Why we waited.

There is a version of Wouch that existed in 2024, almost shipped, and was pulled back.

It worked. It would have worked commercially. It used the same engine architecture as the platform you are reading this on. It used the same research foundations. The product was good.

What was wrong with it was the moment after a user's first match. The screen said the right things - careful language, trauma-informed framing, the same restraint that lives in every screen of Wouch today. But the screen also nudged. A button slightly larger than the other. A subtle suggestion that the next thing was to send a message. A piece of language that, on the third revision, we noticed implied that not messaging would be the wrong move.

It was not a dark pattern in any conventional sense. A reviewer auditing the screen would not have flagged it. It satisfied every internal guideline we had written. And yet, looking at it on a Tuesday in late October 2024, two of us realised independently that what we had built was a quieter version of the same engagement engineering we had spent eighteen months explaining why we refused to build.

The reason it took two years from that Tuesday to today is what this essay is about.

The proximate problem was easy to see. The deep problem was that the proximate problem was not actually local to one screen. If a careful screen had drifted that far without being noticed, the substrate underneath the screen - the way we were thinking about success, the metrics that quietly governed what got shipped, the language we used in internal meetings - must have been drifting too.

So we stopped. Not paused. Stopped. We let the cohort of users we had at the time know what we were doing and why. We refunded subscriptions. We held the team. And we wrote down what we now call the Refusal Register: a small list of things we would never do, with the reasoning for each one, the growth argument we had heard for each one, and the conditions under which we would consider reversing each one. The reversal conditions are deliberately difficult.

The list got tested almost immediately. Several investors who had been close to the round looked at the Refusal Register and explained that some of the refusals would meaningfully cap the platform's growth ceiling. One refusal in particular - that we would never use behavioural observation to drive engagement, only ever to inform safety routing - was named as the deal-breaker for three potential leads.

The Refusal Register held.

What surprised us, two years on, is that the refusals were not the costly part. The costly part was rebuilding the substrate of the team's thinking so that the next quietly-drifting screen would be caught earlier. We hired a clinical governance lead. We instituted daily review of platform behaviour by a clinician, not by an engineer. We rewrote our voice guide with twelve rules of which we now keep eleven, having retired one that we thought was important and turned out to be cosmetic.

The platform you are now reading about is the platform that resulted from that two-year detour. The detour cost capital, momentum, and a meaningful percentage of the team. Several people who joined to build the first Wouch did not stay for the second. The cost was real and we are honest about it.

What we got in return is something that does not advertise well, which is part of why this essay exists. We got a platform that we can describe in plain language without flinching. We can tell you what we do and what we refuse to do, and the explanation does not require us to talk around anything.

A consumer technology product that touches the interior of human relationships should not be built in three months. It should not be built in six. It should not be built by people optimising primarily for capital efficiency. It should not be built by people who are themselves working through the same patterns they are designing around - at least, not without enough clinical scaffolding around them to catch what they cannot catch in themselves.

We waited two years. We are launching now not because the wait has expired but because the substrate is in place. The platform you are using is the version we can stand behind.