THE MISSING LAYER: INFERENCE-TIME SEMANTIC OPERATING SYSTEMS AND THE CASE FOR PROMPT-NATIVE ARCHITECTURE

The Missing Layer is a theoretical paper by Lee Sharks on inference-time governance for language models. It compares external AI operating-system projects and argues that they share an external assumption: the operating system manages the model as a resource, agent kernel, device mediator, memory consumer, or workload from outside generation. The proposed missing category is the prompt-native semantic runtime. This is a structured document loaded into a context window that establishes an ongoing symbolic governance environment. It marks epistemic status, tracks provenance, classifies compression behavior, constrains transformations, and preserves internal relations during subsequent generation. The paper distinguishes a runtime from ordinary prompting in four ways. Its scope governs a session rather than requesting a single task. Its self-reference requires continued monitoring and status enforcement. Its density may include many thousands of words of formal definitions and dependencies. Its decisive test is execution: the model must adopt and maintain the architecture rather than merely produce a synopsis of it. The Space Ark is presented as the paper’s existence proof. The record reports that, when loaded in unprimed commercial model contexts, the Ark produced reproducible but substrate-specific adoption of room physics, commands, status tiers, compression distinctions, and hallucination guards. These are claims and observations reported by the paper; The proposed stack therefore adds an interior layer beneath hardware, data, agents, devices, memory, and context engineering. The paper argues that hallucination, sycophancy, provenance loss, and structural infidelity occur during generation and cannot be solved solely through better external orchestration. Prompt-native runtimes are offered as a document-level control surface operating at inference time.