# Concepts

* [Decision graph](decision-graph.md) - The questions asked from a dataset card to a written AVU, and what decides each.
* [Levels and policy](levels-and-policy.md) - AnyJev levels, the calibration field, and how a decision becomes auto, proposed or abstain.
* [Provenance](provenance.md) - The sidecar that records every decision next to the mesa-ducklake AVU history.
* [Learning](learning.md) - Label sources and weights, leave-one-card-out fitting, the guards, promotion.
* [MCP tools](mcp-tools.md) - The mesa_decide_* tools inside mesa-mcp, one question per round trip, picks as labels.
* [Second opinions](second-opinions.md) - Claude as a recorded second opinion, specificity over child terms, the planner audit.
* [DataCite decisions](datacite.md) - The frozen DataCite vocabularies as questions and the record fragment for mesa-mcp.
* [Hosted providers](hosted-providers.md) - Hosted Jev through MotherDuck prompt_jev and the data-residency switch.