Wicked Decisions

Overview

Wicked Decisions is PublicRisk.ai's implementation of an Issue-Based Information System (IBIS) for navigating complex policy challenges that resist traditional problem-solving approaches. Built on Horst Rittel's theory of "wicked problems," it helps decision-makers explore strategy spaces under deep uncertainty using copula-based statistical modeling and pseudo-quantum entanglement weights.

What Are Wicked Problems?

Horst Rittel's Definition (1973)

In their seminal paper "Dilemmas in a General Theory of Planning", Horst Rittel and Melvin Webber identified 10 characteristics that distinguish "wicked" problems from "tame" (solvable) problems:

The 10 Properties of Wicked Problems

1. No Definitive Formulation

   • The problem can't be definitively described
   • Understanding the problem IS the problem
   • Example: "How do we prevent sexual abuse in schools?" encompasses culture, reporting, support, prevention—each could be THE problem

2. No Stopping Rule

   • You can always do more analysis, gather more data
   • No signal that says "you're done"
   • Example: School safety planning could continue indefinitely—more training, more counselors, more cameras

3. Solutions Are Not True or False, But Good Enough

   • Not provably correct/incorrect
   • Judged as "better/worse" or "good enough/not good enough"
   • Example: Evacuation plan for terrorist attack—can't prove it's "right," only that stakeholders accept it

4. No Immediate Test of Solution

   • Consequences play out over years or decades
   • Can't run controlled experiments
   • Example: Culture change interventions take 5-10 years to show effects; by then, context has changed

5. Every Solution Is a "One-Shot Operation"

   • No opportunity for trial-and-error learning
   • Each attempt creates consequences that constrain future attempts
   • Example: Can't practice responding to school shooters; first attempt sets precedent

6. No Enumerable Set of Potential Solutions

   • Infinite possible interventions
   • No exhaustive list of "all options"
   • Example: Infinite combinations of training intensity, reporting systems, support resources

7. Essentially Unique

   • Every wicked problem is novel
   • Past solutions don't transfer cleanly
   • Example: Sexual abuse prevention at Stanford ≠ prevention at rural K-12 district

8. Can Be a Symptom of Another Problem

   • Nested hierarchy of problems
   • "Solving" one level reveals deeper issues
   • Example: Low reporting rates → symptom of distrust → symptom of past mishandling

9. Discrepancy Can Be Explained in Numerous Ways

   • Multiple competing explanations for why problem exists
   • Choice of explanation determines solution approach
   • Example: Olympic security failure—bad training? Coordination? Technology? Funding? All valid

10. Planner Has No Right to Be Wrong

    • Stakes are high—lives, trauma, public trust
    • Unlike engineers, policy planners face moral/political consequences for failures
    • Example: School shooter response failure → career-ending, lawsuit-inducing

Issue-Based Information System (IBIS)

Theoretical Foundation

IBIS was developed by Horst Rittel in the 1970s as a deliberation framework for wicked problems. It structures argumentation to make reasoning transparent and collaborative.

Core IBIS Elements

1. Issues (Questions)

• Central decision points
• Framed as questions: "How should we allocate resources?"
• Example: "What mix of training, reporting, and support prevents abuse most effectively?"

2. Positions (Alternatives)

• Possible answers to the issue
• Competing strategies
• Example: "High training + High reporting + Med support" (Bundle A)

3. Arguments (Pros/Cons)

• Reasons supporting or opposing positions
• Evidence, values, constraints
• Example Pro: "High training reduces incident rate by 40% (Doe et al. 2023)"
• Example Con: "High training costs $2M annually, exceeds budget"

4. Sub-Issues

• Arguments spawn new questions
• Recursive structure—issues nest inside issues
• Example: "Which training model is most cost-effective?" (spawned from cost argument)

How Wicked Decisions Implements IBIS

Issue: Resource allocation for complex policy problem

Positions: 4 strategy bundles (A, B, C, D) with varying resource levels

Arguments: Outcome probability distributions (Culture Change, Early Intervention, Compliance)

Sub-Issues: Correlations (copula parameters), uncertainty (entropy), sensitivity analysis

Wicked Decisions Methodology

Simulation Framework

Wicked Decisions uses probabilistic modeling to explore the strategy space when:

• Cause-effect relationships are uncertain
• Outcomes emerge over long timeframes
• Stakeholders disagree on problem definition
• Trade-offs are value-laden

Three-Layer Architecture

Layer 1: Strategy Bundles (Positions)

4 Bundles representing resource allocation across:

Resource Levels:

• High: 80-100% of maximum feasible investment (e.g., annual training, 24/7 hotline)
• Med: 40-60% investment (e.g., quarterly training, business-hours hotline)

Layer 2: Outcome Distributions (Arguments)

Each bundle generates 3 outcome probabilities (High/Med/Low) for:

1. Culture Change

Definition: Fundamental shift in norms, attitudes, behaviors around the issue

High (0.35-0.60):

• Bystander intervention becomes norm
• Reporting viewed as duty, not betrayal
• Leadership openly discusses issue

Med (0.30-0.50):

• Some awareness, inconsistent action
• Pockets of change, others resistant
• Superficial compliance

Low (0.05-0.15):

• Status quo persists
• Denial, minimization
• No behavioral change

Measurement: Trust surveys, incident rates over 3-5 years, climate assessments

2. Early Intervention

Definition: Threats detected and neutralized before escalation

High (0.40-0.60):

• 70%+ of concerning behaviors flagged early
• Rapid response (< 24 hours)
• Successful de-escalation

Med (0.30-0.50):

• 30-60% detection rate
• Delayed response (2-7 days)
• Some preventable incidents occur

Low (0.05-0.20):

• Reactive, crisis-driven
• Late detection (after incident)
• Missed warning signs

Measurement: Red flag reports, time-to-response metrics, prevented incident logs

3. Compliance

Definition: Adherence to policies, regulations, and protocols

High (0.40-0.60):

• 85%+ policy adherence
• Regular audits pass
• Documentation complete

Med (0.30-0.50):

• 50-75% adherence
• Gaps in coverage
• Inconsistent documentation

Low (0.05-0.20):

• Non-compliance widespread
• Regulatory violations
• Failed audits

Measurement: Audit results, regulatory inspection reports, staff surveys

Layer 3: Correlations & Uncertainty (Sub-Issues)

Copula Modeling

Purpose: Capture how outcomes influence each other (not independent)

t-Copula Parameters:

• ρ (rho) matrix: 3×3 correlation matrix

         Culture  Early  Compliance
  Culture   1.0    0.40      0.25
  Early    0.40    1.0       0.35
  Compliance 0.25  0.35      1.0

• df (degrees of freedom): Tail dependency (df=5 means extreme outcomes cluster)
• λ (lambda) entanglement weights:
  • λ_TR: Training-Reporting amplification (0.15-0.30)
  • λ_RS: Reporting-Support amplification (0.15-0.30)

Interpretation:

• ρ = 0.40 → If Culture Change is HIGH, Early Intervention is 40% more likely to be HIGH
• λ_TR = 0.28 → Training boosts Reporting effectiveness by 28%

Entropy (Shannon)

Purpose: Measure outcome uncertainty

Formula: H = -Σ p(x) log p(x)

Values:

• Low entropy (< 1.0): Predictable, low variance
• Medium entropy (1.0-1.5): Moderate uncertainty
• High entropy (> 1.5): Highly unpredictable, wide distribution

Interpretation: Higher entropy = more uncertainty = more risk

Pseudo-Quantum Metaphor

Concepts:

1. Superposition: Before decision, all strategies exist simultaneously
2. Amplitudes: Probability weights for each strategy (like wave function amplitudes)
3. Entanglement: Outcomes correlated across strategies (copula models this)
4. Collapse: Decision "collapses" system to one bundle (measurement)
5. Manifest: Metadata about the quantum-inspired simulation state

Manifest Fields:

• entropy: System uncertainty (Shannon entropy)
• seed: Random seed for reproducibility
• streamId: Unique identifier for this bundle's "trajectory"
• amplitudes: Probability weights [a1, a2, a3, a4]
• collapsePolicy: When wave function collapses ("per-scenario" = at decision time)

Use Cases

1. Sexual Abuse Prevention (K-12/Higher Ed)

Wicked Problem Characteristics:

• ✅ No definitive formulation (culture? training? reporting? all?)
• ✅ No stopping rule (can always add more safeguards)
• ✅ Solutions are "good enough" (can't prove 100% prevention)
• ✅ No immediate test (culture change takes years)
• ✅ One-shot (can't practice responding to abuse reports)
• ✅ Unique (each institution different)

Strategy Bundles:

2. Olympic Games Risk Management

Wicked Problem Characteristics:

• ✅ Essentially unique (each Olympics has different threat landscape)
• ✅ No immediate test (can't stage rehearsal Olympics)
• ✅ One-shot operation (no trial runs)
• ✅ Planner has no right to be wrong (global scrutiny, terrorism risk)

Strategy Focus:

• Training: First responders, security personnel, venue staff
• Reporting: Threat intelligence sharing, crowd monitoring, suspicious activity hotlines
• Support: Medical surge capacity, evacuation infrastructure, communication systems

Outcome Interpretations:

• Culture Change: Inter-agency collaboration, security mindset
• Early Intervention: Threat detection before attack
• Compliance: Security protocol adherence across 50+ venues

Example Decision:

• Bundle A: Maximum security (high cost, intrusive)
• Bundle B: Intelligence-led (fewer personnel, more technology)
• Bundle C: Visible deterrence (high training, public-facing)
• Bundle D: Balanced (moderate all dimensions)

3. Terrorist Attack Response (Urban)

Wicked Problem Characteristics:

• ✅ Symptom of another problem (geopolitics, radicalization, inequality)
• ✅ Discrepancy explained in numerous ways (ideology? mental health? foreign actors?)
• ✅ No stopping rule (can always add more surveillance, security)

Strategy Focus:

• Training: Multi-agency coordination, ICS/NIMS, tactical response
• Reporting: Fusion centers, social media monitoring, community tips
• Support: Mass casualty protocols, trauma care, family reunification

Outcome Interpretations:

• Culture Change: "See something, say something" adoption
• Early Intervention: Plot disruption before attack
• Compliance: Unified command adherence, proper evidence handling

Example Scenario:

• High-casualty bombing in downtown district
• 4 bundles explore resource trade-offs between prevention (training), detection (reporting), and response (support)
• Copula models coordination effects (training amplifies reporting effectiveness)

4. School Shooter Event (K-12 Active Threat)

Wicked Problem Characteristics:

• ✅ No right to be wrong (children's lives, community trust)
• ✅ One-shot operation (can't practice live shooter response)
• ✅ No immediate test (may go years without incident)
• ✅ Solutions are "good enough" (can't guarantee zero casualties)

Strategy Focus:

• Training: Lockdown drills, threat assessment teams, SRO collaboration
• Reporting: Behavioral threat assessment, anonymous tip lines, social media monitoring
• Support: School counselors, trauma response, parent communication

Outcome Interpretations:

• Culture Change: Students/staff comfortable reporting concerns
• Early Intervention: Warning signs detected, interventions deployed
• Compliance: Lockdown procedures followed, doors secured

Example Decision:

• Bundle A: Intensive training + 24/7 tip line + full-time counselors
• Bundle B: Moderate training + technology (AI monitoring) + rapid response
• Bundle C: Emphasis on threat assessment teams + community partnerships
• Bundle D: Basic drills + standard protocols + limited counseling

How to Use Wicked Decisions

       Culture  Early  Compliance
Culture   1.0    0.40      0.25
Early    0.40    1.0       0.35
Compliance 0.25  0.35      1.0

Interpreting Results

Probability Distributions

What "High = 0.58" Means:

• 58% chance of achieving HIGH culture change (fundamental norm shift)
• 36% chance of MED culture change (modest improvements)
• 6% chance of LOW culture change (status quo)

Not a guarantee: Probabilities represent uncertainty given assumptions

Copula Parameters

What "ρ = 0.40" Means:

• If culture change outcome is 1 standard deviation above mean, early intervention outcome is expected to be 0.40 standard deviations above mean
• Positive correlation: outcomes tend to move together

Not causation: Correlation ≠ "training causes reporting to improve"; rather, they covary

Entropy

What "H = 1.42" Means:

• Measured in bits (information theory)
• Maximum entropy for 3 outcomes = log2(3) ≈ 1.58
• 1.42 is close to maximum → high unpredictability
• Less than 1.0 = predictable, greater than 1.3 = highly uncertain

Not a quality metric: High entropy isn't "bad"—it reflects genuine uncertainty in complex systems

Limitations & Caveats

What This Tool IS:

✅ Structured thinking framework for wicked problems
✅ Trade-off exploration tool (sensitivity to assumptions)
✅ Deliberation facilitator (IBIS argumentation capture)
✅ Uncertainty quantification (entropy, correlation awareness)
✅ Stakeholder communication aid (visualize strategy space)

What This Tool IS NOT:

❌ Predictive model (probabilities are illustrative, not empirical)
❌ Optimization algorithm (no "best" solution for wicked problems)
❌ Decision automation (requires human judgment and values)
❌ Empirically validated (probabilities not calibrated from real incident data—yet)
❌ Quantum computing (quantum metaphor is metaphorical, not literal)

Appropriate Use Cases:

✅ Strategic planning sessions (exploring "what if?")
✅ Stakeholder workshops (structuring debate)
✅ Grant proposals (justifying resource allocation)
✅ Policy analysis (comparing alternatives)
✅ Training exercises (teaching systems thinking)

Inappropriate Use Cases:

❌ Legal defense ("model said 0.58, so we're not liable")
❌ Sole basis for resource allocation (need real data + expert judgment)
❌ Replacing subject matter experts (model complements, not replaces)
❌ Short-term tactical decisions (model is strategic, not tactical)

Future Enhancements

Backend Integration (PRODUCTION READY)

Status: ✅ Deployed to Modal Cloud Platform

Live Endpoints:

• GET /api/wicked-decisions/decisions - List all saved scenarios
• GET /api/wicked-decisions/{id} - Retrieve specific scenario
• POST /api/wicked-decisions/generate - Generate new scenario with copula modeling
• POST /api/wicked-decisions/simulate - Run Monte Carlo simulation (10K+ runs)
• GET /health - Service health check

Features Implemented:

1. ✅ Python statistical engine (scipy multivariate t-distribution)
2. ✅ Monte Carlo simulation (10,000+ runs for confidence intervals)
3. ✅ Database persistence (PostgreSQL on Supabase)
4. 🔄 Empirical calibration (Bayesian updating - roadmap Q1 2026)
5. 🔄 Sensitivity analysis (Sobol indices, tornado charts - roadmap Q1 2026)

Performance:

• Scenario generation: <2 seconds
• Monte Carlo simulation: <5 seconds (10K runs)
• Cold start: 30-60 seconds (Modal container initialization)
• Warm requests: <1 second

Architecture:

• Backend: Modal Cloud Platform (Python 3.11, FastAPI)
• Statistical Engine: scipy.stats (multivariate_t, copula modeling)
• Database: PostgreSQL (Supabase)
• Authentication: JWT tokens via Supabase Auth
• Deployment: Auto-scaling serverless functions

For implementation details, see Backend Requirements.

Data Integration

Potential Data Sources:

• Education: Department of Education CRDC (Civil Rights Data Collection)
• Healthcare: HHS Breach Portal, HIPAA violation logs
• Public Safety: FBI UCR, NIBRS incident reports
• Regulatory: EPA, OSHA, state agency inspection databases

Calibration Example:

• Upload 500 K-12 sexual abuse incident reports (2020-2024)
• System calculates empirical distribution (% High/Med/Low culture change)
• Bayesian update: Prior (expert estimate) + Empirical (data) → Posterior (calibrated probability)

Advanced Features

1. Scenario Comparison: Side-by-side bundle comparison across multiple scenarios
2. Temporal Dynamics: Model how outcomes evolve over years (ARIMA time series)
3. Multi-Objective Optimization: Pareto frontier analysis (trade-off curves)
4. Stakeholder Weights: Allow users to weight outcomes by importance
5. Real-Time Collaboration: Shared IBIS tree with live editing (Google Docs-style)

Theoretical Foundations

Key References

Horst Rittel & Melvin Webber (1973) "Dilemmas in a General Theory of Planning"

• Defined wicked problems
• Introduced IBIS framework
• Policy Sciences, Vol. 4, pp. 155-169

Jeff Conklin (2006) "Dialogue Mapping: Building Shared Understanding of Wicked Problems"

• Practical IBIS guide
• Facilitation techniques
• Wiley, ISBN: 978-0470017685

Nancy Roberts (2000) "Wicked Problems and Network Approaches to Resolution"

• Network theory + wicked problems
• Collaborative governance
• International Public Management Review, Vol. 1

Sklar & Dietrich (2001) "Copulas and Markov Processes"

• Mathematical foundation for copula modeling
• Risk aggregation
• Springer

Related Concepts

Decision Theory:

• Multi-criteria decision analysis (MCDA)
• Robust decision-making (RDM)
• Scenario planning

Systems Thinking:

• Causal loop diagrams
• Stock and flow models
• Leverage points (Donella Meadows)

Deliberative Democracy:

• Consensus conferences
• Citizens' assemblies
• Participatory budgeting

Best Practices

1. Involve Stakeholders Early

Why: Wicked problems are socially constructed—different stakeholders see different problems

How:

• Workshop with 10-15 diverse stakeholders
• Present 4 bundles, ask: "Which resonates? Why?"
• Capture arguments in IBIS structure

Example:

• Principal says: "Bundle A best—culture change is long-term investment"
• CFO says: "Bundle D best—can't afford $500K"
• Title IX coordinator says: "Bundle B best—victims need support NOW" → Surface value disagreements explicitly

2. Update Probabilities with Data

Why: Default probabilities are placeholders, not predictions

How:

• Gather incident data (past 3-5 years)
• Calculate empirical distributions
• Use Bayesian updating to combine expert judgment + data

Example:

• Prior: Culture change High = 0.58 (expert estimate)
• Empirical: 142/500 incidents showed high culture change = 0.284
• Posterior: 0.2 × 0.58 + 0.8 × 0.284 = 0.343 (calibrated)

3. Test Sensitivity

Why: Wicked problems are sensitive to assumptions

How:

• Generate 10 scenarios with randomized amplitudes
• Observe if Bundle A consistently outperforms
• If results flip-flop, strategy is not robust

Example:

• Scenario 1-5: Bundle A wins
• Scenario 6-10: Bundle D wins → Conclusion: Decision is sensitive; need more data or compromise (Bundle C)

4. Document Reasoning

Why: Future decision-makers need to understand rationale

How:

• Capture IBIS tree (Issue → Positions → Arguments)
• Export to PDF/Markdown
• Store in institutional knowledge base

Example:

• 2025: Choose Bundle A based on culture change priority
• 2030: New leadership questions decision—review IBIS tree, understand context

5. Revisit Periodically

Why: Wicked problems evolve; yesterday's solution is today's problem

How:

• Quarterly review of bundle performance
• Update probabilities with new incident data
• Adjust resource allocation as context changes

Example:

• Year 1: Bundle A implemented
• Year 2: Culture change showing early success (increase early intervention investment)
• Year 3: Budget cut—shift to Bundle C (maintain training/support, reduce reporting)

FAQ

Why "pseudo-quantum"?

A: The quantum metaphor illustrates key concepts:

• Superposition: Multiple strategies coexist before decision
• Entanglement: Outcomes correlated (copula models this)
• Collapse: Decision "measures" system, forcing one outcome

It's pedagogical, not physical. No actual quantum mechanics involved.

Are probabilities empirically validated?

A: Not yet. Default probabilities are expert estimates (calibrated guess). Backend integration (planned) will enable Bayesian calibration from real incident data.

Can I use this for legal compliance?

A: No. This is a planning tool, not a compliance audit. Probabilities are illustrative. Use subject matter experts + legal counsel for compliance decisions.

How do I choose between bundles?

A: No algorithm can choose for you (wicked problems lack objective criteria). Use this tool to:

1. Surface stakeholder priorities (culture vs. early intervention?)
2. Quantify trade-offs (cost vs. outcome probability)
3. Test robustness (sensitivity analysis)
4. Document reasoning (IBIS tree)

Then make value-based judgment.

What if my problem doesn't fit the 3 outcomes?

A: Current implementation is fixed to Culture Change / Early Intervention / Compliance. Backend (planned) will support custom outcomes. For now, map your problem to these categories:

• Culture Change = Long-term systemic shift
• Early Intervention = Immediate threat detection
• Compliance = Rule adherence

Can I export results?

A: Yes (once backend deployed). Planned export formats:

• PDF (board presentation)
• Excel (further analysis)
• JSON (API integration)
• Markdown (documentation)

Next Steps

Glossary

Amplitude: Probability weight in quantum-inspired model (like wave function amplitude)

Bundle: Resource allocation strategy (position in IBIS framework)

Copula: Statistical function modeling joint probability distribution and correlation

Entropy: Measure of uncertainty (Shannon entropy: H = -Σ p log p)

IBIS: Issue-Based Information System—argumentation framework for wicked problems

Lambda (λ): Entanglement weight (amplification factor for correlations)

Manifest: Metadata about simulation state (entropy, seed, amplitudes)

Position: Alternative solution to issue (IBIS term)

Rho (ρ): Correlation coefficient in copula matrix

Wicked Problem: Policy challenge with no definitive formulation or solution

Last updated: December 6, 2025
Theory: Horst Rittel (1973), Jeff Conklin (2006)
Implementation: PublicRisk.ai Wicked Decisions Module