What 10,000 preview requests taught me
1 September 2026
The generators re-render a preview every time you change a number, and every render leaves a line in a log. I sat down with eleven days of those lines, 551 visitors and 9,109 previews, a number that will have crossed ten thousand by the time this post is indexed. I expected traffic statistics. I got a lesson in how people actually treat a parametric tool, and some of it stung.
The ratio that reframed everything
For every file someone generates, there are 27 previews. For every download, about 44. My first reaction was disappointment, all that rendering and only 205 downloads to show for it.
Then I looked at the biggest sessions. One visitor previewed a single vase 234 times. Another spent 217 previews on a rugged box. Nobody nudges a slider 234 times by accident; that is someone sculpting, watching the belly of a vase swell and shrink until it looks right. The preview is not a step on the way to the product. For a lot of visitors the preview is the product, a CAD session with all the CAD removed, and plenty of those sessions legitimately end with "not today" instead of a file. A parts bin does not need reshaping 234 times. A vase apparently does.
That reframing had a practical price tag, by the way. Each preview costs real server time, which is exactly why the running total feeds the cost math I use for the site. The 234-preview vase session cost me about three cents. Cheapest CAD seat on the market.
Where the funnel actually leaks
Of the 551 visitor-days, 27 percent ended in at least one generated file. Of those, roughly six in ten downloaded the ZIP. That second gap bothered me more than the first, because someone who clicked Generate already decided the model was right, and then something between the click and the file changed their mind.
The logs point at two culprits. Generating takes seconds where previews take milliseconds, and a few sessions simply end during that wait. And some visitors generate twice or three times in a row, 1.64 generates per download on average, which smells like people spotting something in the final render that the 2D preview did not show them. The 3D preview tab exists exactly for this, and the numbers say I should make it louder.
Errors are telemetry with feelings
The most useful single discovery was not a ratio but a burst. One night a visitor collected fifty validation errors on the hinge box in 25 minutes, the same error, over and over: lid taller than the base. The sliders I had added made it effortless to drag the lid height past the base height, and the backend dutifully refused every time. Fifty refusals is not a user problem. It is a slider that permits what the geometry forbids, and it took one evening in the logs to find and one clamp to fix.
Same story with the enclosure vents, which a four-word bug report exposed as never having worked at all. The pattern holds: my errors arrive in logs, other people's errors arrive as patience running out.
Small print
Peak hour is mid-afternoon, the quietest is five in the morning, and the gap is only about twelve to one, which means someone somewhere is resizing a drawer organizer at almost any hour. A quarter of the traffic comes from the US, another quarter from Poland and Germany combined. And the top of the popularity table is relentlessly practical, boxes, enclosures, vases, drawer organizers, while the decorative generators trail far behind. People do not come here to browse. They come with a measurement.
The next thing the numbers ask for is obvious enough that I have already started: making the 3D view part of the default flow instead of a tab you have to find. If the 44-to-1 crowd is sculpting, the least I can do is hand them better clay.