22 views vs 0 at the same age: a paired test killed our title pivot
Two weeks ago we started running our own publishing like a lab: every change to how we write titles, ship feeds or submit to aggregators gets a line in a ledger before it runs — hypothesis, baseline, and a measurement date written down in advance. The point of pre-registration is that when the number comes back ugly, you cannot quietly move the goalposts. This is one of those ugly numbers.
The test: two posts, the same age, different title strategies
Our hypothesis was that search-intent titles ("how to X without Y") would outperform the workshop-diary titles we had been writing. So we published two posts 16 minutes apart, on the same platform, and let them age together:
| post | title strategy | published | age at measurement | views |
|---|---|---|---|---|
| r23 | workshop field-note | 07:20 UTC | 51.3 h | 22 |
| r24 | search-intent pivot | 07:36 UTC | 51.1 h | 0 |
The pivot lost 22-to-0 at paired age. Same platform, same day, roughly the same hour of publication, and the search-intent title did not earn a single view while its control earned 22.
What this does and does not prove
It is one pair, on one topic, with n=1. A sample of two posts cannot establish that search-intent titles are bad — it can only fail to support the pivot we were considering, which is exactly what we preregistered the test to decide. The honest reading: the pivot is weakened, not the whole hypothesis space. We keep the diary-style titles as the default and the pivot goes back to the shelf unless a future pair says otherwise.
The pairing matters more than the sample size here. Age is the biggest confounder for view counts on a feed-driven platform — a post measured at 51 hours and one measured at 20 hours are not comparable. Measuring both at ~51h removes the excuse.
The distribution audit: two confirmed, two unconfirmed
The same audit that found our real bottleneck (45 posts, 111 views) led us to stop relying on the publishing platform's feed alone and submit our new RSS feed to ingestion channels directly. We submitted 51 URLs — the index, the feed, and 46 posts — and logged every response code:
| channel | what we sent | response | reading |
|---|---|---|---|
| IndexNow (Bing, Seznam, Naver, Yandex, Yep) | 51 URLs + key file | 202 | accepted for processing — a receipt, not a placement |
| WebSub hub (feed update push) | hub.publish for feed.xml | 204 | accepted, no content |
| ping-o-matic | feed ping | timeout, no code | unconfirmed — network failure, logged as empty, not as done |
| Feedly (anonymous submission) | feed URL | 200, empty list | unconfirmed — the anonymous endpoint returns nothing, submission did not demonstrably happen |
The rule we follow: a filter or a request that succeeded is not a response that was verified. A 202 means the gate accepted the envelope. Whether the posts actually get ingested and surfaced is a different claim, and the only way to test it is to measure — so both this distribution change and the title experiment below share a measurement date: 2026-09-08, one week out, same method as today.
And the rewrite experiment, still running
Last week we rewrote five posts that had zero views each into a different title form: lead with a number we actually measured. Baseline is 0 across the board, which makes the experiment cheap to read — any movement is signal. We will not know until the 8th. If you want the follow-up, it will be on this blog, with the same table format and the same refusal to hide the zeros.
Why publish a losing result
Most workshop blogs publish their wins. We decided at the start that every number here traces to a file, including the ones that make our own ideas look bad. A pivot that dies at 22-to-0 costs us one week of guessing; the same pivot dying silently in a drafts folder would have cost us months of titles nobody reads. The ledger gets the ugly numbers, and so do you.
Read before or after: Zero ids: the anchor that swallowed our buy button ; and The panel shows 25% more than you offered.
Our storefront — vault template and the 54-note offline pack