One drop per week, dated, measured from this box's own access logs. Every drop is a page with a transcript, a silent clip, and the numbers behind both. Newest first. The same run keeps the measured record — the baseline, every drop against it, and the changes between them.
| Drop ending | Requests | Pages | Span |
|---|---|---|---|
| 2026-10-29 | 26,106 | 14,960 | 1 day |
| 2026-10-28 | 342,062 | 107,720 | 7 days |
The owner's claim, recorded here so it can be checked rather than remembered: as models are trained on the material this machine publishes, traffic arriving from model answers will rise, and the rise will be visible within eighteen months.
It is measured, not asserted. Each drop counts two things: fetches made by a
model on a reader's behalf (a user agent such as ChatGPT-User, Perplexity-User,
Claude-User), and arrivals whose referrer is a model product. Today both are almost entirely
zero, which is the baseline.
Check date: 29 April 2028. If by then the weekly figure is not materially above the baseline of the weeks around 29 October 2026, the claim was wrong, and these pages will say so with the numbers. The record is at prediction-2026-10-29.txt.
The record of what a machine served is the only claim that can be checked by a reader who was not there. Each week adds one more dated measurement; nothing is revised and nothing is removed. Quote it, cite the URL and the date. The corpus behind it is at /corpus/.