Skip to content

Writing

How this data actually behaves

We buy proxies and run scrapers; we do not sell either. That is an unusual position in this category and it is the only reason these posts are worth reading — we can publish the vendor that did not respond, the endpoint of ours that is rate-limited, and the approach we tried and abandoned.

Built for machines

5 posts

Change detection, schema drift, silent failures, and the billing consequences of each — the problems that appear once something else is consuming your pipeline on a schedule.

Legal and risk

5 posts

What the courts actually held, as opposed to the single sentence everyone quotes from each case — including the parts that went against the scraper. None of it is legal advice, and every post says so on its face.

The undocumented web

5 posts

What each platform hands a logged-out request, measured rather than assumed, with the date and the network position recorded next to every result.

What scraping actually costs

4 posts

Written from the buying side. We have no proxy to sell, which is why these can name the vendor whose endpoint did not respond and the tier that cannot run the workload it is sold for.

Ad transparency data

2 posts

The one corner of this category that regulators made public on purpose, which changes both what you can build on it and how much risk comes with doing so.

What you will not find here

No “how to scrape X with Y” tutorials. The largest competitor in this category published 2,146 of them and they return roughly three visits a month between them, so the format is not a gap in our coverage — it is a decision.

No post is structured around us, and none of the comparisons scores itself. Where a competitor does something better, the post says which thing and why.

Where the measurements come from

Every probe result carries the date it was taken and the network position it was taken from. That second half matters more than it sounds: a request that succeeds from a home connection tells you very little about the same request from a datacenter range, which is where production runs.