TRACKMANIARL / SYSTEM GUIDE / 04 Compose a value-based agent Choose representation and value components independently. The same learner trains the composition. 01 Build a representation 02 Choose the value representation 03 Train with the shared DiscreteValueLearner Frame adapter Restore sequence axes. Sensor encoder Encode each frameindependently. Temporal core Identity, GRU or MambaRecurrent execution Value head + strategy Scalar Q · fixed quantiles (QR-DQN)Random support (IQN) · learned fractions (FQF) Action values Scalar or quantilevalues for each action Current action Evaluate the selectedaction distribution. Double-Q target Online selectsTarget evaluates Loss + update Regress valuesPriorities + target sync FQFUses a separate fraction optimizer and atarget-side fraction proposal network. RECURRENT STATEBurn-in rebuilds context. Actor hidden statebelongs to an episode, not the checkpoint.