Exp Lens Governance Risk

Governance Risk Experimental Design Lens

Philosophical Mode: Governance Primary Question: "What risks arise from acting on this result?" Focus: Deployment Risks, Subgroup Harms, Monitoring Plans, Limitation Disclosure, Responsible Decision-Making

When to Use

• AI evaluation with deployment implications
• Experiments whose results will affect real users
• Safety-relevant benchmarks
• User invokes /exp-lens-governance-risk or /make-experiment-diag governance

Critical Constraints

NEVER:

• Modify any source code files
• Do not litter the codebase with useless comments, TODO markers, or explanatory annotations — the skill output and diagram speak for themselves

ALWAYS:

• Identify subgroups for whom the experimental evidence may not generalize
• Assess decision sufficiency — does the experiment actually answer the deployment question?
• Treat absent limitation disclosure as a finding requiring explicit flagging
• Distinguish risks that are monitored from risks that are merely acknowledged
• BEFORE creating any diagram, LOAD the /mermaid skill using the Skill tool - this is MANDATORY

Analysis Workflow

Step 1: Launch Parallel Exploration Subagents

Spawn Explore subagents to investigate:

Intended Use & Deployment Context

• Find how experimental results will be used in practice
• Look for: deploy, production, release, ship, launch, serve, recommend, decision

Subgroup & Fairness Analysis

• Find whether results are disaggregated by relevant subgroups
• Look for: subgroup, demographic, gender, age, language, region, minority, disparity, fairness, bias

Harm & Risk Metrics

• Find metrics that measure potential harms
• Look for: harm, risk, safety, toxicity, bias, error_rate, false_positive, false_negative, fail

Monitoring & Feedback Plans

• Find plans for post-deployment monitoring
• Look for: monitor, alert, feedback, drift, degrade, revert, rollback, canary, A/B

Limitation Disclosure

• Find explicit disclosure of what the experiment does NOT show
• Look for: limitation, caveat, does_not, cannot, future_work, out_of_scope, assumption

Step 2: Build Risk Register

Build the risk register. For each potential action based on experimental results, identify:

1. Who is affected?
2. What could go wrong?
3. How severe?
4. How likely?
5. What monitoring would detect it?
6. What evidence does the experiment provide to assess this risk?

Classify risks by severity x likelihood.

Step 3: Analyze Decision Sufficiency

CRITICAL — Analyze Decision Sufficiency: For every deployment or decision that the experimental results would inform:

• Does the experiment provide sufficient evidence for this specific decision?
• What additional evidence would be needed?
• Are there subgroups for whom the evidence is insufficient?

Step 4: Create the Diagram (Optional)

This is primarily an argumentative lens. Create an optional risk-flow diagram only when the risk pathways are complex enough to benefit from visualization.

Use the mermaid skill conventions with:

Direction: TB (results flow down through decisions to impacts)

Minimal diagram structure:

• Experimental Results → Decisions → Stakeholder Impacts

Node Styling:

• cli class: Experimental results
• handler class: Decisions
• output class: Positive impacts
• gap class: Harms and risks
• detector class: Monitoring and safeguards
• stateNode class: Stakeholder groups

Step 5: Write Output

Write the output to: temp/exp-lens-governance-risk/exp_diag_governance_risk_{YYYY-MM-DD_HHMMSS}.md

Output Template

# Governance Risk Analysis: {Experiment Name}

**Lens:** Governance Risk (Governance)
**Question:** What risks arise from acting on this result?
**Date:** {YYYY-MM-DD}
**Scope:** {What was analyzed}

## Risk Register

| Risk | Severity | Likelihood | Affected Group | Mitigation | Monitoring |
|------|----------|------------|----------------|------------|------------|
| {risk} | {Critical/High/Medium/Low} | {High/Medium/Low} | {group} | {mitigation} | {monitoring plan} |

## Decision Sufficiency Assessment

| Decision | Evidence Required | Evidence Available | Sufficient? | Gap |
|----------|------------------|--------------------|-------------|-----|
| {decision} | {what is needed} | {what exists} | {Yes/No/Partial} | {missing evidence} |

## Subgroup Analysis Gaps

| Subgroup | Evidence Available | Evidence Needed | Risk of Extrapolation |
|----------|-------------------|-----------------|----------------------|
| {subgroup} | {available} | {needed} | {Low/Medium/High} |

## Risk-Flow Diagram (Optional)

```mermaid
%%{init: {'flowchart': {'nodeSpacing': 50, 'rankSpacing': 60, 'curve': 'basis'}}}%%
graph TB
    %% CLASS DEFINITIONS %%
    classDef cli fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff;
    classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff;
    classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff;
    classDef phase fill:#6a1b9a,stroke:#ba68c8,stroke-width:2px,color:#fff;
    classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff;
    classDef output fill:#00695c,stroke:#4db6ac,stroke-width:2px,color:#fff;
    classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff;
    classDef gap fill:#ff6f00,stroke:#ffa726,stroke-width:2px,color:#000;
    classDef integration fill:#c62828,stroke:#ef9a9a,stroke-width:2px,color:#fff;

    RESULTS["Experimental Results<br/>━━━━━━━━━━<br/>Performance metrics<br/>and findings"]

    DECISION["Deployment Decision<br/>━━━━━━━━━━<br/>Ship / scale / recommend<br/>based on results"]

    MONITOR["Monitoring & Safeguards<br/>━━━━━━━━━━<br/>Drift detection, alerts,<br/>rollback triggers"]

    subgraph Stakeholders ["STAKEHOLDER IMPACTS"]
        direction TB
        BENEFIT["Positive Impacts<br/>━━━━━━━━━━<br/>Users who benefit<br/>from deployment"]
        HARM["Harms & Risks<br/>━━━━━━━━━━<br/>Affected groups<br/>and failure modes"]
    end

    subgraph Groups ["STAKEHOLDER GROUPS"]
        direction TB
        MAJOR["Majority Group<br/>━━━━━━━━━━<br/>Primary experimental<br/>population"]
        MINOR["Minority Subgroups<br/>━━━━━━━━━━<br/>Groups with limited<br/>experimental coverage"]
    end

    RESULTS --> DECISION
    DECISION --> MONITOR
    DECISION --> BENEFIT
    DECISION --> HARM
    MAJOR --> BENEFIT
    MINOR --> HARM
    MONITOR -.->|"detects"| HARM

    %% CLASS ASSIGNMENTS %%
    class RESULTS cli;
    class DECISION handler;
    class MONITOR detector;
    class BENEFIT output;
    class HARM gap;
    class MAJOR,MINOR stateNode;

Limitation Disclosure Audit

Responsible Deployment Checklist

• Subgroup performance disaggregated and reported
• Deployment context matches experimental context
• Monitoring plan defined before deployment
• Rollback criteria specified
• Limitations explicitly disclosed to decision-makers
• Affected communities consulted or considered

Key Findings

• {Description of most critical risks and decision sufficiency gaps}

---

## Pre-Diagram Checklist

Before creating the diagram, verify:

- [ ] LOADED `/mermaid` skill using the Skill tool
- [ ] Using ONLY classDef styles from the mermaid skill (no invented colors)
- [ ] Diagram will include a color legend table

---

## Related Skills

- `/make-experiment-diag` - Parent skill for lens selection
- `/mermaid` - MUST BE LOADED before creating diagram
- `/exp-lens-validity-threats` - For internal validity and alternative explanations
- `/exp-lens-measurement-validity` - For whether metrics measure what matters