What shipped in the last three weeks, and what I'm nervous about

Writing this down before tomorrow so I can be held to it.

The last three weeks were almost entirely the boring half. Making the same prompt land in the same place twice. Making the model picker say why it swapped instead of swapping silently. Making a Look load the whole recipe in one click rather than most of it. None of that demos well and all of it is the difference between a tool and a toy.

What I'm nervous about isn't the product, it's the first screen. Anything that does six jobs opens looking like a control panel. I've watched people scroll it, make nothing, and leave, while every metric I had said they were engaged. That's the failure I can't instrument my way out of.

Second, credits. I think credits are a bad pricing model and I'm shipping one anyway, because the underlying cost is real and variable. If a generation comes back wrong you already paid for it. I don't have a clean answer to that yet and I'd rather say so here than improvise it tomorrow.

Third, the picker. It's the thing I'm proudest of and the thing most likely to be misunderstood. If it picks well nobody notices, and if it picks badly once people stop trusting it entirely. Consistency matters more here than peak quality and I'm not certain I've got the balance right.

If you use Arteza tomorrow and the first screen loses you, that's the thing I most want to hear about.