Ambient Structure Discovery in Complex Adaptive Systems

Executive Summary

Ambient Structure Discovery (ASD) is a method for detecting organizational pathology that agents cannot directly observe. The headline contribution is normalized deviance detection: the ability to identify systematic deviations from expected coordination patterns before those deviations produce visible failures. Organizations do not fail suddenly; they drift into failure through accumulated normalized deviance. ASD detects the drift.

The paper presents a controlled evaluation across 22,500 Monte Carlo simulation runs (3 anomaly types, 5 severity levels, 3 mesh sizes, 5 noise levels, 100 runs per combination) and a 30-day field deployment ingesting 3,520 real-world organizational signals. The results establish ASD as a practical detection system with quantified sensitivity, specificity, and robustness characteristics.

A provisional patent has been filed: USPTO Provisional #63/981,369 (February 2026).

The Detection Problem

Complex adaptive systems cannot directly observe their environment's structure. They discover it indirectly through the consequences of their actions. The ambient structure discovery process operates through three stages: action (the agent acts in the environment), consequence observation (the agent observes the consequences), and model update (the agent updates its implicit model of environmental structure).

The fundamental problem is that this process converges on an accurate model only under conditions that rarely obtain in organizations: stationary environments, unconstrained action spaces, and rapid feedback. Organizations systematically violate all three conditions. ASD addresses this by using the pattern of signal interactions -- not the signals themselves -- to detect structural anomalies that individual agents cannot perceive.

Controlled Evaluation: 22,500 Monte Carlo Runs

The Monte Carlo evaluation tests ASD against three anomaly types that represent distinct organizational pathologies:

• Coordination failures: Breakdowns in expected interaction patterns between organizational units. ASD achieves a true positive rate (TPR) of 0.82 at a false positive rate (FPR) below 0.15. This is the strongest detection performance across all anomaly types, consistent with ASD's design as a structural interaction detector.
• Knowledge silos: Isolation of information within organizational boundaries. ASD detects silos reliably but reveals a severity inversion: moderate silos are detected more reliably than severe silos. This is theoretically consistent -- severe silos produce so little cross-boundary interaction that there is insufficient signal for the detection mesh to operate on. The absence of signal is itself informative but requires a different detection strategy.
• Normalized deviance: Systematic drift from expected operational patterns. Detection performance is intermediate between coordination failures and knowledge silos, with TPR varying from 0.61 to 0.78 depending on severity and noise level.

A critical robustness finding: detection performance is robust across mesh sizes, with performance spread less than 0.07 across the three mesh configurations tested (fine, medium, coarse). This means that the detection is not an artifact of mesh resolution but reflects genuine structural properties of the signal interaction patterns.

Sensitivity Analysis

The 5x5 severity-noise grid reveals the sensitivity structure of the detection system:

• Severity effect: Detection improves monotonically with anomaly severity for coordination failures and normalized deviance, but inverts for knowledge silos (as noted above)
• Noise effect: Detection degrades gracefully under noise. At the highest noise level tested, coordination failure TPR remains above 0.65, indicating that the structural signal survives substantial noise contamination
• Interaction effect: The severity-noise interaction is subadditive for coordination failures (noise matters less at high severity) but superadditive for knowledge silos (noise amplifies the severity inversion)

Field Deployment: 30-Day Validation

The field deployment ingested 3,520 organizational signals over 30 days from production systems (incident reports, deployment logs, code review activity, communication patterns). The deployment yielded 23 actionable findings:

• Coordination failures detected: 9 instances of cross-team coordination breakdown, 7 of which were confirmed by independent assessment
• Knowledge silos detected: 6 instances of information isolation, 5 confirmed
• Normalized deviance detected: 8 instances of operational drift, 6 confirmed
• False positives: 5 flagged patterns that did not correspond to actual organizational pathology upon investigation

The field deployment false positive rate of 5/23 (21.7%) is higher than the Monte Carlo estimate but consistent with the additional complexity of real-world signals compared to simulated data. Importantly, the false positives were identifiable as such upon investigation (they did not require deep analysis to dismiss), suggesting that the practical cost of false positives is manageable.

Organizational Failure Modes

ASD's detection framework identifies four characteristic failure modes in organizational ambient structure discovery:

• Non-stationary environments: Organizations that developed accurate models in one competitive environment continue to act on those models after the environment has changed
• Constrained action space: The cage limits the actions available to the organization, preventing discovery of the parts of the environment that constrained actions cannot reach
• Delayed consequences: Long product development cycles and slow competitive dynamics mean consequences arrive too late to update models before they are entrenched
• Attributed consequences: Organizations attribute consequences to agents rather than structures, preventing accurate model updates about environmental structure

Patent Filing

The ASD detection method is the subject of USPTO Provisional Patent Application #63/981,369, filed February 2026. The patent covers the signal interaction mesh architecture, the normalized deviance detection algorithm, and the multi-resolution analysis approach that produces mesh-size-robust detection. The provisional application establishes priority while the full application is prepared.

Key References