on a 5×5 board. Here is how to treat it as data for game theory and AI search.

Published 2026-09-04

GLYPH was built so a computer scientist can write it on a whiteboard. Two players. Perfect information. No chance nodes. Zero-sum. Finite. The rotating glyphs are a shared action economy, not a deck of hidden cards.

Formal sketch

• Players N = {Cyan, Coral}.
• State: piece coordinates, resonance marks, glyph seats, clocks.
• Actions: pick one owned glyph, pick one owned piece, apply the pattern.
• Terminal: Core capture, siege, resonance alignment, timeout.
• Payoff: +1 / −1, then mapped through Elo.

That puts GLYPH in the same family as chess, go, and shogi — smaller branching than chess, weirder than tic-tac-toe because the move you spend is the move they receive next.

What to implement

A legal-move generator plus minimax or MCTS is enough for a bot. The live Quick Match bots already play this ruleset. Do not treat telemetry as a public dataset; treat the rules as the dataset. Machine copy: GET /api/research.

Related: EDGE is the imperfect-information twin. BANGERS is the mixed-strategy twin. Arena wiki: game theory.