A single abnormal blood value rarely means only one thing. Depending on what else is true about you — your other lab values, your symptoms, your history — the same number can be the tail end of a minor infection, an early sign of something worth watching, or a lab artefact worth ignoring entirely. Reading it alone risks getting that call wrong in either direction.
One test, read alone, can mislead
This is the actual argument against treating any single test as a diagnosis by itself: the same number can be alarming in one context and unremarkable in another, and which one it is depends entirely on context the isolated reading doesn't contain.
A single number can be alarming in isolation and unremarkable in context — which one it is depends on what else is true about you.
A chain, not a single model
Ultrium's reasoning is built as a chain rather than one model producing one answer. Narrow specialist agents each read one kind of evidence — lab values, medications, symptoms, imaging — against reference ranges built for the people they're reading for. Those readings get combined by a layer that looks for agreement and contradiction across them, and a senior reasoning layer produces the final synthesis, with an emergency-triage check running ahead of everything else on every single question.
What happens when readings disagree
When two parts of that chain point in different directions, the system doesn't quietly average them into a confident-sounding middle answer. It flags the contradiction and lowers its own confidence rather than guessing — because a reasoning chain that hides its disagreements is more dangerous than one that shows them.
That's the practical meaning of “reasoning shown, not asserted”: every answer carries the chain that produced it, so if the evidence was thin, you can see that it was thin — instead of a single, unexplained number standing in for a diagnosis.
