Cloudflare request sampling: stage a rule test
The October 2 update enables request sampling. Here is how to prepare a staged rule test without confusing requests with an audience experiment.
On October 2, 2026, Cloudflare announced hash_in_range() availability across HTTP products on all plans. The function selects a portion of requests. Primary announcement and date.
Region: global. Audience: advertisers and publishers whose sites use Cloudflare. This is a third-party capability, not a new built-in AffiliateCore tool. This article is published October 5; the event occurred October 2.
What to consider
Editorial interpretation: a rule change can first be assessed on a limited share of eligible requests. That helps a team avoid changing treatment of the entire stream simultaneously. Sampling itself does not identify malicious requests or guarantee traffic quality.
Separate the share of requests from the share of people. One visitor can make many requests; this selection should not automatically be described as stable assignment of users to groups. For a complete experiment, define the assignment unit and the way outcomes are linked separately.
Prepare a limited trial
Write down the rule, eligible paths, owner and rollback procedure. Identify technical calls including postbacks, APIs, authentication and payment. Do not blindly apply browser challenges or blocking to these paths: first check whether the calling system can complete the proposed flow.
Set a small initial share and an observation window. Compare errors, latency and destination availability, then reconcile confirmed events separately. Events disappearing after a protection change may indicate a broken integration rather than reduced fraud.
Decide whether to expand
Increase coverage only after checking technical errors and reconciling events. Restore the previous rule if availability breaks, then investigate. Keep the time, selection conditions and verification result: otherwise a later before-and-after comparison will be ambiguous. These are editorial workflow recommendations, not a supplier promise of effectiveness.