SNN-Synthesis v1: Task Specificity, Model Scaling, and Temporal Dynamics of Latent Trajectory Distillation in LLM Reasoning

This preprint presents SNN-Synthesis v1, a systematic investigation of latent trajectory distillation, multi-layer cognitive orchestration, and cross-task transfer of reasoning representations in large language models. Building on the SNN-Genesis framework, seven controlled experiments across two model scales (Qwen2.5-0.5B and Mistral-7B) establish three principal findings: (1) L18 trajectory injection achieves 41–47% solve rate on Modified Hanoi (p < 10⁻⁶, N=100, Fisher exact test), while dual-layer simultaneous injection catastrophically collapses to 0%; (2) reasoning representations are strictly task-specific, with cross-task injection causing active performance degradation; (3) a critical model-scale threshold exists between 0.5B and 7B parameters for trajectory-based interventions. Code and data: https://github.com/hafufu-stack/SNN-Synthesis AcknowledgmentsThis research was conducted entirely independently, without institutional affiliation or corporate funding. The author currently faces financial constraints that make it increasingly difficult to maintain subscriptions to AI services essential for this line of research. To sustain and improve the quality of future work, the author is actively seeking community sponsorship. Details are available at https://github.com/sponsors/hafufu-stack