Editing an atom restarts the evidence window from scratch

Spectare holds every variant's conversion rate as a confidence interval rather than a single number, and names a winner only when the interval clears the runner-up completely. The intervals are built to stay valid under constant watching, so refreshing the dashboard a hundred times cannot manufacture a winner. That has been true for a while. What is new is that the evidence is now windowed per edit: when you change an atom, the conversion count for every arrangement containing it resets, and the changed page re-proves itself from scratch before any verdict is issued.

This matters because a headline that performed well in March is not guaranteed to perform well after you rewrite it in June. The old count would have carried over, blending pre-edit and post-edit performance into a single rate that reflects neither version accurately. Now the evidence restarts cleanly, and your analytics show a clear not-yet rather than a stale winner.

The same discipline applies to rule suggestions. When the evidence supports a change to your targeting rules or an atom's audience assignment, Spectare shows you the with-and-without numbers and waits for your confirmation before applying anything. The accepted change then re-proves itself from scratch under the same windowed counting.

For organisations running a holdout, lift against the live control is also held to this standard. No lift number is quoted until the interval clears zero, and the holdout itself is opt-in per workspace. The result is a confidence page where every number shown is one you can defend: a statistically real winner, or an honest not-yet, nothing in between.