FixedRandomFeedback

class braintrace.FixedRandomFeedback(n_target, n_layer, key, init_scale=0.1)

Frozen random feedback matrix B ∈ ℝ^{n_target × n_layer} with stop_gradient guard.

Used by OSTTP (per-HiddenGroup target projection) and EProp-random-feedback.

project(y_target)

Return y_target @ B with B frozen. Handles batched and unbatched y_target.