Journal · 18 April 2026

Sparse GPS and funnel drop-off

A clean trace is a privilege. Most first sessions in consumer apps are a handful of pings with holes. If you drop those rows, your activation rate is a sampling artefact.

OS battery policy, basements, elevated rail, and users who tap “allow once” all produce sparse GPS. The warehouse temptation is to keep only traces that look continuous enough to chart. Those traces disproportionately belong to power users who already converted. You have just built a funnel that congratulates itself.

Keep the holes, label them

We ask studios to add a confidence field on visit events: observed, inferred, or unknown. Observed means pings on both sides of the action within a stated window. Inferred means a plausible path with a gap. Unknown means you should not quote the row in a review.

Attribution windows then become explicit. A conversion can be credited to a fence only if confidence is observed or inferred, and the slide must show the mix. Finance teams dislike this for a week. Then they notice the number finally moves with receipts.

Drop-off that is not drop-off

Many “exited zone without converting” steps are simply the SDK going quiet indoors. That is an indoor session cut, not a user walking away. Mixing the two inflates leak metrics and sends PMs to redesign a checkout that was never the problem.

A modest rule

Never delete a trace solely because it is gappy. Filter in the query, with the filter named, when a specific analysis needs continuity. Default tables should be noisy and true. Pretty is a view, not a fact.

This is week three of District Signal Studio. Teams who arrive certain their drop-off is real often leave less certain, which we count as progress.