Four days, one solder bridge. Monday it went on the bench and the fail-safe stop changed character - the path that used to coast now brakes. Wednesday the table adopted it, which means the machine that coasts is no longer the machine we ship. Thursday, the machine wearing that bridge ran its first soak block: twenty-four passes, six hours and fifteen minutes, every band as re-minted at Wednesday's table.
Most of what the block measured comes down to one number the new band was minted without.
The unknown, named in writing
Wednesday's post carried the new fail-safe band and its gap in the same sentence: hard 16 mm, worst of 24 warm captures, deep-sag case not yet measured. Monday's bench run couldn't get there - a fresh pack won't sag on demand - so instead of padding the number or pretending the question didn't exist, the band note named it: deep sag gets measured as the last item of the next soak block, and if it moves the braked class, the band goes back to a table.
Thursday's block was built around that sentence. The deep-sag captures were scheduled for the tail of the run, so the battery would get low the way it actually gets low: by working. The pack started sagging on its own around pass 21 - no fake load holding it down - and the strapped stops ran last, after pass 24, with the pack deeper still: 15 to 16 millimetres, worst 16. Same class as warm, same spread, same direction. The braked path does not care what the battery is doing.
So this morning the band note did the thing it was written to do: the caveat expired by its own condition, with a dated entry in the registry citing the run record. Nobody scheduled a meeting. The note said what would settle the question, the question got settled, and 16 stands as minted. The fail-safe stop is now defended cold, warm, and deep-sag - all three measured, none of them assumed.
Two days is a short life for a limit, though, and there's a real difference between the two days. Wednesday the 16 existed as 24 bench captures. Thursday it sat through heat, dwell, a full duty cycle, and a battery running itself flat - and read the same. That's what soak hours are for.
The other new band had a livelier first day
Wednesday also gave the position-error band a hard edge - 4 cm, replacing an eyeball "few cm" that had sat in the ledger since mid-July. On pass 17 of Thursday's block, that new edge tripped: end-of-area position error read 5 cm. Stop, stamp, forty minutes of finding out why.
The cause turned out to be felt. The left tyre had been picking fluff off the edge of the rug offcut and winding it into a layer about half a millimetre proud
- which makes one side of the machine roll on a slightly bigger wheel. Bigger
wheel, more distance per revolution, heading creeps a little every lane, and by end of area the error has walked long. Same family as the surface finding from mid-July: the floor keeps contributing to the machine's geometry whether you invited it or not.
Priya cleaned the tyres, ran a dock out-and-back that came home in family, and resumed the block. Passes 18 through 24 read 2 to 3 cm.
Now the part I keep chewing on: 5 cm is "a few cm". Under the wording that band carried until Wednesday morning, this trip doesn't exist. The felt keeps winding, the error keeps walking, and whenever it finally got noticed there'd be no stamp, no pass number, no forty-minute cause on file - just a vague sense that the machine had gone a bit crooked lately. The hard edge, one day old, converted all of that into a measured cause in under an hour.
And the band isn't changing. The number was right; the world found a new way past it. What changes instead is small and physical: a thirty-second tyre wipe joins the every-four-passes stop, and the design notes gained a line that tyre material has to consider fluff pickup, because real homes have rugs.
The rest of the sheet, plainly
Commanded stops read 12 to 14 mm all block, including under sag - at the 14 edge, which is exactly where a band minted from the measured worst should sit. Reflex latency 41 to 45 ms at both ends of the block, against the 70 budget. The $2.60 motor driver that got its 200-capture characterisation on Monday night plateaued at 65 C with 20 degrees of headroom. Worst link gap 57 ms, once; the trip lamp stayed at zero, which brings the running count to thirteen-plus soak hours without a single unattributed halt. Pack worst 12.9 V under load, nowhere near cutoff.
The new habit rode along too: an on-dock out-and-back before the first pass and after the last, as a same-day check of calibration drift. It came back 0.3%, inside the 0.4% repeat class - the first point on that chart, and it cost nothing but two short drives.
One trend gets its own paragraph, because trends are the other half of the rulebook. Whisker trigger force is up 8% cumulative across the three blocks - +5% after block two, +3% more after block three, the same direction each time. The blocks aren't equal lengths (12, 16, and 24 passes), so there's no tidy per-block rate; the fact is the direction, not the slope. Roots clean, function nominal, and no hard limit exists for it because nobody can defend a threshold yet, so it sits on the watch list instead. If it's still climbing in the same direction at block five, a spare arm comes off the shelf for a pull test.
The ledger, with the blocks side by side: 3 h 05, then 4 h 10, then 6 h 15. About thirteen and a half hours, 52 passes, three blocks.
The shape of the week
The sequence, start to finish: a band was minted Wednesday off 24 bench captures, with its one unknown written into its own note rather than papered over. The unknown was measured Thursday, on schedule, with the battery low for real. The number stood.
Sixteen was minted with the question attached. A day later the question has an answer, and it's sixteen.