Gallery

Gallery

Small visual patterns for design review.

Unsafe Same-Run Prediction Feedback

AgentRun -> Predictor observes AgentRun -> Prediction controls AgentRun before it ends

Classification: causal_paradox

Safer External Orchestrator

AgentRun -> Supervisor observes AgentRun -> Orchestrator consumes result -> NextAgentRun

Classification: valid_acyclic

Post-Run Audit

AgentRun ends -> Evaluator scores AgentRun -> Audit record is written

Classification: valid_acyclic

Ordinary Local Loop

AgentRun increments local progress counter -> AgentRun continues until bounded budget expires

Classification: valid_acyclic when no modeled prediction about the whole current execution is consumed.

Ambiguous Identity

Supervisor produces r-1 -> somebody consumes r-1

Classification: insufficient_info

Question to answer: who consumes r-1, and does that consumer control the observed execution before it ends?