Runtime Research for Autonomous AI Systems

Technical papers on deterministic replay, trajectory drift, execution validity, and runtime-state evolution in autonomous AI systems.

Research Papers

Trajectory Drift and Execution Validity in Multi-Step LLM Workflows

Deterministic replay and analysis of structural execution behavior across drift, branching, convergence, and multi-step workflow progression.

TRAJECTORY RUNTIME CONTROL
Available on:Zenodo
Source:GitHub

Efficiency Collapse in Multi-Step LLM Execution

Analysis of execution trajectories across sequential steps showing efficiency is highest at the initial step and declines sharply thereafter.

EXECUTION SYSTEMS X-Ray
Available on:
SSRN|Zenodo
Source:GitHub

Governance Maturity in Autonomous AI Agent Systems

Evaluation of governance primitives across AI agent systems using the Autonomy Accountability Framework (AAF)

AAI
Available on:
SSRN|Zenodo
Source:GitHub

Accountability Framework

ACCOUNTABILITY FRAMEWORK

Autonomy Accountability Framework (AAF)

A governance model for understanding and managing accountability in autonomous AI agent systems. Introduces core constructs including the Agent Accountability Stack (AAS) and Governance Debt (GD).

AASGovernance Debt
Available on:
Zenodo|SSRN
Source:GitHub

Runtime Research Areas

RUNTIME RESEARCH AREAS
01

Execution-State Analysis

02

Continuation Control

03

Runtime Constraint Enforcement

04

Execution Accountability Systems