Findings No. 61
Measure time-to-interactive on a real phone before putting a bot check in front of content
Enabling BotID Deep Analysis on CCRTA's maps took mobile time-to-interactive from about 0.5s to about 20s on a real phone. It was reverted the same morning.
A verification step that is correct and cheaper can still be unusable. The only way to know is to time it on the device your users actually have.
To control the CCRTA website's Google Maps bill, the interactive map loads only after a verdict that the visitor is real. We tried Vercel BotID's Deep Analysis mode for that verdict. On a real phone, time to an interactive map went from about 0.5 seconds to about 20 seconds, on the agency's most-used page, for the device that accounts for 93% of real visits. We had verified that the change was correct and that it cut cost. We had not timed it on a phone before shipping.
We also found that an AbortSignal.timeout(3000) meant to cap the verdict request did not cap it. Our best explanation is that BotID wraps fetch and holds the request before our timeout applies, but we have not confirmed the mechanism. Deep Analysis was reverted the same morning and replaced with a gate based on the input device: phones load the map immediately, and desktop activates it on the first mouse move, scroll or keypress.
The general rule: before putting any challenge or verification in front of content people are waiting for, measure time-to-interactive on a real mobile device. And do not assume your own timeout bounds a third-party fetch wrapper you do not control.
Receipt: CCRTA Website, July 26, 2026 (docs/superpowers/specs/2026-07-26-pointer-aware-activation-design.md; CHANGELOG, desktop map activation; FINDINGS.md, "Google Maps / cost").