Deductive Adversarial Review Integration
Input: $ARGUMENTS
Overview
Bridges the deductive strategy system with adversarial review testing. The deductive system provides logical derivations (proofs). Builder-breaker provides adversarial attacks. Together, they produce strategies that are both logically derived (not guessed) AND battle-tested (survived attacks).
This is the highest level of strategy confidence.
Steps
Step 1: Derive Strategy Deductively
First, derive the strategy using the logical proof system:
- → INVOKE: /lps (logical proof system) with the input
- Obtain:
- The derived strategy
- The axioms it rests on
- The inference chain
- The proof strength
- The critical assumptions
DEDUCTIVE OUTPUT:
Strategy: [what was derived]
Proof strength: [strong / moderate / weak]
Critical assumptions:
1. [assumption] — confidence: [H/M/L]
2. [assumption] — confidence: [H/M/L]
Weakest inference: [which step is least certain]Step 2: Convert Weaknesses to Attack Targets
From the deductive analysis, identify what to attack:
| Attack Target | Source | Why It’s Vulnerable |
|---|---|---|
| [assumption 1] | Critical assumption | If false, strategy collapses |
| [inference X] | Weakest step | Relies on [type of reasoning] |
| [axiom Y] | Problem definition | May not be complete/accurate |
| [missing factor] | Completeness gap | Not in the proof at all |
Step 3: Run Adversarial Review
→ INVOKE: /advr (adversarial review) targeting the identified weaknesses:
Builder presents: The derived strategy with its proof chain Breaker attacks: Each identified vulnerability, plus any additional weaknesses found
For each attack:
ATTACK: [description]
Target: [which part of the proof]
Severity: [fatal / serious / minor]
Builder response: [refute / repair / reinforce / concede]
Proof impact: [proof holds / proof weakened / proof collapses]Step 4: Update Proof Based on Attacks
After adversarial review, reassess the derivation:
| Original Assessment | After Attack | Change |
|---|---|---|
| Proof strength: [X] | Proof strength: [Y] | [stronger/same/weaker] |
| Assumption 1: [confidence] | Assumption 1: [new confidence] | [up/same/down] |
| Assumption 2: [confidence] | Assumption 2: [new confidence] | [up/same/down] |
| Strategy: [derived] | Strategy: [revised?] | [unchanged/modified/abandoned] |
Step 5: Classify Final Confidence
| Level | Criteria | Description |
|---|---|---|
| Proven | Strong proof + survived all attacks | As confident as we can be |
| Robust | Moderate proof + survived most attacks | High confidence, some uncertainty |
| Plausible | Weak proof but survived key attacks | Reasonable but not certain |
| Fragile | Strong proof but fell to attacks | Logically sound but practically vulnerable |
| Dubious | Weak proof and fell to attacks | Low confidence — reconsider |
| Refuted | Proof collapsed under attack | Strategy should be abandoned |
Step 6: Handle Each Outcome
If Proven/Robust: Proceed with the strategy. Document the proof and attacks it survived.
If Plausible: Proceed cautiously. Identify what additional evidence would upgrade confidence. Test the weakest assumptions first.
If Fragile: The logic is sound but reality may not cooperate. Design experiments to test the vulnerable assumptions. Have contingency plans.
If Dubious/Refuted: Do NOT proceed. Either:
- Find new axioms and re-derive
- Abandon this approach and try a different strategy
- Gather more information before re-attempting
Step 7: Report
DEDUCTIVE-ADVERSARIAL INTEGRATION:
Strategy: [what was analyzed]
Deductive phase:
- Proof strength: [level]
- Critical assumptions: [N]
- Weakest inference: [which]
Adversarial phase:
- Attacks attempted: [N]
- Attacks survived: [N]
- Fatal attacks: [N]
- Strategy modified by attacks: [Y/N — how]
Final confidence: [Proven / Robust / Plausible / Fragile / Dubious / Refuted]
If proceeding:
- Strategy: [final version after modifications]
- Key risk: [weakest surviving point]
- Monitor: [what to watch for]
If not proceeding:
- Why: [which attacks were fatal]
- Alternative: [what to do instead]When to Use
- After deductive strategy discovery
- Before final strategy selection on high-stakes decisions
- When you need maximum confidence in a strategy
- → INVOKE: /lps (logical proof system) for the deductive phase
- → INVOKE: /advr (adversarial review) for the attack phase
- → INVOKE: /cv (convergent validation) for additional validation methods
Verification
- Deductive derivation completed BEFORE adversarial review
- Weaknesses from derivation used as attack targets
- All attack types attempted (evidence, reasoning, completeness)
- Proof level updated based on attacks survived
- Final confidence level is honest assessment
- Next steps appropriate to confidence level