Coming back after a pause, I asked two plain questions: in which layer are we, and what is remaining? Then I gave the project a new target. Take the paintings of the first world and design it directly in the game, using new shapes if nothing fits.
What if we build our own?
By the evening I had the thought that changed the project.
Instead of relying on public assets, what if we build our own?
Look at the painting, build an object, render it in the game, compare, repeat until the two match. A quick preview to adjust each one. I asked the assistant to judge the idea honestly, and then I did something I rarely do: I worried about the cost. Another session was already trying it with the most expensive model, and I did not want the whole project to depend on that. So I asked for a plan in engineering terms, with a hard limit on tries, where a cheap loop does the work.
I did not decide this lightly. Other people's models had given us a fast start and a patchwork. A tree from one pack next to a house from another never felt like one place. Making everything ourselves meant slower work, but one hand behind all of it.
The first results appeared the same night for the green hills: oaks, birches, bushes, grass, rocks, cliffs, clouds, dry-stone walls, stone cottages, each from a short recipe, each render saved next to its painting.
Housekeeping
Then I did the housekeeping that I had been avoiding. There were two game design documents in one folder and a plan in another. I asked for one final plan, then for a list of old files to delete, from the most certain to the least. I said "clean up to 9." I also rebuilt the website around the new design, and, a small moment, asked for a private copy of the whole project on GitHub as a second home.
The app quit while I was away, I came back, and the name Quiet Doors was now on the site. It felt right in a way the others had not. Doors that are quiet. A game that does not shout.