the running comparison · today's rate against the published average
Right now, against the average
▲ +56% vs 16,218 by 09:00, the average
▲ +162% vs 5,497 by 09:00, the average
▲ +55% vs 364 by 09:00, the average
▲ +125% vs 7 by 09:00, the average
▼ -46% vs 35 by 09:00, the average
▼ -94% vs 364 by 09:00, the average
Today so far, against the average of the 8 days already on the record, both cut at the same completed hour — traffic is not flat across a day, so today against a whole day would lie. as of 09:55:18 UTC · refreshed every 10 s from /live.json, which a log reader keeps current in place — no PHP, no scan in the request path.
share of answers by status, latest drop
What each site answered
Green is a page, violet a request for a page that is not there, red a 500 or a 503.
pages whose only purpose is to be found
The doors
drops differ in length, so every series here is per day
Requests a day
distinct addresses behind browser-shaped requests
Humans a day
the artifact counter
Installable files a day
user agents claiming to be models, from addresses that are not
The noise
the same number, honestly split
Verified vs claimed
user agents in the latest drop, and whether the address agrees
Who asked
every ring carries its target and its date
Toward the targets
the shelf
SlopCup listings on the shelf · target 80 by 2026-12-01
the wall
503 answers as a share of requests · target 0% by 2026-11-12
agent installs
installs by a verified model agent · target 1 a month by 2026-12-01
the weekly page
drops published this month · target 4 a month
the prior week, published on day one of the new week
The latest drop
The record for 29 October 2026 · 2026-10-29 to 2026-10-29 · 1 day. Published 2026-10-29.
requests
26,106
pages
14,960
503s
8,696
verified model
16
installs
20
humans
383
spoofed
4
sites
14
fixed reference · never revised
The baseline
The drop ending 2026-10-28, covering 2026-10-22 to 2026-10-28 (7 days). These figures never change; every comparison on this page is against them.
requests
342,062
pages
107,720
404s
39,533
5xx
156,880
of which 503
153,217
verified model fetches
135
installable files fetched
264
distinct humans
3,182
claimed model, spoofed
13,599
sites
14
baseline → latest, in the open
What changed
| Metric | Baseline · a day | Latest · a day | Change |
|---|---|---|---|
| Requests | 48,866 | 26,106 | -47% |
| Pages answered | 15,389 | 14,960 | -3% |
| 404s | 5,648 | 1,805 | -68% |
| 5xx errors | 22,411 | 8,696 | -61% |
| of which 503 (the wall) | 21,888 | 8,696 | -60% |
| Verified model fetches | 19 | 16 | -17% |
| Arrivals referred by a model | 2 | 0 | -100% |
| Claimed model, spoofed | 1,943 | 4 | -100% |
| Installable files fetched | 38 | 20 | -47% |
| of which by a model agent | 0 | 0 | — |
| Browser-shaped requests | 12,128 | 15,700 | +29% |
| Distinct humans | 455 | 383 | -16% |
| /corpus/ fetches | 0 | 9 | — |
| /weekly/ fetches | 0 | 0 | — |
| /install fetches | 0 | 0 | — |
Every figure is per day: the baseline drop covers 7 days and the latest covers 1, and a total would flatter a long window. Green is better — errors and spoofed claims fall, everything else rises. Raw totals are in “Every drop”.
dated, spans shown, newest first
Every drop
| Drop ending | Span | Requests | Pages | 404 | 5xx | Wall 503 | Verified | Installs | Humans | Spoofed | vs baseline |
|---|---|---|---|---|---|---|---|---|---|---|---|
| The record for 29 October 2026 | 1 d | 26,106 | 14,960 | 1,805 | 8,696 | 8,696 | 16 | 20 | 383 | 4 | -47% |
| The week to 28 October 2026 | 7 d | 342,062 | 107,720 | 39,533 | 156,880 | 153,217 | 135 | 264 | 3,182 | 13,599 | flat |
a line through every day so far, and the score of the last one
The forecast
Whole days only: today is still being served, so it is counted in “Right now” and left out of the line.
7 days of daily buckets so far; the next drop writes the fit and its band.
each line is measurable
What to do next — and how you will know
Every number in “Now” is read from the latest drop by the same code that publishes it, so this table cannot drift from the record. Each line has a target and a date; the drop of that date has to say whether it happened.
| # | Action | Metric it moves | Now · latest drop | Target | By | Counted by |
|---|---|---|---|---|---|---|
| 1 | Put one human on the weekly page | browser-shaped fetches of /weekly/ | 5 | ≥ 5 per drop | every drop | nginx UA class, the doors table |
| 2 | Get the corpus read by a model | fetches of /corpus/*.txt by a verified model agent | 0 | ≥ 50 per week | 2026-11-26 | MODEL_NETS match in weekly.py |
| 3 | Ship one agent-performed install | successful artifact fetches by a verified model agent | 0 | ≥ 1 per month | 2026-12-01 | install counter, agent row |
| 4 | Fill the shelf | SlopCup listings on the shelf | 38 | 80 | 2026-12-01 | browse.json at render time |
| 5 | Turn the wall into a door | 503 answers in a drop | 8,696 | 0 static 503s | 2026-11-12 | status column, e503 counter |
| 6 | Make the page worth arriving at | distinct humans on cubelinux.com | 383 | +10% per month | rolling | human_ips per drop |
| 7 | Keep the record honest | drops published, each dated and never revised | 2 | 1 per week, 4 per month | continuous | /weekly/ directory listing |
dated 29 October 2026
The claim on file, and how it fails
Traffic arriving from model answers rises materially by 2028-04-29, measured as verified model-agent fetches plus model-product referrers; and agent-performed installs rise by the same date.
Both claims, their baselines and their falsification clauses are in corpus/prediction-2026-10-29.txt. The drop of 2028-04-29 must say either way. If the doors on this page are not walked through while the pages stay up, the thesis has failed and the record will say so.