Description: An article-based anomaly game logged seven starts and two finishes on an external device, while the mode recorded two visits and 41 page views. A small case study in reading engagement metrics properly.
A website turns ordinary articles into a spot-the-anomaly horror game. Headlines, images, and margins may shift; eight correct judgments lead to a final proofreading screen. Then the dashboard makes an even stranger appearance: two visits, 41 page views, seven game starts, two completions.
Maybe the first monster lives in the spreadsheet.
Why two visits and one device can both be right
An interim report retrieved on October 17, 2026, at 23:05 JST showed two visits, 41 page views and 62 qualifying ad impressions for the anomaly mode. A separate game counter, grouped by UTC day, reported one playing device, seven starts and two completions.
Cloudflare defines a visit in a way that permits multiple page views per visit. The game's device metric is an anonymous daily first-play count, not a stable personal identity. The two datasets also cannot establish that one device produced every single one of those 41 page views. Visits, devices, starts, and page views are simply different things.
Seven starts, two completions, eight judgments to escape
Each finish requires eight anomaly decisions. Two finishes are significantly more than merely clicking a game link. The report also logged one failure each at stages one, two and five. Someone encountered obstacles and played again.
Two divided by seven gives a 28.6% finish-to-start event ratio. It is not the percentage of unique players who won. Seven starts do not mean seven completed circuits; 41 page views divided by eight is not a valid way to infer five runs. The median and 75th-percentile playtime were both in buckets of two minutes or less, consistent with repeated short attempts rather than verified hours of play.
The week's impressive total had an unexpected contributor
Across seven days, the mode registered six visits, 190 page views, 37 starts and nine finishes. That is 31.7 page views per visit versus about 1.12 for ordinary reading. It sounds like something worth promoting immediately.
But a substantial part of those weekly starts may have been the operator testing the game themselves. The creator had become one of the most enthusiastic players. Funny, yes; conclusive evidence of broader audience enthusiasm, no.
There is one important distinction: the device counted playing on October 17 was confirmed not to be the operator's. The external seven starts and two completions remain a real, if very small, observation.
Can 41 page views become 2,050?
Multiply 20.5 views per visit by 100 visits and the arithmetic yields 2,050. The forecast is far shakier. The sample contains only two visits, and a game built around navigating pages will naturally produce more page loads than a conventional article. Sixty-two ad impressions are not 62 ad clicks or a revenue figure.
A sensible next experiment would separate internal tests, bring in new outside visits and track four steps: arrival to game start, start to completion, failure to retry and completion back to normal article reading. One hundred outside visits could be a practical first checkpoint, not a magic threshold.
The game has not gone viral. But someone outside the workshop has already finished two proofreading shifts.
Nobody even offered a salary.
Source and scope: Anonymous interim site report retrieved October 17, 2026, 23:05 JST (game figures grouped by UTC day); Cloudflare's visit and page-view definitions. Device counts are not verified human counts.
