The work behind the work

Hosting, DNS, TLS certificates, mail routing and deliverability, access control and monitoring — across a portfolio of live sites, running continuously rather than as a one-off project.

This is the least glamorous thing on the page and the most relevant to you. The reason your infrastructure is safe here is that this already runs every day, at a scale where the boring failures — an expired certificate, a mail path that silently stops delivering — have already happened and already been fixed properly.

Large-scale collection, joining and analysis: property and parcel records, market data, and public datasets turned into something you can actually query.

The engineering that matters here isn't the volume. It's the correctness — making sure a number that arrives late doesn't quietly get counted as though it arrived on time. That class of fault is invisible to everything that would normally catch it, which is why auditing for it is part of the work rather than an optional extra.

Training, evaluating and shipping custom models — including on our own hardware, rather than renting someone else's and hoping.

Two models are published openly and free for anyone to download and check. The model cards lead with their failure modes, because a model you can't characterise is a model you can't deploy. Locally, this rarely turns out to be the answer. When it is, it gets built here rather than resold to you.

An open protocol specification, published in full: schemas that resolve at their own public addresses, a conformance suite anyone can run, and two independent implementations held to it.

Two implementations interoperating is the only real test a specification can pass. It's the same instinct as testing a restore instead of trusting a backup report, applied to a different problem.