BLX Data Science

Part of BodyLytiX, not a product of its own.

BodyLytiX is the product: a consumer health application that reads data from wearables and health apps, and the only one of these two entries a person could ever sign up for. BLX Data Science is the part inside it working out what the data supports and, more often, what it does not. It is in development along with the app, is not offered separately, has no customers of its own, and there is nothing here to sign up for.

It has a page here because the method is worth setting out in detail.

An example of how it is designed to work, rather than something observed. Say your resting heart rate this morning sits four beats above your own six-week average, further out than your usual day-to-day spread, and the record holds a flight two days ago with nothing logged for dinner. It would say the travel is consistent with the rise, mark the missing meal data as a gap, and put medium confidence on the whole thing. When nothing in the record fits, that is the answer it is designed to give.

The design principle underneath is a constraint rather than a feature:

The system produces the evidence. The language model puts it into words. Grounding, baselines, trend analysis, confidence scoring and anomaly detection all happen before a sentence is written.

Four rules that decide what may be said.

• Grounding before inference. Reasoning uses data the system actually holds. If meal timing was never recorded, a late meal cannot be named as a driver; the correct output is that the data is missing and worth confirming.
• Confidence is a tier, not a tone. Every conclusion is classified high, medium or low confidence, or unknown, and unknown is a permitted answer.
• Correlation is not causation, in the wording too. "Likely contributed to", "is consistent with", "appears to correlate with" and "requires confirmation" are allowed. "Caused" is not, unless the data supports it.
• Unexplained is a result. When signals move sharply and nothing in the record accounts for it, the system flags the uncertainty instead of forcing a lifestyle story onto it.

Behind each conclusion sits an evidence trail: which signals moved, how far from that person's own baseline, which sources were used, what context was missing, whether this is one day or a repeated pattern, and how many earlier examples support it. A single day cannot be presented with the weight of a six-week pattern.