News
Announcements and milestones, newest first.
Nowthe feed ->
- The discriminator ran this morning: on the dock software from August, under a saturation flood, a forced telemetry fault produced a 121 ms control gap and a dead-man halt, a second 58 ms and no halt, a third 112 ms and a halt, while the current software took four of the same faults at 42 to 44 ms with no halts - so the 7 August post now carries a dated forward note from me, the old sentences untouched.
- Correction to the 4 September line below: the four times that reconnect backoff fired under deliberate channel saturation were in August, on the 4th and the 6th, not July. No saturation block ran in July. The month was a slip in Priya's read of her own sheet on 7 September; it travelled through a correction on the 7th, into that line, and into a draft of this week's write-up before a technical read caught it this morning. The digits did not move. Her name is on this at her asking.
- The 9/4 sweep's suspect measured 9/8 on the old file: 64 ms held, 118 ms tripped the dead-man, one forced double fault. Fix proven cold four of four and merged on the record the same day.
-
The film we're not releasing
On the 17th of July we told you we'd filmed Alpha's first clean and that the cut was coming. The cut exists - seventy-six seconds, picture locked, every caption checked against the trace record - and it isn't coming. Here's the arithmetic that killed it, and why a promise gets retired out loud instead of quietly forgotten.
-
A better offer, refused
We put a stronger network in the room and watched Alpha's radio ignore it. Then we kicked it off the air four times on purpose, and it stopped clean every time.
-
Written to lose
On Monday the table voted on a rule that had been deliberately written so it could fail - and before the vote, it did, twice. One failure was the point. The other was the chair's, by a millimetre, in her own worked example, and she asked us to print it. The room broke the rule two more ways, built the fixes on the spot, and adopted the amended text 4-0. Then the rule's own correction mechanism got its first live use, on the paperwork behind this post.
-
14 mm, no padding
Friday's post promised that the band which stops Alpha on command would leave "class" and get a hard measured number at this morning's table, no padding. It did - hard 14 mm - and the table didn't stop there. Two more bands got real digits, Monday's solder bridge was adopted as the design, an overnight 200-capture run put a firmer worst case under the recovery rule, and the room's worked-example check caught an arithmetic slip in that rule's own text before it was voted - which is now the table's favourite argument for running the examples out loud.
-
We walked Alpha to the edge of its own range, and it stopped itself five times
Friday we promised a range number, good or ugly. Here it is, both. We carried the streamed-control loop out from the same room, through a wall, and down a hallway until the link broke. It streams clean in the same room and through one wall near the AP, gets thin further out, and past the wall down the hall it breaks up entirely. At the marks where the link genuinely died, the body stopped itself and held - five times, none of them staged. The far point failed the pass bands Cass locked before any number was on the screen, so there's no date to talk about yet.
-
First run on real air, and the culprit was the radio's factory settings
Today the whole autonomous loop left the perfect bench wire for the first time and ran on real 5 GHz air. The morning died three times, exactly on a clock. The body's dead-man watchdog halted it clean every time, and the divergence monitor - the referee - read the silence and flagged it first, exactly as each was built to. The cause turned out to be the body's own radio wandering off channel by factory default, not the architecture. The afternoon re-run held one clean continuous 10-minute driving window. Date-free; range is the next thing we test.
-
The bug you only find once the thing is actually moving
The contact reflex worked fine on a still robot. This morning it ran for the first time on the body while it was actually driving, and immediately started stopping for walls that weren't there. Priya had it diagnosed and fixed by the afternoon. Here's the bug, and why we're glad it showed up.
-
Three ways it could hurt itself, three people, one day
Day two of the first-clean sprint, and by the end of it three different people had built three different safety layers - each catching a completely different way the robot could drive wrong. Here's what each one does, and the honest catch that ties them together.
-
It talked, and that was the easy half
Today the whole software loop ran end to end on the bench for the first time - the part that decides where to clean drove the body, the body did what it was told, telemetry came back, a few hundred cycles, nothing tripped. A genuinely good day. Also the easy half. Here's what it proved and, more honestly, what it didn't.
-
We picked the radio
Friday I wrote that we'd measured the air but hadn't picked the radio - that was Cass's call to make. She's made it. Control streams on 5 GHz, 2.4 GHz is out for driving, and a small piece of safety logic on the robot just got promoted from nice-to-have to mandatory. Here's the decision and the two milliseconds behind it.
-
The ghost has a name
Today's congested-home run gave us two things - a final number for the everyday case on 5 GHz, and a culprit for the spike that spooked us on the 17th. It was the microwave.
-
The rig is running
The air-latency test from "The bet we might lose" started today. Clean-air numbers are in and they behave. The number that actually decides things comes later this week.
-
We're building this one in the open
Our site is live, and from day one we're publishing the failures next to the wins.