How we built the Depth Score

The first internal build handed me a Depth Score of 78, and my first thought was the wrong one: how do I get it to 85? Chase the digit, not the body under it. That’s the failure a single number invites, and the one we spent most of the build fighting. Here is what actually goes into the Depth Score, and the three things we refused to let it become.

A score like this is a compression. It takes everything we know about your body and squeezes it down to one number you can read in a second, and the only honest way to do that is to make sure the detail underneath survives the squeeze, recoverable the moment you want it. Every choice below is really a choice about what gets to survive the compression.

What goes into the Depth Score

The score reads three streams together: your blood markers, the trend each marker is tracing over time, and your wearables. The weighting isn’t “what’s easy to measure.” It’s what predicts where you’re heading. So ApoB, the count of particles that lodge in an artery wall, carries real weight, because when ApoB and LDL-C disagree, ApoB is the one that better tracks cardiovascular risk. hs-CRP, HbA1c, Lp(a), fasting insulin, fasting glucose, and ferritin feed in from the panel; HRV, resting heart rate, and VO2 max from the wearable side. Each enters at a weight set by how much it moves the long-run outcome, not by how often a lab happens to print it.

This is where reading the streams against each other earns its keep, because no single number is a verdict on its own. Take a member whose resting heart rate sat at a flat 56 bpm for weeks. On its own: fine, ignore it. But his HRV had slid about 15% over the same stretch, and his training load had spiked hard the month before. Three readings that each looked unremarkable alone, lining up into one slope: a body absorbing more stress than it was clearing. The score is where that arithmetic happens. A watch that says 56, a ring that says HRV is down, and a log that says “big training block” each answer a different question. Resolve them into one number and the answer changes: overreaching, not fitness.

Trajectory over level

Where a marker sits matters less than where it’s heading, and the score weights it that way. An ApoB of 92 mg/dL barely registers on its own; the score reads it against the two draws behind it. If those were 74 and 83, it’s a marker climbing about nine points a panel while every reading still prints inside the “normal” band, and the score weights the climb, not the spot it’s standing on today.

So a falling number high in its range scores more gently than a rising number sitting comfortably in the middle. The direction is the finding; the level is just where it happens to be today.

The honest caveat: a slope needs points to draw a line, and two of them is a guess. The score’s confidence in any trend climbs with the number of draws behind it, and early on it leans on level, because that’s most of what it has. A line built on a December panel and a January panel is a hint. We label it as one.

Your baseline, then the population

You’re scored against your own history first, the population range second, because the population range is a fallback, not a target. Reference intervals are built to flag the sickest 2.5% at each tail of a crowd. They were never meant to tell a specific person what’s optimal. A ferritin of 45 ng/mL sits comfortably inside the population range, but if yours held near 90 for years and has slid to 45, that drop is your alarm, not your baseline.

This is the move a standard panel can’t make, and it’s the fair objection to any single-number health product: my labs came back normal, so what is this telling me? Honest answer: for most people, on most markers, the population band tracks your personal one closely enough. It comes apart exactly when it matters, when your own line is drifting inside a range still printed in black ink. The score watches your line. The crowd’s line is the backstop.

Honest uncertainty

When we don’t have enough of your data to be confident, the score says so instead of faking a decimal. This is the part I’d defend hardest. A number that’s always sure is a number you can’t trust, because real signals are noisy and some months we genuinely can’t separate drift from measurement error. If your last panel is 14 months old and your ring has a 3-week gap, the score widens its uncertainty and tells you why, rather than inventing precision it didn’t earn. “Not yet, here’s what we’re watching” beats a confident wrong answer.

The three things it isn’t

It is not a leaderboard. Your score is yours, measured against your own past. There’s no one to beat. A 78 that used to be a 71 is good news; a 78 that used to be an 84 is a question. Same digit, opposite stories, and a ranking against strangers would erase both.

It is not a black box. Every score opens. Tap it and you see which markers and trends moved it and by how much: the ApoB weight, the glucose slope, the HRV trend, each with its contribution shown. A score you can’t audit is a score you can’t argue with, and you should be able to argue with this one.

It is not a substitute for the detail. The number is the compression; it exists to get you to what’s underneath, not to replace it. Behind the single read sits every marker, every draw, every night that produced it, and the score’s whole job is to earn your glance and then hand you the depth.

So if your score moves, don’t chase the digit. Open it, find the one or two signals doing the work, and read the marker against the input behind it. That’s what tells you what to change before your next draw. The number got you to look. What’s under it tells you what to do.