Executive Summary
Context fences — reset instructions that clear prior context before new coordination — emerged in Papers XIX and XX as the single most effective coordination mechanism. But why do they work? The intuitive explanation is garbage collection: the reset removes accumulated noise from the receiver's context, freeing capacity for new information. This paper tests that hypothesis directly against an alternative: mode switching, where the reset triggers a transition in the model's processing mode rather than clearing accumulated state.
The garbage collector hypothesis is falsified on three counts. First, distance from neutral state does not predict cross-entropy (Spearman rho 0.100) — if garbage collection were the mechanism, contexts farther from neutral should benefit more from reset. Second, more tokens in the context correlate with better results (rho -0.672), the opposite of what garbage collection predicts. Third, reset benefit is constant at approximately 0.076 nats regardless of context length — garbage collection would predict increasing benefit with longer contexts.
The correct mechanism is mode switching. A standard 18-token reset achieves CE 0.503, while silence (no coordination at all) yields CE 0.699. The mode and domain components are separable (Spearman rho 0.858), meaning the reset triggers a mode transition independently of the subsequent domain priming. This two-phase architecture — mode switch followed by domain activation — is the mechanistic foundation for the optimal coordination protocol identified across the entire experimental series.
Key Findings
- Garbage collection falsified (1): Distance to neutral does not predict CE (rho 0.100)
- Garbage collection falsified (2): More tokens = better results (rho -0.672), opposite of garbage collection prediction
- Garbage collection falsified (3): Reset benefit constant ~0.076 nats regardless of context length
- Mode switching confirmed: Reset triggers processing mode transition, not state clearing
- Standard reset performance: 18-token reset = CE 0.503; silence = CE 0.699
- Mode and domain separable: Two-phase composition with rho 0.858 independence
Key References
- McEntire (2026) — Ritual Shape: reset + prime as optimal protocol (Paper XIX)
- McEntire (2026) — Shepherd Agents: reset dominates shepherd content 22:1 (Paper XX)
- McEntire (2026) — The Coordination Problem Is Interference: unified framework (Paper XXI)
- McEntire (2026) — Hop Scaling: chain degradation across multi-agent hops (Paper XXIII)