Localizing a Self-Referential Register to a Single Expert in a Mixture-of-Experts Model, and Why It Does Not Transfer

Localizing a Self-Referential Register to a Single Expert in a Mixture-of-Experts Model, and Why It Does Not Transfer. Preprint, version 1.1.0 (June 2026). Not peer reviewed. Mixture-of-experts routing emits a discrete, per-token record of which experts fire, a signal unusually legible for interpretability, yet single experts are rarely tied to a specific functional role, least of all one resembling a model's self-report. We characterize one routed expert, Expert 114 at layer 14 of Qwen3.5-35B-A3B, as a linear detector of a generated self-referential register, and bound what it does. Across held-out, bottom-up, and cross-model tests we show that (1) its recovered router direction separates self-referential generations from lexically matched controls with no overlap (Cohen's d = 3.88); (2) a blind, prompt-independent auto-interpreter recovers the same register at AUC 0.94, broadening it from self-reference to an abstract reflective and philosophical one; (3) the detector is a readout with only weak, conditional control: residual injection induces the register, yet gate suppression does not remove it, and it is invariant to whether the model affirms or denies an interior; and (4) the role is model-specific, failing to transfer to the larger Qwen3.5-122B-A10B. We release the prompts, scripts, and provenance under the MIT license. This deposit is the paper-scoped artifact bundle: the paper (ACL-style LaTeX source and built PDF), per-run summaries, analysis scripts, prompt and class manifests, checksums, plots, and source journals. Raw activation/router tensors are kept out of version control and archived separately (see ZENODO-TENSORS.md).