Search

A worked example of what a taco looks like partway through a piece of work, rather than tidied up at the end.

What a recipe fork actually gives you: the slots, and nothing else.

The interesting fact about software buying decisions is not that people choose badly. It is what the utilisation data shows happens to carefully-chosen software

A fork of someone else's incident shape, continued in a direction they did not take: what happens when an incident outlives the shift that started it.

What two of the best-documented outages in public record show about carrying an incident: Cloudflare's 2019 regex outage, and GitHub's 43-second partition.

Measured figures for the Model Context Protocol ecosystem, taken directly from the official registry on 21 August 2026 — plus the counting trap that makes most

What is actually known — and, more usefully, what is not known — about making an agent's understanding survive the end of a session. Researched on 21 August 202

What the international reporting standard for systematic reviews demands, read as a specification for what a review has to have been carrying all along — agains

What the most-copied discovery method in software actually did — and why the number everyone quotes is the part that matters least.

A raise is the worst-case context problem in a startup's life, and the shape of that problem is measured rather than felt. From DocSend's seed data (2023) and P

There are two onboarding numbers, and the distance between them is what this shape is for.

How we chose what to aim at for our first ninety days after launch, and why the broad launch is deliberately not part of it.

Why we priced a product to be effectively free, what we metered instead, and the rule we adopted to stop ourselves quietly walling the one action the product ex