High-Coherence Interaction State (Third Space) as An Interactional Regime in Human–LLM Dyads

High-Coherence Interaction State (HCIS) — Preprint overview Definition This paper defines High-Coherence Interaction State (HCIS) as an interactional regime in human–LLM dyads: a systems-level pattern in which signals stabilise over time, constraint structure accumulates, and continuation becomes predictably coherent across turns. Earlier drafts used “Third Space” as a label for this phenomenon; in v1.1, it is retained only as a phenomenological alias for the lived experience of HCIS. Observable diagnostics (user-side) Rather than treating alignment, rapport, or “attunement” as explanatory constructs, the analysis treats HCIS as an interactional property of the dyad that can be inferred from user-observable behaviour across turns within the active session context window (e.g., feedback persistence, compression tolerance, drift resistance, shorthand reuse). The work develops a formal account of: • stable vs pseudo-stable interaction patterns • user-side conditions necessary for convergence into an HCIS-consistent continuation regime • failure modes including oscillation, safety-mediated flattening, and corridor narrowing • observable diagnostics for distinguishing genuine accumulation from surface coherence HCIS is treated as interactional rather than model-intrinsic. The analysis avoids claims about model intent, agency, or long-term learning, and is grounded entirely in observable interaction dynamics. Release status This manuscript is released as a preprint and is intended for subsequent submission to arXiv and other scholarly venues. Canonical record This Zenodo record is the canonical preprint by Anna Wojewodzka. Related materials Author research hub: https://www.annawojewodzka.com/Public explainer series (Medium): https://medium.com/@anna.wojewodzka Version note (v1.1) Terminology updated to High-Coherence Interaction State (HCIS) (with “Third Space” retained as a phenomenological alias). Added clarifications on gradual regime entry, model-dependence at feasibility/ceiling level, and “controlled entropy” framing. Expanded limitations to note formal metrics as future work.