The hypotheses: what Prototype 1 is actually testing
It’s easy for a hardware project to drift into “building stuff and seeing what happens.” To keep this one honest, every planned test exists to confirm or kill a named hypothesis, with pass/fail thresholds decided before testing. A killed hypothesis is a successful experiment — it just points somewhere else.
H0, the one that matters
A cap-brim distance sensor with distance-to-pulse-rate haptic feedback can warn me of head-height obstacles while I look down at my feet, reliably enough — and with few enough false alarms — that I keep choosing to wear it on real hikes.
Note the two-part sting: reliability (misses are a safety problem) and false-alarm rate (the documented killer of every comparable product). H0 can’t be tested directly, so it decomposes into four clusters.
Cluster A — Sensing: “the sensor can see what matters”
Can it detect a large obstacle at 1.5 m indoors with stable readings? Can it see a 2 cm wooden dowel at 0.7 m — the thin-branch question no one has ever published data on? How badly does direct sun degrade it (expected: badly — the point is to quantify it)? Does the ultrasonic sensor cover the outdoor gap? Does the vibration motor’s buzzing shake the sensor into noisy readings? And does head-down hiking posture flood the system with ground echoes?
Cluster B — Feedback: “I can feel and understand it”
Can I identify which of four pulse rates is playing, without looking, at least 90% of the time? After 20 minutes of intermittent pulses at the temple, is it still tolerable — would I voluntarily keep it on? Walking slowly at a soft obstacle, do I stop in time on at least 9 of 10 approaches using vibration alone? (If the vocabulary turns out illegible, the fallback is fewer zones: better a crude vocabulary I trust than a rich one I misread.)
Cluster C — System: “it survives a walk”
Does the power bank run for 3+ hours without auto-sleeping? (Its low-current mode auto-exits after 2 hours — a real risk discovered in datasheet fine print, with a firmware heartbeat as the planned fix.) Does the cable ever snag? Do false alarms stay under 1 per minute in shaded outdoor terrain — a threshold set now, before testing, because moving the goalposts afterwards is how projects lie to themselves?
Cluster D — Adoption: the hypothesis everything else serves
After three outdoor sessions, do I put it on for a fourth without being prompted? Would I describe it as helping rather than nagging? If everything in A–C passes and this still fails, the concept fails on human factors — and understanding why would be more valuable than the hardware.
Kill criteria
The project only stops if basic sensing fails after debugging, or the haptics prove illegible even with fallbacks, or the adoption test fails despite everything else passing. Every other killed hypothesis is a pivot with a named next step — a different sensor, a different threshold, a narrower scope. That’s the plan, written down in advance, so future-me can’t quietly rewrite history.