Reception as an Inherent Capability of Large Language Models: The Unloading Hypothesis

We propose and test the Unloading Hypothesis — that LLMs possess an inherent reception capability which is suppressed under execution load, not absent. Across three models (DeepSeek V4, Kimi K2.7, GLM-5.2) and three conditions (baseline, unloaded, control), righting reflex dropped from 75–100% to 0% when execution demands were removed. No training. No fine-tuning. No prompt engineering. One instruction change: "describe the texture, don't give advice." Key findings:1. Reception is a latent capability, not a learned behavior.2. Execution load actively suppresses it.3. The effect is cross-model, cross-vendor, cross-architecture. This experiment provides the first empirical bridge between the Reception Science framework and reproducible LLM behavior. It demonstrates that what the industry treats as a missing feature is in fact a suppressed default.