One trafficking change per day
About 5,000 impressions and 20 clicks a day is too little signal if two things moved. Write the revert command before you apply the change. REVERT is a normal outcome.
Published 2026-09-09 · Search job: how to test campaign settings
How to test campaign settings on a small remnant flight: change one thing per day, write the revert command first, and decide on IVT and click-to-page-view, not on CTR.
This fleet delivers about 5,000 impressions and 20 clicks a day. That is far too little signal to attribute an outcome when two settings moved. If you change pacing and an exclude on the same day, tomorrow's readout cannot tell you which one did the work. The standing daily job is therefore: check campaign status, verify the app, ship fixes, and make exactly one trafficking improvement.
Revert is a written object, not a memory
No trafficking change ships without its revert command written into the experiments ledger first. A revert path that exists only in the head of whoever made the change is not reversible in practice. The next operator will not be that person.
REVERT is a normal, neutral outcome. The state to avoid is applied, never judged, which silently becomes permanent policy. If you cannot write the revert, you do not understand the change well enough to ship it.
Also write the metric and the window that will decide the experiment. Given about 20 clicks a day, a click-based metric needs many days. An IVT or delivery metric can move in one. Until IVT measurement is attached and populated, a click decline cannot be read as better or worse traffic. Fewer clicks may mean fewer bots. Any experiment decided on clicks alone must say so.
Health first
Run the loop health check before you touch trafficking. An experiment on a broken loop produces a result that means nothing. Two verdicts that are routinely misread as failures:
auto_paused_daily_capis healthy for a fixed-daily fleet. See Prepaid runway vs a paused portfolio.- A young campaign with no delivery is too early to tell, not broken.
Judge the change on IVT rate and click-to-page-view once measurement is on. Never on CTR. That is the same rule as How to traffic a campaign you can actually judge.
One change. One revert command. One metric. That is the whole method.
Write the experiment row before you apply anything. If the revert is a Beeswax PATCH that restores a bidding block, paste that block. If the revert is "point the campaign back at targeting set X," write the id. A prose note that says "undo it" is not a revert command. The ledger exists because this fleet is too small to recover signal after two overlapping edits, and because the person who ships the change will not be the person who has to take it back.