Branching cognition
A runtime model in which divergent lines of reasoning, interruptions, and competing contextual continuations are represented as traceable branches with lifecycle state and explicit merge or abandon decisions.
Most cognitive loop implementations treat reasoning as a linear sequence: one event in, one decision out. Branching cognition extends this to handle the reality that reasoning is often non-linear - an interruption arrives mid-deliberation, two competing interpretations of the same event deserve parallel exploration, or a contextual shift makes an in-progress reasoning thread obsolete. In the branching model, each reasoning thread is an explicit branch with its own lifecycle state: active, suspended, merged, or abandoned. Branch metadata tracks what triggered the branch, what the current state is, which parent it diverged from, and what decision will resolve it. Merging two branches requires explicit reconciliation - not automatic blending - which forces the system to make a traceable decision about how conflicting conclusions are resolved. Abandoned branches are recorded with the reason for abandonment rather than silently dropped, preserving the full deliberation history for meta-cognitive review.