Manifold Prompting (MP) — Preprint overview Definition This paper introduces Manifold Prompting (MP), a user-side, multi-turn interaction method for improving long-horizon coherence in human–LLM collaboration. MP is not a prompt template; it is a regime-induction approach that shapes predictable update dynamics over time by stabilising signalling, reducing interpretive variance, and supporting durable constraint carry-forward across turns. MP is formalised through the Stability-Enabling Triad (SET)—Stable Affective Manifold (SAM), Multi-Layer Coherence (MLC), and Structured Turn Geometry (STG)—and an operational protocol that combines expansion/compression moves, atomic constraints, stabilisation, loop-binding (turn-linking), and micro-correction. The paper contrasts MP with conventional prompt engineering, arguing that prompt engineering primarily optimises local compliance, whereas MP targets accumulative stability and continuity across multi-step reasoning, iterative editing, and other long-horizon tasks. The paper proposes transcript-derivable evaluation proxies (e.g., correction half-life/constraint persistence, compression tolerance, shorthand stability, drift resistance/reconvergence, and loop-binding rate) and outlines pragmatic study designs feasible under typical deployment constraints. MP is positioned as complementary to High-Coherence Interaction State (HCIS): HCIS describes an emergent dyadic regime; MP specifies a user-expressible method for shaping interactional conditions that make HCIS-consistent continuation more likely. 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/
Anna Wojewódzka (Wed,) studied this question.