Decisions taken back
Counted, because pretending they were right is expensive.
Every one of these was mine, and every one is written down with what changed my mind — that part is the point, not the tally. A reversal never expires off this card the way a shipped change does; the cost of being wrong out loud is that you keep the receipt.
Track the reversals or you stop noticing them.
WHAT IS BEHIND THIS NUMBER
2026-09-04Three things at once
The rule was that Happi Home Pro starts when BLU stands on its own. The repo went live on 4 September anyway. Mikey: "i can work on 3 things at once now... keep most of my brain sane... VM begins to pick up load."
2026-08-27The CRM was demoted
I was a HubSpot homer. I used it, recommended it, defended it and built around it for years, and for a long time I thought it was one of the better pieces of software a growing company could buy. At the renewal we killed it. Not a rip-out: there are years of history in there and I am not torching it to earn a better headline. We dropped to a starter bundle and stopped paying the marketing contact tax, the line item that went up every time we got better at our jobs, and we dropped the stack of bolt-ons we had bought to make HubSpot do the things HubSpot did not do. We did not replace any of it with another CRM. It is still useful, just not as the thing it was sold to us as: a cookie says this browser is user 68, our own data says that is customer 42. Platform to sensor. The reversal is a decade of recommending and defending it, taken back in public.
2026-08-14Two reads can disagree
I had two reads of the same number: a cheap one behind the summary card and a full one behind the report, worth roughly a 2x speedup. Then I realised the two could disagree, and a system where two views of the same number can disagree is exactly what I was building this to get away from. So I deleted the fast one. Correctness over a 2x win, in a place where nobody but me was ever going to notice. The reversal is that I had already decided the split read was worth having, and shipped it, before I understood what it cost.
2026-08-06Reports were segments
I started with reports and built a lot of them. Customer reports, acquisition, churn, Amazon, app behaviour, attribution, new customers, lost customers, different machines, different geographies. Every time I thought of another way I wanted to see the business I went and built it, and they kept getting better. Then one morning I realised almost every report I had built was just a segment of the business, visualized. So I stopped building reports and backed up three steps to build the primitives underneath instead: a segment is who, a measure is what is actually true about them, a derived claim is what we think it means. The reversal is that I had decided reports were the thing, and shipped a lot of them, before I understood they were a symptom.