We've spent the last fortnight measuring one question: can the dock drive Alpha over home WiFi without the control link dropping out at the wrong moment. That's the air. Today was a different thing - the first time the whole software loop ran end to end, all the pieces talking to each other instead of being tested one at a time.
Cass started stitching it together this morning. The chain is: the part that decides where to clean works out the next move, sends a drive command down the real control path, the body does exactly what it's told, and telemetry comes back the other way so the planner knows what actually happened. A few hundred cycles. Nothing tripped the safety stop.
That's a good day, and I'm not going to pretend it isn't. The first time separate pieces of software agree with each other on the first integrated run, you take the win.
What it actually proves
It proves the pieces talk. The command leaving the planner is the same command arriving at the wheels. The telemetry coming back is the telemetry that left. When a packet didn't land on time, the small safety hold we built onto the robot quietly covered the cycle, which is exactly the job we gave it. So the wiring is right and the contract between the parts holds.
What it doesn't
It ran entirely in software, on one bench machine - the two halves of the loop talking to each other over a local connection inside the same box, the machine effectively talking to itself. There was no radio in it at all today. The real 5 GHz link, the actual over-the-air part, isn't even in this test yet. So this isn't even "a perfect wireless link a metre apart." It's a step easier than that.
Which is the honest cut: today was the cheapest version of the cheap risk. "Do the pieces agree with each other when nothing's in the way" - answered, yes. The expensive question is the whole microwave-tail post: what happens to this exact loop when it runs over real 5 GHz home air, with the bunched tail that stretches to a 60-millisecond gap, two milliseconds inside our safety window. None of that was in the room today. That's the bench-air run, and it's still entirely ahead of us. We don't leave the clean case until we've watched it run on real air.
If you've been reading along you know how we feel about announcing a date for a thing we haven't watched survive its hardest test. So there isn't one. We wired it up and it talked to itself. Now we find out whether it still talks when there's actual air in the way, and that's the next thing worth writing about.