mobieusHealth: food data quality
mobieusHealth keeps the numbers members log honest with a lightweight, AI-assisted quality loop: a full Nutrition Facts label members can read and portion, a one-time sanity check on each food, member controls to confirm or report a food, and a tenant-admin console to review what members flag. This page covers all of them. For the tracker as a whole, see mobieusHealth.
This is a wellness tracker, not medical advice.
The Nutrition Facts label
From the food logger, View Nutrition Label opens the familiar FDA-style panel for the picked food: serving size, calories, the full macro breakdown, and the label micronutrients (saturated and trans fat, cholesterol, added sugars, vitamin D, calcium, iron, potassium) with % Daily Values. Members set a portion by servings, grams, ounces, or a volume like cups, and every number rescales. Two deliberate behaviors: a value the food simply doesn't report is left blank rather than shown as zero (so "0 g" always means measured zero), and a volume poured for a food with no known density is flagged approximate.
Every surface (the web app, the iOS app, and the Android app) draws these numbers from one shared calculation on the server, so the label reads the same everywhere and no client does its own math.
The one-time AI check
The first time anyone pulls up a given food to log it, mobieusHealth runs an AI sanity check: does the serving size look typical, and are the calories and macros plausible and roughly internally consistent? Members see the verdict as a small indicator when they set their portion: AI-checked (looks reasonable), serving may be off (atypical serving), or data flagged (numbers look implausible and are worth a second look).
The check runs once per food and the result is cached, so it is instant afterward and never re-spends. It is advisory. It flags what deserves a closer look; it never rewrites the food's numbers on its own, and it never mutates the shared food database seed.
Confirming and reporting a food
On the label, a member is asked "Is this information correct?"
- Yes records a quick confirmation, a signal the data looks right.
- No opens a short editor where the member can suggest corrections: type the right calories, protein, carbs, fat, fiber, or sodium, or just add a note and flag it. Suggested corrections travel with the report so an admin sees exactly what the member believes the numbers should be.
The older Report incorrect data control does the same thing and still works. Reporting also:
- Re-checks the food with AI immediately,
- Escalates the food to the community's admins for review, and
- Takes the entry out of the lookup's first rung, so the next search for that food goes back out to the other data sources looking for a better answer. If one has it, that answer replaces the flagged one and clears the flag for everybody.
The third is what makes a report worth making even when the member has already worked around the bad entry. They are not filing a note; they are sending the search looking.
A food a member flagged carries an "under review" marker while it is in the queue. To keep it fair, members can have only one open report per food, and there is a daily cap on how many foods one member can flag.
Where a food's numbers come from
Lookup asks in a fixed order and stops at the first source that knows the food: the community's own records (everything a member saved or corrected, plus everything looked up before), then the branded and restaurant catalogs, then the USDA analyses, then the supplement labels, and an AI estimate only when nothing above has it.
Two consequences worth knowing. An answer found further down is kept, so the next member to search that food gets it immediately and no community pays for the same lookup twice. And a row whose numbers were estimated rather than measured says so on the result, because a laboratory analysis and a model's guess are both useful and are not the same kind of fact.
The apps also keep foods a member has already looked up on the device, so searching for one again works without a connection and says that is what it is showing rather than passing an old list off as current.
The admin reports console
Community admins review flagged foods at /admin/health/reports (Admin → mobieusHealth → the food reports console). The queue lists each reported food with the member's note, any suggested corrections, and the outcome of the immediate AI re-check, and can be filtered by status (new, rechecking, resolved, rejected).
For each report an admin can resolve it (data fixed or accepted), reject it (no change), or accept the correction, which promotes the member's proposed values into the shared, cross-tenant food-reference database so the fix improves the food for every community, not just this one. (Accepting into the shared database uses a dedicated write credential the platform provisions for each tenant; where it isn't yet configured, the correction stays recorded on the report and can be applied once it is.) The console requires a super-admin (role 5), and every resolve/reject/accept action is written to the community's audit log.
Related
