Delivered 2026-09-27 by spice-sp-fidelity-researcher: plans/spway/6a/TermSet.ModelTypes.xml (22 terms, 2 levels: EU AI Act general-purpose / systemic-risk; TypeSafe's System One decision-model with Choice/Score/Noul children; Hugging Face task categories with pipeline-tag leaves; ISO/IEC 22989's four tasks cross-referenced), TermSet.AgentTypes.xml (12 terms: single-agent, tool-using, workflow-agent with sequential/routing/parallelization/orchestrator-workers/evaluator-optimizer carrying Anthropic's and Microsoft's names as OtherLabels, human-in-the-loop, reactive/deliberative, the SPICE-only harness) and report.md. Shape correction: SPICE term sets are flat AddListItem rows against ContentType.Term (Path, ParentPath, OtherLabels), provenance on SolutionGallery rows, not a TermSet element. Disagreements left open in the terms' descriptions, not resolved silently (foundation model vs general-purpose AI model vs text-generation; Score's shape; no first-party TypeSafe page for Jev).