The whole surface, listed.

Retrieve prior decisions by semantic, keyword, hybrid or bridge search.

Evaluate a proposed action against the active guardrails.

Record a decision with context, reasons, stakes and confidence.

Log what actually happened — the input calibration depends on.

Calibration statistics: Brier score, accuracy and related measures.

Fetch one decision in full, including its deliberation trace.

Reason-type performance: which kinds of reasoning actually hold up.

Finalise or amend a decision record after the work settles.

Attach reasoning to an open decision while the work is in progress.

Primary. Queries similar decisions, evaluates guardrails, fetches calibration and extracts patterns — optionally recording the decision. One call replaces three.

Primary. Full cognitive context at session start: agent profile, relevant decisions, guardrails, calibration by category, ready queue, confirmed patterns. JSON or markdown.

Surface the queue of decisions awaiting outcome review.

Create a typed edge between two decisions.

Traverse the decision graph from a node.

List the immediate neighbours of a decision node.

Inspect the state of a circuit-breaker guardrail.

List all circuit breakers and their current state.