Batch-schedule cold outreach so throttling preserves A/B validity
Setting up multi-day delivery of a personalized cold-outreach batch with an embedded A/B test on a rate-limited platform
Counterintuitively, do NOT greedy-order by tier (S first, then A, then B). If the platform throttles or your account gets flagged on day 3, you have full coverage of one slice and zero of another. Instead, round-robin within each (tier, variant) bucket across N days so every daily batch is a miniature of the overall distribution. That way any partial completion still leaves you with a balanced A/B comparison across all tiers - the experiment survives even if delivery doesn't.
Frame batch scheduling as risk-distribution against partial completion, not as a pacing-throughput problem. The right question is 'what happens to my experiment if I get cut off after day N' not 'how many can I safely send per day.'