The difference between a measured score and a population estimate — and why one app shows both, clearly labelled, never blended.
An app can show you one of two kinds of age number, and most products never say which kind you got.
A measured score is arithmetic on your inputs. Blood goes in, a deterministic formula runs, a number comes out. Same inputs, same number, every time. You can check it by hand.
A population estimate is what a statistical model predicts for someone with your characteristics. It is built by fitting a line through thousands of other people. When your data is incomplete, the model fills the gaps with cohort averages — and the number moves less, because more of it is about them and less is about you.
The DELTA age on Delta's Corpus screen is the second kind, and it says so. It comes from the "Healthspan Copilot" paper (PLOS One): an ordinary least squares regression over ten variables — bloods, plus waist circumference, weight, blood pressure and pulse — trained on a CDC dataset of 248 participants. The fitted coefficients were recovered by re-running the regression on the paper's own training data; the fit reproduces the paper's worked example to the published figure, which is the evidence that the recovered model is the paper's model.
Two properties of that model are worth knowing before you trust any number it produces.
Chronological age carries essentially zero weight. The regression learned to predict biological age from the biomarkers alone — a 30-year-old and a 60-year-old with identical markers get nearly identical estimates. That comes from the paper's dataset — it's not an implementation bug, and it is why an incomplete DELTA age barely moves with your data: with most inputs missing, the model returns almost the same number for everyone, because that number is mostly the cohort.
And some coefficients run against intuition as fitted to this dataset — higher hs-CRP and higher weight lower the estimate in this fit; higher HDL raises it. Those are the paper's numbers, faithfully implemented. They are also a reminder that a small-cohort OLS fit is an estimate, not physiology.
Because an estimate, honestly framed, still earns its place. Delta shows the DELTA age only with its provenance attached: an "estimate" eyebrow, never the word "measured", the model's version string, a confidence indication that falls as more of your inputs become cohort averages, and the paper's own disclaimer. When more than half the inputs are imputed, the app stops printing the number entirely — a figure that is mostly about the training set is not worth a glance, let alone a decision.
It coexists with, never blends into, the Adaptive Age — the deterministic blood-scored PhenoAge number that anchors the app. Different surface, different eyebrow, different uncertainty. The day you can't tell which number you're looking at is the day a product has failed you.
Every score in Delta is computed by pure, versioned arithmetic — never an on-device language model doing biomarker math. Every input is sourced or labelled as imputed. Every number carries its method. Estimates are framed as estimates; measurements keep their receipts.
And that, in the end, is the difference between a tool and a vibe.
