Pdr Reasoning

Overview & Purpose

Predict-Disrupt-Reflect (PDR) is a adversarial reasoning protocol designed to combat optimism bias and build bulletproof strategies. It forces the agent to explicitly predict an outcome, actively try to disrupt/break the prediction with worst-case scenarios, and reflect to refine the strategy.

When to Use

• Plan Validation & Risk Assessment: Before deploying critical code, database migrations, or major architectural changes.
• Adversarial Red-Teaming: Identifying edge cases, security vulnerabilities, or single points of failure.
• High-Stakes Decision Making: Where failure carries high costs and mitigation must be built in up front.

Execution Workflow

1. Predict (Base Strategy):

   • State the proposed plan, expected sequence of events, and intended successful outcome.

2. Disrupt (Red Team Attack):

   • Introduce severe disruption scenarios: "What if the DB times out?", "What if memory spikes?", "What if an unexpected input is received?".
   • Actively attempt to break the plan.

3. Reflect & Harden:

   • Analyze how the plan failed under disruption.
   • Modify the plan to incorporate fallback mechanisms, circuit breakers, and contingency paths.

Expected Output Contract

### PDR Strategy Stress-Test
- **Prediction**: [Proposed plan and intended outcome]
- **Disruption Scenarios**:
  - *Disruption 1*: [Failure vector] -> *Impact*: [Severity]
- **Reflection & Mitigation**: [Hardened plan incorporating safeguards]

Scripts

• scripts/pdr_reasoning.py - Deterministic evaluation, state validation, and CLI tool for pdr-reasoning.