Field Notes

Early Fathom builds had a depth reading in one corner and a progress bar across the top. Children ignored both. Adults watched the bar and stopped looking at the water, which was a remarkably efficient way to drain the sea from a deep-sea game.

Letting the depth gauge absorb the progress bar fixed it. One vertical scale now runs down the left edge, with the player as its marker. Zone boundaries are small ticks. They draw closer as you descend, which is wrong as physics and right as a feeling. Nobody in testing asked what it meant.

The problem left behind is the bottom of a zone. A bar that fills promises an end; a depth gauge cannot, because the sea carries on. We are testing a faint pressure line that appears about twenty seconds before the transition—enough warning to be felt, not enough to become another bar.

A day in Coach’s Diary gets opened at the door before a session, in the ten seconds after one, and again on Sunday evening. The obvious solution is one list that serves all three. We drew three different views of the same Tuesday instead and kept only what survived all three.

At the door, the urgent facts are who is next and what was said last time. Straight after the session, the screen has to take attendance, a one-line note and the reminder for next time, one-handed. On Sunday, the day becomes a week and the useful question changes to who needs a message before Monday.

What survived was the next-time line. Every version puts it first, above the time, above the client’s history, above everything else. That turned out to be the design. Everything else is arrangement.

We reviewed published measures of dietary diversity to see whether an ordinary food log could produce one honestly. Most were built for surveys: a trained interviewer, a fixed recall window and a population rather than a person. In that setting the scores are robust. Applied to one household over a fortnight, they become numbers wearing borrowed authority.

The recurring trouble is the denominator. Any diversity score needs a list of what counts as distinct, and every published list draws the boundary where its own study needs it. Three treat all leafy greens as one item. One separates tinned from fresh. Legumes seem to start an argument wherever they appear.

The working conclusion is to show no single score at all. Ikat displays the groups a week touched and the ones it missed, then lets the reader draw the line. It is less gratifying than a number and considerably harder to be wrong with.

We tried four ways to put a trace of noise over the site: a tiled PNG, a repeating radial gradient, pseudo-elements on each section and one fixed layer over the viewport. The fixed layer won every test that mattered.

The PNG was heaviest and alone in showing a seam on wide displays. Section by section, the noise moved with the content and read like texture printed on the page. The fixed layer stays still while everything passes behind it, so the texture seems to hang in the air. That is an unexpectedly large difference for something at three per cent opacity.

Opacity is the whole game. At five per cent, people notice an effect. At three, they notice a suspicion. It is also the first thing removed for reduced transparency: the texture is decorative; the cost of seeing it is not.

The display face loads as a variable font with swap, so the first paint borrows whatever serif the machine has to hand. On a slow connection, the change to the real face moved the opening line by eleven pixels and sent a small shudder through everything below it.

The fix is a second font face aimed at a local fallback, with its measurements overridden: size adjustment, then explicit ascent and descent. The borrowed face now occupies almost the same box as the real one, so the swap changes the letterforms without rearranging the page. Getting within a pixel took three rounds of measurement per face. It was worth doing once.

Two caveats. The overrides are tuned against the fallback we expect to be matched first, so a machine that has none of them will still shift a little. And the numbers are specific to these three families; they are not a general recipe and should be re-measured if a face is replaced.

The idea was simple: several people at the same event would join one session, share an exposure reading and know together when to step outside. It demoed beautifully in a kitchen with three phones lined up on a table.

Then everyone put a phone where phones actually go. A pocket, a table and a hand near a speaker produced readings so far apart that the shared number no longer measured anything. Averaging made matters worse: the result was stable, plausible and wrong, the most dangerous shape a health figure can take. Fixing placement would require hardware we do not have, or asking people to hold a phone just so for an entire evening.

The idea is shelved rather than buried. If a future watch exposes calibrated readings, everyone in the group will be wearing a sensor in roughly the same place and the feature may become honest. Until then, Muriel measures one person at a time, where she can be sure of herself.