How we verify our numbers
What we check: that the AI repeated your data faithfully, not that your data is medically right.
Nucora’s chat answers and deep-dive reports are written by AI. When our app writes a number into a sentence, for example that your average sleep score last week was 82, we check that number before you see it. This page explains what that check does, what it does not do, and why we word our headline stat the careful way we do.
What we check
Every bolded number our app writes into a chat answer or a deep-dive report gets checked automatically, on every response, every time. Those two surfaces, chat and deep-dive, are where our AI puts your numbers into words. When the app writes a number into a sentence, our verification layer looks back at the exact data that answer was built from and confirms the number matches a value in that data. Every bolded number goes through the check, not a sample of them.
This covers the numbers our AI writes into sentences. It does not cover the tiles and charts on your dashboard. Those are a separate surface, and they are not part of the count on this page.
What "verified" means, and what it does not
“Verified” means one specific thing: the number you were shown matches the data the answer was built from. It is a check on the last step, the hop from your data to the words on your screen. It confirms the app repeated your data faithfully instead of inventing a figure.
It is not a claim that the underlying number is medically true. If a lab value or a device reading is off, and the app reports it faithfully, our checker calls that verified, because the app did its job of carrying the number through faithfully. We verify that the app is honest about your data. We do not certify that your data is right.
The headline, honestly stated
Our headline stat is “0 confirmed fabrications.” Here is exactly what stands behind it. Across more than 16,000 bolded numbers we have checked in chat and deep-dive answers as of July 2026, our checker flagged 43 of them for a closer look. Flagging is deliberately cautious, so most flags turn out to be fine. A person then reviewed every one of those 43 by hand. On review, none were confirmed as invented, which is why our headline reads “0 confirmed fabrications,” not “zero fabrications.”
We always show that number with its scale and its date, the way you see it above. A count of confirmed fabrications only means something next to how many numbers it came from and when we read it.
The honest limit
Here is the part we will not soften, because it is the whole reason we word things carefully.
Our checker can only catch a made-up number if it is good enough to notice it. We measure how good it is by continually attacking it ourselves, trying to slip fabricated numbers past it and fixing every gap we find. We test this way constantly, and we keep finding gaps and closing them. Because we keep finding them, we assume some still exist that we have not found yet. That is an honest read of any system like this, and it is why we say “0 confirmed fabrications” and never “zero hallucinations” or “100% accurate.” Those phrases would claim a kind of proof we do not have, and one we do not believe is achievable for any system like this.
Our review, not an independent audit
The reviews behind that zero are done by our team, not by an outside auditor. We are straightforward about that. To keep our own review honest, our reviewers follow one rule: when they are not sure whether a number was invented, they count it as a confirmed fabrication. Among the numbers we flagged, that rule can only push the confirmed count up, never down. The convention never flatters us. (This is only about the numbers the checker caught and flagged; the limit in the section above still stands for the ones it never noticed.)
A live, timestamped number
This figure is not fixed. It grows as people use the app, so we read it fresh, pin it to the month we read it, and tie it to the version of this methodology it was measured under. When the count or the method changes, the date and the version change with it. That way the number you see is always one you can place in time.
Methodology version 1, July 2026.
Nucora is a wellness and fitness tool, not a medical device. It is not intended to diagnose, treat, cure, prevent, mitigate, or monitor any disease or medical condition.
Nucora is in private beta. Request Beta Access.