The camera bug that looked like user behavior

The funnel showed iOS users opening the scanner and never scanning — 6 camera grants, 4 users who "held the viewfinder and walked away." Section 7 of the funnel analysis concluded the explorer hypothesis was confirmed: curious non-drivers with no real parcels in hand.

Section 8 overturned it. Slicing the same events by appVersion showed the "open camera, never scan" users were all on iOS v0.5.8 — while 0.6.2-iOS users scanned normally (11/11 attempts landing). The real culprit was a build configuration: PERMISSION_CAMERA=1 missing from the Podfile, silently suppressing the OS camera prompt (fixed 2026-05-23; the store-distributed fix rode the 0.6.x releases). What looked like a user-psychology finding was one line of iOS build config.

The analysis document does the rare thing: it rewrites its own conclusion in place and closes with the lesson as a standing rule — "埋点自动带 appVersion —— 任何行为归因前先按版本切,否则会把'旧版 bug'读成'用户行为'." (Slice by version before attributing anything to behavior.) A second bug from the same episode: capture failures were swallowed by a bare debugPrint with no analytics event — "GA4 全瞎 —— 本次差点误判的根源" — the silence of the instrumentation was itself the misdirection.

Interview archetype: "tell me about a time you almost reached the wrong conclusion" — I published the wrong conclusion to myself, then my own version-slicing overturned it, and the correction became a methodology rule.