Tinybrains History Audit and Formation Extension

This report asks a narrow mechanistic question: what kind of history is already present in the public tinybrains model, and what changes when a persistent relation-forming layer is added above its connectome-derived neural dynamics? The result is positive but bounded. Native tinybrains behavior is path dependent through explicit transient state, yet the difference disappears when all explicit state is matched. A separately specified formation matrix produces prospective history effects after neural and acute state reset. Nonlinear transformations gated by the current input preserve cue specificity substantially better than linear history mixing. Multiplication is the implementation of that gate, not the primary classification axis. A locked discrete closed-loop experiment then separated action- contingent world return from matched exposure. A continuous two-agent experiment retained both brains, bodies, positions, and partner-generated acceptance or rejection throughout development. Live closure produced prospective differentiation after acute reset and survived live-yoked, timing, mirror, and P1-silencing controls. Exact replay of the complete locally available trajectory reproduced the effect, however, exposing an operational equivalence boundary rather than evidence for external causal provenance. A final localization-compatibility test compressed formation into a literature- constrained mushroom-body-like compartment code. Gated formation, transfer, correction, and readout dependence survived, but a capacity-matched scrambled basis performed at least as well and DAN, octopamine, and dual gating were indistinguishable. The experiments therefore support biological architectural compatibility, not anatomical localization or a unique neuromodulatory mechanism. They do not establish phenomenal consciousness or biological validation.