thomas kim! That’s a good question.

Yeah, I’m pretty aligned with this hot take.

In the real world, the “line” almost never shows up as a clean ±X mm number — because the body doesn’t fail gracefully in millimeters. It fails in contexts. You can be off by 1–2 mm and be totally fine for an hour, then be off by 10+ mm the moment the band rotates, the user sweats, or they change posture. That’s why paper specs feel so disconnected from lived use.

Based on my review and analysis of various sources, the approximate threshold appears to be:

For short, constrained interactions (VR controllers, rehab exercises, guided sessions):
You can get away with tighter bounds. A few millimeters of relative consistency over a short window is enough, because you’re constantly re-anchoring the model. Absolute accuracy still isn’t real, but drift doesn’t have time to compound.

For all-day wear (watches, rings, bands):
Millimeters stop mattering. If your signal is stable enough to detect change over time — trend, deviation, asymmetry — you’re winning. Being “wrong” in the same way all day is often more useful than being occasionally “right.” This is where ±5 mm vs ±1 mm is a meaningless debate, because skin shift alone exceeds both.

For behavior classification (walking, typing, slouching, fatigue):
UX truth beats geometric truth every time. If the user says “yeah, that’s when I started slouching,” your sensor is accurate enough. If the model needs constant recalibration or fails when clothing changes, it’s not. The threshold here is: does the output stay interpretable without explaining the sensor to the user?

So yeah, “good enough as long as the UX feels right” isn’t hand-wavy — it’s actually the most honest metric we have. The hard part is that it’s not a single number you can put on a slide. It’s a system-level property: attachment, time-on-body, algorithm design, and expectation setting all at once.

Chasing millimeter-level accuracy on a deforming, sweaty, self-repositioning surface is basically pretending humans are CNC fixtures. The teams that ship stop asking “how accurate is the position?” and start asking “when does this break trust?” That’s the real line.

This is a great breakdown. One thing I’m still wondering though — how do you communicate this internally? Like, PMs and stakeholders always ask for a number.

If “±X mm” isn’t the right answer, what do you actually give them without sounding hand-wavy?

Yeah, that’s the hard part.

What’s worked for me is reframing accuracy as reliability over time instead of spatial precision.

I’ll say things like: “We’re confident detecting changes within a user across a day,” or “This signal stays stable across posture and strap movement.”

Then I back it up with failure cases, not best-case demos. Once people see how quickly millimeters fall apart on-body, they usually stop insisting on a single number and start asking the right question — “when does the user stop trusting this?”