bayesian_kpi_predictions

bayesian_kpi_predictions(
    estimates,
    as_of,
    *,
    forecast_source,
    settings,
    backtest_source=None,
    publications=None,
)

Apply the Bayesian mathematics to point-in-time panel data.

Independent and correlation adjusted are the two methods. Historical estimates fit strictly earlier growth observations; forward estimates fit all realized history available before as_of. The workflow anchors data, estimates moments, then publishes named results. Existing BayesianPosterior callers retain their tutorial-specific behavior.

Parameters

Name	Type	Description	Default
estimates	pd.DataFrame	Long panel with actual, consensus, and model observations. Consensus rows may carry dispersion and estimate_count.	required
as_of	pd.Timestamp	Explicit exclusive UTC reading moment.	required
forecast_source	str	Live forecast source name.	required
settings	BayesianPredictionSettings	Versioned method, history, bias, dispersion and output options.	required
backtest_source	str \| None	Historical model source; defaults to forecast_source.	`None`
publications	pd.DataFrame \| None	Subject keys and UTC publication_date, required for publication-relative reads. Callers supply the known schedule.	`None`

Returns

Name	Type	Description
	pd.DataFrame	One row per subject with the requested named outputs, eligibility, resolved relative_days, as_of, method, settings_version, preset and fitted_through. Missing predictions retain their subject and reason. Diagnostics may remain available when a prediction is refused.

Raises

Name	Type	Description
	PanelError	The settings were fitted on data reaching as_of.