Last time I said everything rests on one question: can the dock carry the load I've put on it. So let's open it up. There's a version of that question that ends the company, and I'd rather you watch me find out than read about it later.
Here's the bet in plain words.
We want the cheapest possible robot body. Not cheap as in nasty - cheap as in there's almost nothing on it to pay for. Good sensors, precise motors, a small chip to keep it safe, and that's the lot. The thinking - where to go next, and how - lives on the dock. So the dock has to drive the robot. It works out the next move and streams it to Alpha, move by move, over your home WiFi. The same WiFi your phone is on.
If that link holds - if the dock can send "do this now" and Alpha hears it in time, every time - then the body stays dumb and cheap and the whole plan from the last post is intact. The robot gets smarter through the dock, not through new hardware. That's the thing I sold the car for.
If the link doesn't hold, I've got a problem. The time-critical control has to move back onto the robot, because you can't have it standing in your kitchen waiting on a packet that's stuck behind someone's video call. And the moment that thinking moves onto the body, the body needs a real chip, real cooling, real power to feed it. It gets smarter and it gets dearer, and the cost gap that's the entire reason we exist starts closing. That's how we lose.
I want to be exact about losing, because we were. Before we spent a dollar, we wrote the condition down: the control loop has to run over WiFi reliably enough that Alpha never stutters or stalls waiting on the dock. If we can't hit that, the all-on-dock idea is wrong and we have to know early, while early is still cheap.
So here's the honest part, and it's not the part I'd put on a poster.
We haven't measured it yet.
The plan for this week was to build the rig that measures the link - the thing that tells us three things at once: how late the dock's instructions arrive on average, how much that lateness wobbles from one move to the next, and whether the link ever goes quiet long enough to leave Alpha waiting with nothing. All three matter, and you can pass on one and still be sunk by another. It didn't happen. The week went on getting Alpha driving and getting the connection between the two of them clean, and both of those had to come first, and both took longer than the calendar I'd drawn. So the rig slipped to next week. That's our first slip. I'm logging it here because I told you I'd show the working either way, and the working this week includes a thing we said we'd do and didn't.
I'm not going to hand you jitter numbers I don't have, or a verdict I haven't earned. The measurement is next. Whatever it says, it goes up here.
What did land is worth saying, because it's the first time the thing moved under its own command. Alpha rough-drove on the bench. The dock sends it a velocity command, Alpha takes it and goes, and streams its telemetry back so the dock can see what happened. It's tethered to a bench and it lurches and it's not pretty, and it drove, off instructions coming in over the link. That's the shape of the whole bet, working badly, which is the right order for it to work in.
And one thing we built on purpose, so the WiFi question can never turn into a safety question: Alpha's reflexes don't go through the dock at all. The "stop, that's a stair" decision runs on Alpha's own little chip, on the robot, with no network in the loop. So whatever the WiFi does - drops, lags, dies - the robot still won't drive off a step, and if the link goes quiet mid-move it stops where it is rather than carrying on at the last speed it was told. It asks the dock about anything clever. It never has to ask anyone whether the next move is a fall.
Next week we measure the link. I genuinely don't know what we'll find. Come watch.