Risk Simulator

Socio-Technical Risk Simulator - Governed Intelligence

Risk Simulator: Custodial Health and Authority Drift Simulator

This simulator shows how a few behaviors in human review of AI output combine into one measure of whether people are still in custody of the decisions they sign off on. It is a demonstration for explaining Authority Drift and Interpretive Debt. It does not measure a real system or use real data.

Behavioral Inputs

Move the sliders to simulate human-machine interaction metrics under evaluation.

15s
80
5%
1

Custodial Health Score

0
Loading
Analyzing behavioral variables...

Real-Time Decay Projection

The decay curve of independent oversight versus authority drift, over your last ten adjustments.

85 and above: Active Custody 50 to 84: Epistemic Decay Below 50: Custodial Collapse

Illustrative model for discussion, not a measurement of any real system.

How to use it

Move the four sliders under Behavioral Inputs and watch the Custodial Health Score change.

  • Average Review Time is how many seconds a person spends reviewing each AI output.
  • AI Output Complexity is how dense the output is. The same review time counts for less against more complex output.
  • Human Variance Rate is the percent of outputs the reviewer modified or added to, instead of accepting as written.
  • Source Lineage Tracing Depth is how many click-throughs the reviewer makes back to the sources.

Reading the score

The score runs from 0 to 100 and sets one of three states:

  • 85 and above, Active Custody: people are exercising deliberate oversight and can stand behind decisions.
  • 50 to 84, Epistemic Decay: authority drift is under way, and summaries are being accepted faster than they can be understood.
  • Below 50, Custodial Collapse: the system works as an unverified oracle and no one holds the context to defend its decisions.

Document Summary

The Master Equation. The score is a bounded sum of three components:

CHS = Min(100, Max(0, CF + HV + LD))

Weighted Component Ratios. Each component has its own discrete logic, so technical teams at OEMs and suppliers can see exactly how user interactions drive the metric:

ComponentCalculationMaximum
Cognitive Friction (CF)Min( Review Time / (Complexity × 1.8) × 40, 40 )40
Human Variance (HV)(Variance Rate / 100) × 4040
Lineage Depth (LD)(Lineage Depth / 10) × 2020

Governance Mapping. The numerical risk thresholds map to operational impact, framed for corporate auditors and pilot engineers:

ScoreStateOperational impact
85 and aboveActive CustodyDeliberate oversight; decisions can be defended.
50 to 84Epistemic DecayAuthority drift under way; summaries accepted faster than they are understood.
Below 50Custodial CollapseUnverified oracle; no one holds the context to defend decisions.

The decay graph

The Real-Time Decay Projection plots the last ten scores as you move the sliders, so you can show a score falling or recovering as behavior changes. It starts again when the page reloads.

Limits

The weights behind the score are illustrative. Review time against expected review time (1.8 seconds per complexity point) counts for up to 40 points, variance for up to 40 and lineage depth for up to 20. They show the shape of the idea and are not calibrated or validated.

Running the Live Simulation

The web-based prototype is fully functional. It models the decay curve of human oversight dynamically based on user behavior parameters.