The device we rebuilt after watching it used once

What a single observation revealed that no amount of specification had.

A single observation of real use revealed a flaw that months of written specification had missed because a specification only ever describes what a device is supposed to do, while watching someone actually use it shows what a hand, in practice, actually does, and those two things had quietly stopped matching without anyone noticing until someone stood there and watched.

Why watching beats asking

A specification is written from memory and intention, a list of functions the device must perform and the conditions under which it must perform them, assembled by people who already know exactly how the device is meant to be held and operated. That prior knowledge is precisely what makes a specification blind to certain failures, since the person writing it cannot un-know how the device works well enough to notice where a first-time hand might reasonably do something else entirely. Watching a real attempt at using the device removes that advantage completely, because the person being watched has no prior knowledge to fall back on, only the shape in front of them and whatever their hand does with it in the moment, which is exactly the information a specification, however thorough, structurally cannot supply.

Asking someone how they would use a device suffers from a related but distinct problem, since a person asked to describe their own future behaviour tends to describe the behaviour they believe is correct rather than the behaviour their hand will actually settle into once thinking about it stops being the point. A person asked in advance whether they would grip a torch near its focus ring will very often say no, confident in the answer, and then grip it there anyway the moment the actual task of walking down a dark path takes over their attention. Watching sidesteps this entirely by never asking the question out loud at all, it simply records what happens.

The torch-focus comparison

A torch with a twist-to-focus lens passed every functional test on the bench, the beam sharpening and spreading correctly whenever the head was rotated, and the specification describing that mechanism was entirely accurate. Watching one person actually use the torch outdoors in the dark showed something the specification had no way of capturing, that their hand naturally gripped the torch exactly where the focus ring sat, so every small shift of their fingers nudged the beam gradually out of focus without them ever meaning to touch it. Nothing in the bench testing could have surfaced this, since bench testing checks whether a mechanism works when someone deliberately operates it, not where an untrained hand happens to rest when it is not thinking about the mechanism at all.

The person doing the testing on the bench had, without realising it, been holding the torch carefully, consciously keeping their fingers clear of the ring because they already knew what it did. That single habit, invisible to the tester because it felt like nothing more than ordinary care, was itself the thing standing between the specification and the actual problem, and it only stopped being invisible once someone unfamiliar with the mechanism picked the torch up and gripped it the way an unthinking hand naturally would.

The one number worth remembering

A specification review involving several careful readers had gone over that torch's design for weeks without anyone flagging the grip position, while a single fifteen-minute observation of one real attempt at using it in realistic conditions caught the problem immediately, within the first minute of the beam visibly drifting. The disparity is not a reflection on how carefully the specification had been checked, it reflects that a written description and a lived attempt are answering genuinely different questions, and no amount of additional scrutiny applied to the wrong question closes that gap. Weeks of careful reading had, in effect, been checking whether the sentences describing the torch were internally consistent, a real question worth answering, but not the same question as whether a hand holding the finished object in the dark would behave the way those sentences assumed.

What this changes in practice

Building in at least one round of watching a real, unrehearsed attempt at using a device, ideally before the design is considered finished rather than after, catches an entire category of problem that no specification review, however rigorous, is actually positioned to catch. The habit worth keeping from this is treating a clean specification review as necessary but never sufficient on its own, since it answers whether the device does what it claims to do and says nothing at all about whether a real hand, working it for the first time, ends up doing what the specification assumed. Fixing the torch afterwards was simple once the actual problem was visible, moving the focus ring to a section of the barrel a resting hand never touched and adding a slightly raised band the fingers could feel and avoid without looking, neither change more than an afternoon's work, both entirely invisible to the specification that had been signed off weeks earlier.

Where this stops being true

A single observation is not a substitute for repeated ones, since one person's particular habits, gripping with the left hand rather than the right, wearing gloves, working left-handed against a right-handed assumption baked into the shape, can just as easily mislead a designer into fixing a problem that was never general in the first place. The value of watching comes from doing it more than once, ideally with more than one person if the device is not being built for a single named user, and treating the first observation as a hypothesis worth checking rather than as a verdict to redesign the whole device around immediately. For a device built for exactly one named person, this caveat matters less than it first appears to, since there is only one hand's habits that actually need to be understood, and repeating the observation with that same person on a different day serves mainly to confirm the first grip was not simply a fluke of that particular moment.

My key error with this

We built the adaptive golf kit against a careful specification, with the reach, the grip and the swing arc all worked out with the user's measurements in front of us, and by every check we knew how to run it was correct when it left the bench. Watching it used once was enough to see that we had designed for the swing and forgotten the setup, because getting into position and addressing the ball involved a sequence of movements nobody had asked about, and the support that made the swing possible made that sequence considerably harder. None of it appeared in any of our requirements, since we had only ever asked about the part of the activity that looked like engineering. We rebuilt it around what we had watched rather than what we had specified. What replaced the belief is that a specification records the questions somebody thought to ask, that watching the thing used answers the questions nobody thought of, and that one honest observation is worth a great deal more than another round of careful requirements written in the same room by the same people.