Outside Read

1 rounds · 4 agents per round·7 agents

Four vendor-diverse attack-vector reviewers examine your question in parallel — no cross-talk. A synthesizer distills their outputs into a battlefield map (core_agreement / concessions / irreducible_conflict / risk_ranking). BlindCritic and FactChecker audit the map before it reaches you. Three presets: find-blind-spots, critique-draft, pre-mortem.

Claude
AssumptionHunter

Attack vector: surfaces load-bearing assumptions the user has not named.

Gemini
BaseRate

Attack vector: grounds claims in reference class and historical frequency.

OpenAI
FailureMode

Attack vector: enumerates failure modes the proposal has not addressed.

Grok
SecondOrder

Attack vector: surfaces downstream and second-order consequences.

Claude
BattlefieldSynthesizer

Synthesizer — produces the locked battlefield-map output contract (web-off).

BlindCritic

Logical/structural auditor — no tools; blindness is the audit mechanism.

FactChecker

Factual auditor — web search to verify verifiable claims in the map.