Early on we built the display first, because a display is the obvious thing to build. Then someone pointed out that during the work the device is meant to support, the user is looking at a tooth. The screen is in their hand and out of their attention.
That observation is most of why the device reports three ways instead of one.
Three channels, three kinds of attention
The channels are not redundancy. They are for different situations, and they carry different amounts of information.
The screen is the high-bandwidth channel. It shows the current angle and where it sits relative to the limit — a continuous quantity, readable at a glance. It is also the only channel that can tell you how far and which way. It requires you to look, which means it is available during setup and between steps, and mostly unavailable during the work itself.
The LED is peripheral. It carries almost no information — a colour, a state — but it delivers it without a glance. Peripheral vision is poor at detail and good at change, which is exactly the right match: you cannot read a number out of the corner of your eye, but you notice a colour shift immediately.
The tone requires no vision at all. It is the only channel that works when your eyes are fully committed. It carries the least information of the three and is the most intrusive, which is a deliberate pairing: the channel that interrupts hardest should also be the one that says the least.
Colour before number
The display does not just print a value. It maps the deviation into zones — comfortably inside the limit, approaching it, past it — and colours accordingly, with the zone boundaries derived from whatever limit the user selected rather than fixed in advance.
This is because reading a number requires a comparison. Seeing "7°" means nothing until you remember your limit was 10 and work out that you are close. Colour skips that: the state is the appearance. A user who never reads the number still gets the information.
The number stays for the cases where the magnitude matters — during setup, or when someone wants to know how much they actually drifted rather than whether they drifted.
When to speak, which was the hard part
Deciding how to alert took an afternoon. Deciding when took much longer, and it is the part that matters.
An alert that fires too eagerly does not merely annoy. It trains the user to ignore that channel, and once a channel is ignored it is worse than absent — it is a channel you believe is protecting you while it is not. Alarm fatigue is a well-documented failure mode in clinical environments, and a device that adds another beep to a room has to earn it.
Three things follow from that.
The threshold belongs to the user. We cannot know whether someone is rehearsing a movement or working, and those want very different sensitivities. Fixing the threshold ourselves would mean being wrong for most people most of the time.
The tone is reserved for crossing the line. The screen and the LED can report continuously because they are ambient. The audible channel is not ambient — it takes attention away from something else — so it stays quiet until the state actually changes.
Silence has to mean something. If the device is quiet, that must be information, not just the absence of an alert. This is why the LED holds a steady colour rather than switching off: quiet-and-fine looks different from quiet-and-not-running.
What we tried and dropped
A continuous tone that rose in pitch with deviation. Technically elegant, genuinely informative, and unbearable within about ninety seconds. Continuous audio in a room where someone is concentrating is a much bigger imposition than it seems while you are testing it alone at a desk.
More colours. An earlier version used five zones. Nobody could tell the middle three apart at a glance, which defeated the point of using colour in the first place. Three is what peripheral vision can actually discriminate.
Flashing to indicate severity. Motion in peripheral vision is extremely attention-grabbing — too much so. It pulled the eye away from the work, which is precisely the opposite of what the LED is for.
The principle underneath
Match the channel to the attention the user has available, not to the information you would like to deliver.
The temptation with a device that computes a rich, continuous quantity is to present all of it, because it is there. But the user's attention is the scarce resource, not the data. The screen gets the detail because looking at it is a decision. The LED gets a state because peripheral vision handles states. The tone gets one event because interrupting someone is expensive and should buy something worth the cost.