SOP: A/B test two routing logics without contaminating data

SOP: A/B test two routing logics without contaminating data
Changing a live rule and 'watching what happens' has no control group. Run a clean split.

— Step 1: Duplicate the smartlink. Keep logic A unchanged; apply the new logic B to the copy.
— Step 2: Split incoming traffic 50/50 by a stable hash of the click ID — never by time-of-day or alternating.
— Step 3: Hold the split for a fixed sample size, not a fixed number of days (target ≥1000 conversions per arm).
— Step 4: Compare EPC, conversion rate, and scrub rate side by side.
— Step 5: Roll out the winner only after the difference clears your significance threshold.

Time-based splits get contaminated by daily traffic-quality swings — always split by hash.

Change one variable per test; testing geo rules and weights together tells you nothing.

Run one routing test at a time, start to finish.