Proceedings of the Annual Meeting of the Cognitive Science Society

We model shogi move selection in ACT-R to test how expertise depends on representation and cognitive resource constraints. The model integrates visual exploration, a capacity-limited imaginal stack, and declarative memory retrieval. Beyond absolute board encoding, we implement king-centered relative representations (3$\times$3 local patterns) intended for endgame reasoning. Simulations crossed (i) relative representations on/off and (ii) imaginal capacity (9 vs. 18), while expertise was manipulated by storing expert vs. novice game records. Expert advantage was conditional: it was stable when relative representations were available, but collapsed (and sometimes reversed) when relative representations were absent under low capacity, accompanied by a sharp drop in retrieval use and earlier convergence of game states. Larger capacity partially compensated for missing relative representations. These results imply that expertise in this domain emerges from representation $\times$ resource coupling, not experience alone.