request_status
LoadingLoading ML readiness assets
modeled_rows
Offline baseline comparisons and validation experiments are underway using versioned feature and label marts, while the project deliberately stops short of deployed predictions, recommendations, readiness scores, or coaching.
experiment_status
offline_baselines: active
production_inference: none
The active work compares simple offline baselines and validates modeling assumptions. Inputs, labels, lineage, and missing-data behavior remain inspectable, but this site does not train or serve models, claim performance, or convert experimental outputs into athlete guidance.
Active baseline and validation work uses explicit grains, sources, and labels without presenting experimental results as live intelligence.
mart_route_prediction_featuresmart_weekly_training_featuresThese use cases are being explored against versioned feature and label marts. No model performance, published finding, production prediction, or athlete guidance is claimed.
experiment::01
Baseline comparisons are testing how route shape, prior route history, and recent training explain pace outcomes.
experiment::02
Simple baselines are being compared for next-week distance, duration, and active-week labels using completed rollups and lag fields.
experiment::03
Validation experiments are examining duration and completed-distance labels for recurring routes without turning results into coaching recommendations.
experiment::04
Exploratory grouping is testing whether observed routes, weeks, and sessions expose stable patterns worth expressing as descriptive analytics.
The useful outcome is not automatically a deployed model. Careful analysis can expose better definitions, missing context, and data-quality work.
review::01
Baseline error analysis and feature review can reveal missing context, unstable assumptions, and upstream data-quality gaps.
review::02
A result must remain stable, explainable, and analytically useful before it becomes part of the governed model surface.
review::03
Validated findings can become documented and tested gold metrics or features that improve future agent context and visual analysis.
serving_boundary
Only deliberately promoted gold outputs can become context for the planned read-only MCP interface. Experimental predictions are not passed through directly.
Review Agent Interfacemart_run_sessions, mart_daysmart_routes, mart_run_sessionsmart_route_prediction_features, mart_weekly_training_features