q.experiment roadmap

Implemented

Context-managed runs with RUNNING, FINISHED, and FAILED lifecycle states.
Context-local active-run propagation.
Backend-neutral parameters, metrics, tags, artifacts, datasets, models, and structured records.
Local DuckDB metadata and copied-file artifact backend.
MLflow 3 tracking through mlflow-skinny, without SQL storage, server, UI, data-science dependencies, or framework-specific model flavors.
Explicit access to the underlying MLflow client.
Redacted q.ai inference provenance and usage integration.
Optional pyqrt[experiment] installation profile.

Run search, comparison, and dataframe reporting APIs for the local backend.
Parent/child run relationships for batch rows, evaluation suites, and distributed jobs.
Dataset schema and split fingerprints shared with q.dataset.
Model registry references, aliases, promotion policy, and deployment handoff.
q.model, q.bt, and q.ray emitters using the same active-run contract.
OpenTelemetry trace/span correlation without recording content.
Artifact retention, remote object storage, and cleanup policies.
Authentication/connectivity diagnostics for MLflow servers.
MLflow HTTP-server contract tests and version compatibility matrix.