Optional dashboard (`increment.dashboard`)
increment.dashboard is an optional presentation layer over one bound
experiment: section-level mo.Html helpers and thin adapters over the
public Analysis and CoefTable APIs. Install it with pip install "increment[dashboard]"; core Increment, estimation, power, and the
dataframe entry points never import marimo, CoefTable, or pandas
through this package. See the dashboard guide for
the caller-owned source/config/prepare/render recipe, supported experiment
shapes, and statistical disclosures.
DashboardConfig
Section titled “DashboardConfig”DashboardConfig(expected_allocation, title, source_label, provenance, metric_units)Display configuration a caller supplies alongside their Analysis.
DashboardSnapshot
Section titled “DashboardSnapshot”DashboardSnapshot(experiment_name, binding_fingerprint, title, description, start, end, control_group, treatment_group, primary_metric, metrics, breakouts, estimates, readout_rows, allocation, allocation_refusal, allocation_history, allocation_history_refusal, config, computed_at, group_data)One prepared reading of an experiment: metadata, allocation, results.
Read-only by construction. It holds no connection and is not a portable analysis format; reprepare from the source instead of persisting it.
increment.dashboard.ExploreView is the exported literal type
"cumulative_lift" | "daily_values" | "cumulative_values" | "segments",
accepted by load_explore and render_explore.
prepare_dashboard
Section titled “prepare_dashboard”prepare_dashboard(analysis, config)Run the headline analysis once and capture it as a read-only snapshot.
This is the only place the dashboard computes results. Display controls
downstream never recompute; reexecuting the source cell is an intentional
new snapshot. The caller keeps ownership of analysis and its
connection: nothing here closes or replaces them.
dashboard_styles
Section titled “dashboard_styles”dashboard_styles()The packaged, scoped stylesheet as one style block.
Read from package data, so an installed wheel needs no stylesheet path and no repository checkout.
render_header
Section titled “render_header”render_header(snapshot)Experiment identity, analysis policy, and the headline summary.
render_health
Section titled “render_health”render_health(snapshot)The allocation and caveat section, shown above the results.
render_results
Section titled “render_results”render_results(snapshot)The whole declared family, its readout table, and its disclosures.
render_metric_details
Section titled “render_metric_details”render_metric_details(snapshot, metric)Definition, tested tail, policy, and observed group evidence for one metric.
load_explore
Section titled “load_explore”load_explore(analysis, snapshot, metric, view, completed_windows_only, breakout)Load exactly the one advanced view a caller asked for.
A thin dispatch over the public readout API. It never reruns the headline
family, the allocation check, or materialization, so changing a display
control cannot change a verdict. metric=None selects every declared
metric in snapshot order; a metric name selects just that metric.
render_explore
Section titled “render_explore”render_explore(snapshot, data, metric, view, completed_windows_only)The requested CoefTable view, with context and unavailable-point reasons.
render_details
Section titled “render_details”render_details(snapshot)Declared policy and optional provenance, with no source access.
readout_csv
Section titled “readout_csv”readout_csv(snapshot)Export every unrounded headline row; unavailable numbers stay empty.
Formula markers after leading whitespace and control-prefixed strings receive an apostrophe for safe human-facing downloads.
group_data_csv
Section titled “group_data_csv”group_data_csv(snapshot, metric)Export exactly the rows returned by :func:group_data_rows.