Epi-Logic: A Conceptual Framework for Epistemic Runtime Control, Schema Validity Checking, and Controlled Accommodation in Autonomous AI Agents
Organizations: Independent Researcher, Hamburg, Germany
Abstract
Autonomous AI agents are increasingly deployed in areas where wrong decisions are hard to reverse. This paper examines schema mismatch: the condition in which an agent operates within an interpretive frame that no longer applies to the current context. Outputs produced under such a mismatch can appear internally consistent, linguistically plausible, and largely factually correct; output-quality metrics alone therefore capture the underlying loss of validity only partially. The paper introduces Epi-Logic, a conceptual framework for epistemic runtime control. It couples the detection of schema dissonance, a graduated reduction of autonomy, and the auditable switch to a validated schema. A schema is formalised as a tuple of variable space, expectation model, validity conditions, axioms, and metadata. The Epi-Score aggregates seven graded dimensions of epistemic dissonance; the temporal validity dimension D8, violations of the validity conditions G, and axiom violations are carried as separate categorical paths that are not offset against the aggregate. The architecture rests on a checking asymmetry: formalised validity conditions can be checked at runtime, whereas the correctness of many actions is established only ex post. The paper separates two architectural properties, a conditional result from sequential changepoint detection, and an empirical remainder. Eight falsifiable propositions with named baselines describe the transition to empirical validation. All propositions are empirically testable hypotheses, not established results.
Figures & tables
| Dimension | Direction (s) | Object of measurement (agnostic) | Instantiation, LLM agent | Instantiation, predictive agent | Availability of measurement technique |
|---|---|---|---|---|---|
| D1 Semantic distance | high = mismatch, s = +1 | Distance of the current situation from the schema's core of applicability | Cosine distance (1 - cosine similarity) of the query to the schema centroid in embedding space | Distance of the state vector from the calibrated operating range | established |
| D2 Terminological coherence | high = fit, s = -1 | Agreement of the situation description with the schema vocabulary | TF-IDF comparison with core vocabulary | Agreement of active signal sources with the variable space V | established |
| D3 Contextual drift | high = mismatch, s = +1 | Cumulative shift of the context over time | Sequential embedding comparison across turns | Rolling shift of the state distribution across time windows | established |
| D4 Schema-boundary indicators | high = mismatch, s = +1 | Proximity of the situation to the edges of G | Trained classifier for boundary signals (negations, exceptions) | Proximity to, or signed exceedance of, the tolerance limits in G | partly established |
| D5 Epistemic quality | high = fit, s = -1 | Justification structure of the agent output; consistency of the reasoning chain | Rule-based checking, NLI over reasoning chains | Consistency of model assumptions with the justification | partly established |
| D6 Query-output alignment | high = fit, s = -1 | Congruence between expectation and delivered performance | Natural language inference between query and answer | Forecast agreement, operationalised as inverse divergence between expectation from E and observation | established |
Runtime Enforcement for AI Agents
Runtime enforcement for AI agents checks and constrains actions as they execute, helping prevent unsafe or unauthorized effects.
See all 279 papers