zicato-design-experiment

To design a pre-run hypothesis for multi-agent system experiments, analyze telemetry, drift metrics, and failure-mode profiles to connect observed loss patterns to permitted harness mutations. This prevents inventing predictions after seeing results by requiring read-only grounding and structured metric predictions.

Experiment design requires validating mutation points to ensure operators select only permitted harness targets and justify how changes affect observed failure patterns. This validation enforces strict separation between predictions and post-run outcomes, preventing confabulation and maintaining evidence-based reasoning.

The best way to structure experiment hypotheses using telemetry analysis is producing a schema-valid hypothesis with predicted drift movements, metric movements, pass-rate expectations, and risk analysis. Ground these predictions in read-only telemetry data and epoch goals before proposing changes.

Yes, you can use epoch analysis to predict pass-rate expectations for mutation planning by examining outcome failure-mode profiles and drift metrics. This grounds your predictions in observed loss patterns and validates them against allowed mutation points before running the experiment.

Evidence-based hypothesis authoring works for research task confabulation risk by formulating pre-run hypotheses that tighten researcher instructions and predict both quality improvements and possible tool-use costs. This approach requires read-only grounding and structured drift predictions to maintain integrity.

You should not use pre-run hypothesis authoring when telemetry data is unavailable or when mutation points cannot be validated against permitted harness targets. The approach requires read-only grounding, structured drift metrics, and strict separation between predictions and post-run outcomes to function correctly.