Machine learning and AI backlog¶
Open¶
ENH-003 - Ground semantic-model agents and gate ontology agents¶
- Priority / effort: Idea / M
- Outcome: Agents have business descriptions, synonyms, safe-use instructions, approved questions, and capability-aware rebinding.
ENH-004 - Prepare the semantic model for richer Copilot experiences¶
- Priority / effort: Idea / M
- Outcome: Verified answers, AI instructions, narratives, and anomaly explanations are published only after KPI, ML, and governance gates pass.
ENH-007 - Add trustworthy ML lineage and explainability¶
- Priority / effort: Idea / L
- Outcome: Every prediction includes model lineage, intended-use limits, feature evidence, calibration, and prediction-versus-outcome diagnostics.
Settled — do not reopen¶
IMP-008 - Align the active report with a trustworthy ML contract¶
-
Resolution: All 14 outputs have executable contracts and isolated tiers. The required validator and exact-run gate fail closed before Reporting. A live full-demo deployment with more than 540 days of history completed setup, required ML, optional ML, experimental ML, SQL metadata synchronization, semantic-model publication, and report validation. The same required gate used by
standardwas exercised successfully. -
ML outputs are not assumed to exist in a core deployment.
- Ontology and other preview-dependent automation require capability preflight.
- Semantic-model agents and ontology agents have separate support boundaries.