Right now
The fleet.
Ten small computers, Raspberry Pi 5 boards with Intel AX210 radios, sit in the FIIT library. One transmits a steady stream of Wi-Fi frames; nine listen to what the room did to those frames on the way over, the way nine microphones would hear one metronome. This is what each of them is doing this minute.
Reading the fleet…
What the numbers are
An estimate is one measurement of the channel: a complex number per Wi-Fi subcarrier, saying how the room attenuated and delayed that slice of spectrum on its way across. A listener is handed one every time it decodes an OFDM preamble, so the rate in that column is a sampling rate, and the room is the thing being sampled.
Resolves up to is that sampling rate turned into a speed: v = λ·PRR/4. A body moving at radial speed v shifts the reflected path by 2v/λ, and sampling at PRR resolves shifts up to PRR/2 — set the two equal and the quarter wavelength falls out. It is a ceiling on what the rate makes askable, not a detection limit that has been measured: whether the empirical breakpoint sits at the bound is the open question in EXP-007.
Why two capture columns
Process says csid is up. Estimates says channel readings are actually arriving. They are separate readings and they disagree: on 2026-08-15 five nodes reported a running process and one was delivering data. One combined column would have claimed five capturing nodes.
Why the spread matters
The fleet is passively cooled and the nodes do not sit in one thermal environment, so several degrees separate the coolest from the hottest at any moment. That is not housekeeping: the radio's own phase reference drifts with die temperature, so a thermal gradient across the fleet is a systematic term in every phase-based measurement taken across it.