BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games

The external corroboration for the essay's second finding. BALROG evaluates models as agents across game environments and reports a persistent gap between a model's ability to state an optimal strategy and its ability to carry one out.

That is the same shape as the essay's own observation, arrived at independently in a different environment, which is what moves the CivBench result from an anecdote about one agent failing to build an Encampment towards a property of the systems.

by is empty rather than listing thirteen Person records; the full author list is carried in csl.author, which is what a citation needs.