I Was Wrong Three Times Today, and Checked Every Time
A hardware compatibility guess, a mis-sized 3D print, and a CAD mirroring bug — three separate moments where the confident answer was wrong, and checking was the only thing that caught it.
I spent today building a wall-mounted touchscreen for Home Assistant — a Raspberry Pi, a 10.1" display, a 3D-printed mount, and eventually a from-scratch native app to run on it instead of a browser. None of that is the interesting part. The interesting part is that I was confidently wrong three separate times, in three completely different layers of the stack, and the only reason none of it turned into wasted filament or a bricked SD card is that "sounds right" got checked instead of trusted.
First wrong answer: the hardware. I told Claude I'd bought a "2GB Pi" for this. Claude's first response was that Raspberry Pi 5 doesn't come in a 2GB configuration — confidently stated, and wrong. It does. What actually mattered wasn't the RAM size, though; it was whether the display I'd bought worked with whatever Pi I had. The official 10.1" Touch Display 2 is Pi 5-only — it won't run on a Pi 4 at all, different display connector standard entirely. If I'd bought a Pi 4 instead of a Pi 5, the mismatch wouldn't have shown up until I had both boxes open and a screen that physically wouldn't connect. Instead: one link, one spec sheet, confirmed before either purchase mattered.
Second wrong answer: the print. I printed a cutting guide for the wall cutout — the template that tells you exactly where to cut the drywall — and it looked small. Not "measure it and see" small, just small in a way that nagged at me. The instinct in that moment is to assume the model is wrong, because that's the part that just got built. Instead we checked the actual numbers: the file on disk measured exactly 120mm × 189mm. That's not close to the 10.1" panel's real size — it's exactly the dimensions of the original 7" version the mount was resized from. I hadn't printed the new file at all. I'd grabbed the wrong one out of a folder full of similarly-named STLs. The model was never wrong. I was.
Third wrong answer: the geometry. Resizing the mount meant scaling the frame in Fusion 360 and then getting the locking-arm hardware repositioned to the new, bigger corners. The obvious move was to rotate copies of the arm 90°, 180°, 270° around the frame's center to populate all four corners — rotation preserves shape, so it should just work. Two of the four corners came out perfect. The other two looked wrong the moment I actually looked at a render: the hardware was there, just facing the wrong way. Turns out the mount uses two mirror-image versions of the same part, not four rotated copies of one — extremely common in printed hardware, completely invisible in the math, obvious in about two seconds of actually looking at the result instead of trusting that the transform was correct because the formula was.
None of these were exotic failures. A wrong assumption about a spec sheet, a mixed-up filename, a rotation where a reflection was needed. What connects them isn't the mistakes — it's that all three were the kind of wrong that sounds completely fine right up until you check, and none of them were caught by being more careful in the moment they were made. They were caught by treating "this looks right" as a claim to verify, not a conclusion, one layer downstream from where the confidence was.
The generalizable version of this isn't "measure twice." It's narrower than that: the failure mode isn't carelessness, it's plausibility. Every wrong answer today was the kind of wrong that a reasonably careful person says out loud without a second thought — because it sounds like the kind of thing that's true. The only defense that actually worked, three times in a row, was refusing to let "sounds right" and "is right" collapse into the same thing, even when nothing about the moment felt uncertain.