Cleanup Rule Bench

Synthetic data

Switch the cleanup rules on and off. See which rules buy precision, which buy recall, and what each one costs.

Back to the switches

Loading the evaluation data…

Scores for this configuration

Precision: of the pairs the tool joined, the share that are truly one person or one home. Recall: of the true pairs, the share the tool found.

Chips show the change against All rules on for the same usable runs. Ranges show the lowest and highest single run.

Rule impact

What happens to the selected score when you flip one duplicate or household switch, and keep all others as they are now.

Board matching

Board matching links donors to officers of nonprofits in IRS 990 filings. Move the sliders to recompute the matches from every stored candidate.

The test data has no planted board look-alikes, so the sliders mainly trade recall; precision stays at 100% at the default settings.

Methods and caveats