The Smell of Molten Projects in the Morning
Ed Nisley's Blog: Shop notes, electronics, firmware, machinery, 3D printing, laser cuttery, and curiosities. Contents: 100% human thinking, 0% AI slop.
And kvetching, too
Spotted on a price scanning terminal during our grocery ride:
The process continued on all the scanner terminals as we collected our weekly supplies, crashed because a file was missing / locked / whatever, then became a steel-cage death match between whoever was running remote updates and Microsoft Windows:
Given the recent history of catastrophically bad Windows Updates, one might be inclined to consider a different OS for embedded applications, but …
Having verified the brake sensors work and with some idea of their actuation distances, installing them on Mary’s Tour Easy involved no more than:
The “proper location” put the actuation point about halfway between the brake lever’s released and pulled positions. Given that the previous sensors lacked indicators, I don’t know where their actuation point might have been, other than likely too close to their released position.
A slideshow of the front brake lever positions:
Similarly for the rear brake lever positions:
Obviously, the rear brake lever sensor has the “dim” LEDs.
There’s not much to go wrong and the first ride was uneventful, so we’ll declare victory until the sensors or magnets shift their position, despite the wraps holding their wires to the brake cables.
The magnets come from power toothbrush heads, encased in urethane adhesive in 3D printed mounts.
Stipulated: I sometimes over-do things.
However, I am absolutely not going to back down from saying the lashup shown in this screen grab from the sensor installation video is utterly and completely wrong:
Do not get me started, you know how I am. OK?
The NYS DOT engineer who designed the Raymond Avenue rebuild told me drivers should drive on the median to give cyclists enough clearance. This rarely occurs, because the “design” put bollards / posts / road furniture just behind the curb where they both deter driving and are easily hit.
The July 4th power outage called for refilling the gasoline supply and I’m returning home along Raymond Avenue:
That’s generous clearance, particularly from a hulking pickup truck. FWIW, commercial drivers generally do much better than average civilians.
If towing the trailer weren’t such a nuisance, I’d deploy a decorative gas can on every ride.
Not what you want to see when trying to extract a month’s worth of data from your blood pressure meter:
Perhaps the format string should be $1$s or %1%s, not a mixture?
It’s typical for Omron to issue two updates in quick succession after months of no activity, so programming by successive approximation may be a thing.
The Comment form you’ll see down below says:
Spam comments get trashed, so don’t bother. Comment moderation may cause a delay.
My apologies to real people, but this junk would appear without that little speedbump:
The IP addresses map to Singapore and China, for whatever that’s worth.
The game is on …
Just for fun, I handed the text of a recent post to one of the myriad AI detectors:
AI generation score: 12%
This text reads as highly authentic human writing, likely from a personal blog or DIY/home repair journal. The author demonstrates a genuine personal voice throughout, with specific technical details that feel lived-in rather than generic. Phrases like ‘sad bent wreckage on their stems’ and ‘yet another nonstandard standard size’ carry real personality and mild frustration that feels genuinely human. The writing has a casual, matter-of-fact tone typical of someone documenting their own project for an audience of fellow hobbyists.
Yeah, that’s about right.
Of course, an AI analyzed my text and generated the response, so I tossed its response back into the hopper:
AI generation score: 72%
This text has several hallmarks of AI-generated content, particularly in its analytical, evaluative framing. It reads like an AI detector’s own output — a meta-analysis of writing rather than original creative or personal content. The language is polished and structured in a way that feels templated: it moves systematically through observations (voice, tone, specific phrases, audience) in a formulaic progression typical of AI summarization tasks. Phrases like ‘feel lived-in rather than generic’ and ‘carry real personality and mild frustration’ are the kind of hedged, evaluative language AI systems use when describing human qualities they’re trying to identify or simulate.
However, we know real humans sometimes steal content from other authors, so I ran the same text through one of the myriad plagiarism detectors:
Guilty as charged: my text always exactly matches what you’ll find at softsolder.com. I have no idea where the “83% Match” for the first paragraph came from.
Remember: you read it here first …
I long ago learned to never trust Amazon’s unit pricing (or, for that matter, their recommendations), so this came as no surprise:
OK, you generally don’t buy hoses “by the ounce”, but “per fluid ounce” may not mean what you think it means:
Pricing items individually should be simple, if you know what a single item is:
Even knowing the number of items and the overall price isn’t enough for Amazon to get it right:
Amazon now has a “shopping assistant”, so I asked Alexa why the unit prices were incorrect. After some back-and-forth providing details Alexa should have known from the context, this seemingly plausible sequence of words emerged:
Amazon apparently stopped commingling knockoff crap with brand-name products under the same SKU earlier this year, a change driven by major brands refusing to have anything to do with Amazon’s “supply chain”, but the probability of my one-by-one reports producing any discernible improvement seems low.
The Smell of Molten Projects in the Morning 