Research paper 11

Telemetry Without Authority: Roughness and Lacunarity as Scout Signals for Agent Context

Agentic development systems need early warnings when context is fragmented, assumptions are missing, or work is likely to drift. Telemetry can help, but telemetry becomes dangerous when it is treated as authority. This paper studies roughness and lacunarity as scout signals for agent context. Roughness marks local mismatch or inconsistency in a context surface. Lacunarity marks gaps, missing connective tissue, or unevenly distributed evidence. Local historical Fractal Governance artifacts provide pilot evidence. A refreshed roughness-lacunarity scout batch contained 1000 receipts: 500 roughness receipts and 500 lacunarity receipts. It recorded 454 roughness mismatch receipts and 250 lacunarity gap receipts, with bootstrap mismatch and gap rates of 0.908 and 0.5 respectively. A controlled lacunarity task pilot also contained 1000 cases: 500 low-gap contexts with 500 proof passes and no wrong assumptions, retries, or escalations; and 500 high-gap contexts with mean pre-action gap score 0.6675, 1223 wrong assumptions, 1223 retries, 462 escalations, and no proof passes. These results motivate roughness and lacunarity as context-risk telemetry. The claim is deliberately bounded: telemetry may prioritize review, trigger questions, or route work, but it may not approve, deny, promote, or replace proof.