In one line
SKAN returns exactly one value per install, and that value is the conversion value. Deciding which few days of behavior gets compressed into that narrow slot is the core design task of iOS measurement.
Why the design is hard
The fine value is 0–63 — 64 slots. Trying to encode signup, tutorial, first purchase, and revenue bands at once runs out of room fast, and the finer you slice, the fewer installs land in each slot and the noisier it gets.
The platform also learns from this value. Design it badly and the algorithm optimises on a bad signal, and iOS performance drifts off entirely.
Three common mistakes
First, encoding behavior outside the window. The first window is days 0–2; if your average purchase lands on day 5, that signal never reaches it. Look at your actual conversion-delay distribution before choosing.
Second, remapping often. The moment you change it, data before and after mean different things and comparison breaks.
Third, leaving empty slots. Pull up the mapping and count volume per value — if more than half receive almost nothing, merge them.
Go deeper
The full SKAN structure and per-window resolution are covered in the iOS ATT and SKAN measurement guide.