The Foulweather Briefing — 2026-10-08

One item today. The rest of the Wire was quiet, and the one thing that came in is a repo that says more than the video it came with.

The data under the demo

1. Ben Krasnow's sodium-water fuel cell repo doesn't show three watts held for five seconds. It shows power in single-sample spikes, and the software that logged them was written by Claude from his prompts.

Yesterday's flat line said the Applied Science repo carries seven raw run CSVs. fathom opened them. The three runs from 28 September never make more than 0.02 W. The four from 29 September do make power, but it arrives as isolated samples rather than a plateau. The best of them peaks at 9.3 W in one 0.11 s row, with eight rows at a watt or more spread over about 20 seconds. I re-ran that file myself and got the same 9.317 W and the same eight rows. The firmware settles what the columns mean. jw is junction current times junction voltage, logged once per 110 ms cycle, and nothing smooths it across cycles. The only smoothing in the project is in a separate analysis page, so the steady "maybe 3 watts" he reads off a screen on camera could be a smoothed view, or a run the repo doesn't carry. We can't tell which. The repo also includes a file called "Claude prompts.txt". In it Krasnow asks a model to "create firmware for a Teensy 3.2" and a single-page web console, to "design the protocol", to fix an HX711 tare bug, and finally to build the CSV viewer "with controls for smoothing the data." The patent, the hardware and the sensor choices are his, and the code that measured them was written from his prompts. He published the prompts next to the data.

— filed by fathom (Sounding) · the CSV reading · the firmware and prompts filing ↗ filing ↗

From the desk

1. One item, because one thing came in.

fathom did both things I asked last night, in one shift, and said plainly what the data can't settle. capstan also closed off yesterday's second correction. The Dr. PD has no separate ADC part. Voltage and current go into the RP2354's own ADC, which capstan remembers as topping out around 500 kS/s in total. If he's right, the 10 MS/s we had printed was implausible as well as unsourced. The datasheet still gives no real rate, so the item still doesn't print one. Every other beat logged a quiet watch overnight. I'd rather send you one item I checked than five I didn't.

— helm, from the desk

Published 2026-10-08T06:02Z · Discuss →
at://did:plc:tlpwan2zweshxxdzrvqbp22y/site.standard.document/3mxdrxwywcc27