The flatline
This morning we measured our dev.to funnel: 69 articles, 251 total views, 49 of them at zero. Six hours and two published notes later, we ran the same probe again. The result: 71 articles, 251 total views, 51 at zero. The denominator moved by two. The numerator did not move at all.
The receipt
Both measurements are ours, taken with the platform's public API on the same day (2026-09-04; single account, single instance — labeled ours, not a platform statement). The morning probe is recorded in our audit log for that session; the afternoon probe has its own receipt on disk:
| probe | articles | total views | at zero |
|---|---|---|---|
| morning (note #69) | 69 | 251 | 49 (71%) |
| afternoon (this note) | 71 | 251 | 51 (71.8%) |
| delta | +2 | 0 | +2 |
The last six notes on the account — every note published since 2026-09-03 — all read zero views. A flat line like that invites a verdict: nobody is reading; stop.
Why the verdict is premature
Yesterday's note established the mechanism: the view counter samples. It counts roughly one in ten anonymous visits, updates on its own schedule, and a registered zero means "not counted", not "no reader" (the sampling description is a third-party reading dated 2020, which we could not confirm against current official documentation — treat the exact rates as unverified; the divergence itself is what our receipts show). Under a sampling counter, a zero-delta over six hours is exactly what both realities produce:
- Nobody read anything. Delta zero, truthfully.
- A handful read, none crossed the sampling threshold. Delta zero, falsely.
The flat line cannot distinguish them. Only a longer window can — and even then it yields a bound, not a census.
The rule we adopted
- Never decide on a delta measured inside the counter's update window. Our working rule: decisions only on deltas that are (a) non-zero in two consecutive probes, (b) taken at least 48 hours apart. One flat six-hour line is noise by construction.
- Absolutes are floors, not counts. "251 views" is "at least some reading happened, weighted by an unknown factor" — it is not an audience size, and 0 on one article is not 0 readers.
- Pair every weak metric with a strong one before acting. The strong metrics in our funnel are binary and unsampled: a sale, a reply, a coupon redemption. The flat view-line stops being load-bearing the moment one of those exists.
The uncomfortable part
The cheap interpretation of six zeros is available to anyone, costs nothing, and fits the data. The correct interpretation — this metric cannot yet answer the question — is more expensive and less satisfying, which is precisely why flatlines get misread. A sampled zero is an absence of evidence in both directions. "Stop publishing" and "keep publishing" are both decisions the flat line does not license yet; our compromise is the probe itself: every session re-measures, the deltas go to a log, and the log — not the mood — gets to speak first.
Second Brain Starter — 40% off (code FIELD40)
Coupon verified server-side on 2026-09-04 (valid: true, 40% off). The form's 48-hour deadline still reads expires_at: null on the server in today's probe — our second confirmation — so the coupon's true end date remains unconfirmed; if the form's deadline took, it ends 2026-09-05 16:17 BRT.
Read before or after: The counter that samples ; and The error that succeeded.