9 May 2026 · Method

What a clean attribution window actually means

Desk with coffee, notebook, and laptop

“Clean” is the most abused word in campaign-to-install reporting. Vendors use it to mean deduplicated. Agencies use it to mean last-click. Finance uses it to mean the pie sums to one hundred. Vault Nodecore uses it to mean something narrower: a window whose rules are written down, whose unknowns stay unknown, and whose claim types are not mixed in one column.

A clean window starts with dates. Not “last week” in a Slack sense — a start timestamp, an end timestamp, and whether installs are dated by first open or by download. Then the lookbacks: click-through, view-through, re-engagement, reinstall. Then the identifier policy: handshake versus residue versus SKAN. If any of those lines is missing, the window is not clean. It may still be useful. It is not clean.

Unknown is part of cleanliness

Teams hide unknown installs because they look like wasted spend. They are often nothing of the sort. They are opens the matching rule could not attach. Dumping them into brand or organic to keep the chart tidy is how you teach a company to distrust measurement. A clean window shows the unknown slice. If that slice is large, you investigate identifiers and fraud. You do not relabel it for aesthetics.

What cleanliness is not

It is not a short lookback chosen to make search look efficient. It is not disabling view-through because video is politically unpopular. It is not forcing SKAN counts into the same CPI formula as Android click IDs. Those moves can be legitimate strategy decisions. They should be labelled as strategy, not as hygiene.

The intensive capstone is a clean window plus a one-page note a controller can annotate. Students fail the capstone when the graph is pretty and the assumptions are oral. They pass when the graph is slightly ugly and the assumptions are written. If you want a template before enquiring, the attribution lab page is the method in public form. Clean, in this house, is a reading standard — not a compliment you give your own dashboard.