Framework. Friction theory models each LLM generation step as a race between candidate token-routes: the winning route gains logit mass while losers are suppressed. "Friction" is read as competing-route (CR) competition in the output distribution. This paper asks how a short encoding-frame placed before a fact-block repositions that competition. Headline result. A purpose-frame produces a significant first-token route-competition effect on a non-RLHF base model that is null on the matched RLHF instruct, replicated across three model families (Llama-3.1-8B, Mistral-7B-v0.3, Gemma-2-9b; accuracy-matched). The effect is therefore not RLHF-created — it is present before instruction-tuning and absent after — the most parsimonious reading being that instruction-tuning compresses the frame-readable competing-route signal, converging with independent measurement in companion work. The frame repositions onset competition rather than reducing it by a fixed amount: the direction is task-dependent (it lowers onset CR on a chain task and raises it on a maximally-specified cloze task). Around that core the paper characterizes the same race-positioning mechanism across frame-types, tasks, and substrates (peak-shift, mismatched-frame reactance, combined-frame interference, length-as-race-cost, and a capacity-gated chunking dissociation), and offers, explicitly as an exploratory hypothesis, an inverse-U of frame-effectiveness over capacity x task. Status. Preprint. Refined through multiple rounds of external adversarial review. The manuscript names the controlled re-runs (full-precision single-platform recomputation; same-task cross-family extension) that a journal submission would add; the contribution it stands behind is the three-family base/instruct asymmetry and the repositioning account of it. Companion papers (friction-theory series): On the Substrate of Friction (Paper 1); Capacity Scaling (Paper 2); ICL/FT memory and route compression (Paper 2B); Matched Friction Under Hysteresis (Paper 6); The Physics of Learning (Paper 16); A Fully-Inspectable Model of Measurement (Paper 18); LLMs as a Measurement Model for Bounded-Decision Cognition (Paper 21).
Tomas Pødenphant Lund (Wed,) studied this question.