adversarial-cycle

Adversarial Cycle

Validate a claim through structured conclusion, counter-claim, response cycles with defined sufficiency requirements. Each claim passes through a three-phase cycle that must meet explicit sufficiency criteria before the claim is considered validated. Produces a determination of whether the claim survives scrutiny and in what form -- validated, modified, refuted, or early-exit (unfalsifiable within scope).

When to Use

• A specific claim or decision needs stress-testing before commitment
• You have a falsifiable claim with a supporting argument to challenge
• Evaluating a design choice, architecture decision, or protocol rule
• Testing a set of resolved decisions as a composed system (aggregated mode)
• Pre-flight check before investing in downstream impact analysis

When NOT to Use

• The question is about assumptions underlying a decision (use assumption-surfacing)
• You need to compare alternatives against criteria (use tradeoff-analysis)
• You need to trace downstream impact without first validating a claim (use downstream-forecasting directly)
• The claim is too vague to be falsifiable -- restate with commitment first

Execution

This protocol runs as the adversarial-cycle agent. The agent follows the full phase procedure autonomously, using the sequentialthinking MCP tool for phases marked (Sequential-thinking required).

• Protocol selection and composition: use /reason
• Domain research delegation: agent spawns domain-researcher as needed
• Phase validation: enforced by SubagentStop hook
• Without sequential-thinking: agent tags output validation_tier: Fallback

Input

• claim (required): The decision or position being stress-tested, stated as a falsifiable claim
• supporting_argument (required): The reasoning supporting the claim; provides the mechanistic base the counter-claim must attack
• decision_context (optional): Broader context framing why this claim matters; scopes counter-claims
• mode (optional): individual (default) or aggregated for testing composed decision sets

Output

• validation_status: validated | modified | refuted | early-exit
• final_claim: The surviving claim (present when validated or modified; absent when refuted or early-exit)
• cycle_transcript: Formatted summary block per cycle
• validation_tier: Full | Fallback

Composition

• Validate-and-Forecast (V-F): AC output feeds Downstream Forecasting via a validation gate. Refuted claims terminate the pattern; validated/modified claims proceed to impact forecasting.
• T-V pipeline: Receives the winning alternative from Tradeoff Analysis as a synthesized claim.
• T-V-F pipeline: Same as T-V, then forwards the validated claim to Downstream Forecasting.
• A-V-F pipeline: Receives bifurcated input from Assumption Surfacing adapter.
• Use /reason for full composition guidance, pattern selection, and pipeline specifications.