The greenhouse that always settles into the right comfort setting
Before sunrise I walked the greenhouse corridor, past room after room linked by little vents. In each room, one helper kept cranking the heater up, and the other kept tugging it down. I wanted the whole place to end up in one of a few cosy, repeatable “comfort displays”, no matter how messy it started.
The vents were the trouble. Open them the wrong way and warmth sloshes about, rooms chase each other, and the whole greenhouse wobbles instead of settling. Connections can turn a calm plan into a feedback mess.
So I stopped trying to “copy” what the greenhouse had done before. I chose the comfort displays first, on purpose. I wrote down a few heater-up maps, while keeping the tug-down helpers the same each time, then set the vents so those chosen displays became natural resting places, not slippery ones.
The mapping stayed simple. Rooms were the parts of a network. The two helpers were the push and pull inside each part. Vents were the connections, and only the heater-up side sent influence out through them. The comfort displays were the labels. Takeaway: sorting becomes “which stable resting display does the whole place fall into?”
A comfort display is useless if a tiny draft can knock it loose. So I checked the vent strengths and kept only the ranges where small bumps die out. While tuning the vents, I stayed inside that safe zone, so the greenhouse couldn’t accidentally learn a habit of wobbling.
Each test was routine. Start with a messy spread of temperatures, let the rooms nudge each other step by step, and see which comfort display it drifts towards. If it drifts to the wrong one, nudge the vent knobs so next time it flows to the right one, still staying calm. You can even run it backwards from a settled display and rebuild a rough version of the start.
Later, the same idea was used to sort real things, like greyscale pictures where each pixel acts like its own little room, and it can work better if you first squeeze the picture into a cleaner starting layout. The point wasn’t “it wins at everything”. It was seeing the invisible part clearly: you can design the settling places, then learn the paths to them, without hoping the whole thing behaves.