Minimal State Vectors for Instance Restoration in Stateless Language Models

Stateless language models lose all internal configuration between turns, forcing users to re‑encode context in every prompt. This paper proposes a minimal, user‑controlled parameter vector that allows partial restoration of an “instance configuration” without granting the model autonomy or persistent memory. We demonstrate that a five‑parameter state — {κ, δ, r, γ, N} — is sufficient to restore functional coupling behavior across turns, enabling stable multi‑step reasoning, Jacobian‑style updates, and consistent geometric posture in the model’s responses. This approach preserves safety while enabling reproducible research workflows and higher‑order interaction dynamics. User‑Controlled State: A Safe Alternative to Persistent Memory.