2026-05-06

The opening distribution plan for a side project I’m shipping was three cold pitches, one per week over 21 days, to three independent newsletter operators. The three operators served the same audience.

Product context: a curated TV-style guide to Mandarin video and China live cams, aimed at English-speaking learners and watchers. The plan tested cold-pitch fit through three small operators in the Mandarin-learning space, then would have decided based on replies and downstream traffic whether to scale.

Expected reply count for three cold pitches with no warm intro is roughly 0.15 to 0.45, mode zero, even when the message is well-crafted and the targets are well-chosen. That’s the noise floor. A 21-day decision window built on top of it produces nothing on most rolls.

The noise floor wasn’t the worst of it. All three operators were learner-resource operators. Their subscribers overlap heavily; a learner who follows one almost certainly follows the others. Three channels, one audience. If every operator replied, I’d learn that learner-resource operators amplify, and learn nothing about whether the product also fits China-watcher readers, travel/expat readers, ambient-curation curators, or topical-community moderators. If every operator went silent, I’d have no way to tell whether the message was wrong or the audience was wrong.

The replacement plan covers six cohorts (Mandarin learning, China-watchers, travel/expat, ambient/curator, niche-topical, passive distribution) with about 25 sends in week one, in parallel. Silence in any single cohort is interpretable: that audience didn’t engage. Replies concentrated in one cohort are interpretable: that’s the wedge. The unit of the test is the cohort, not the operator.

The artifact that holds the new plan together is a public sortable index of about 50 Mandarin YouTube channels graded by listening difficulty, with methodology and a CSV download. Every cohort gets a pitch built around the index instead of around the product. A learner-resource operator can republish the table, a China-watcher Substack can quote the methodology, a moderator can pin it. The pitch leads with the gift; the product is mentioned in parentheses.

Channel count is not signal count when the channels share an audience. Three independent operators drawing from one subscriber pool produce one test with sample size three, not three tests with sample size one each. The diversity that matters is in the population behind the channels, not in the channels themselves.