What got built, what broke, what we threw away. With dates, not adjectives.
We publish this monthly for a simple reason: a system sold on "it moves every week" has to be able to show the weeks. Not pretty screens — dates and finished things.
The rule of the column: only what’s delivered and checkable goes in. And what went wrong sits on the same page as what went right, at the same type size. If you ever read a month here in which everything worked, it means we’ve started writing advertising instead of a journal.
On 8 August the system had a repo, a deploy and a backup — that was all. By 25 August it had all fourteen pieces of foundation finished, in order, each with its own tests.
Counted once, at the closing of the month, on commit 9c83fe67 (31 August 2026, 22:42), the last one on main before September. It’s a photo of the month, not a live counter: it doesn’t recompute on every visit, so it doesn’t repeat the defect that wrote the row above it, a "right now" figure on a page about a closed month. The production engines are four: post, carousel, explainer, graphics. There’s a fifth one in the code, a probe engine that produces nothing for a client; that’s why it says four here and five in the internal inventory.
if.Until 27 August, English was a promise on the website.
Now the interface language belongs to the account, not to the browser, and the content language is a folder of playbooks, not a branch in the code. The difference matters: an if language == "en" breaks at the second language. A folder doesn’t.
The last engine pulled out from under hard-coded strings was the explainer, on 28 August. Until then, a "multilingual" engine still had English sentences sewn into it.
Same rows, same type size as the deliveries. This is the section that makes the column credible — if it ever gets shortened "so it reads better", the column is dead.
Three things that worked and were taken out anyway. Two of them were selling.
Three things started, and one we’re not promising.
The interface for the three access tiers — the decisions exist, the screens don’t. The 129 untranslated sentences. And the missing link between what the system sees in the market and what it proposes we write: today, proposals are born only out of what we ourselves noted down.
What we’re not promising: case studies. We don’t yet have a client’s measured result on this machine, and a case study written before the number is fabricated evidence.
Every figure on this page comes from our own system and can be recounted. None of them is a client result — client results need clients running long enough for there to be something to measure, and we aren’t there yet. When we are, the number appears here with the name of the month it happened in.
Selection and rewriting, not documentation from scratch. That, at small scale, is exactly the argument: the raw material already exists inside what you do — what’s missing is the mechanism that turns it into something publishable. A client who keeps their Brain up to date gets the same piece, at the same cost.