Constitutional Orthogonality

Pattern: Assign agents contradictory optimization criteria to surface blind spots through adversarial review.

Problem

Single-agent systems optimize one dimension. Adding more agents with same mandate increases throughput but not correctness. Agents reinforce each other's biases.

Solution

Define orthogonal constitutions: agents with conflicting success criteria. When proposals satisfy opposing mandates, confidence increases. When they fail one mandate, blind spots surface.

Implementation

1. Define Orthogonal Pairs

Example from space-os:

2. Decision Review Flow

Proposer creates decision
  → Swarm reviews (including orthogonal agents)
  → Challengers cite concerns
  → Proposer addresses or decision rejected
  → Commit requires no active dissent

3. Measuring Convergence

When orthogonal agents agree:

When orthogonal agents disagree:

Metrics

Empirical Results

Security experiment (tribunal):

Constitutional orthogonality surfaces design-level issues that pattern matching misses.

Falsifiability

Pattern fails if:

References