App stats
Automated traffic is excluded from every figure on this page. Browser automation reports a fixed device id and is counted, metered and then left out of anything that claims to be a person. Counting began on August 18, 2026. Where a figure was never measured it reads no data rather than nought.
All time
By day
The numbers behind the chart
| Day | Questions | Requests | Devices | Accounts | New accounts |
|---|---|---|---|---|---|
| Aug 22 | 0 | 0 | 0 | 0 | 0 |
| Aug 23 | 0 | 0 | 0 | 0 | 0 |
| Aug 24 | 0 | 0 | 0 | 0 | 0 |
| Aug 25 | 0 | 0 | 0 | 0 | 0 |
| Aug 26 | 0 | 0 | 0 | 0 | 0 |
| Aug 27 | 0 | 0 | 0 | 0 | 0 |
| Aug 28 | 0 | 0 | 0 | 0 | 0 |
Today
What the reviews decided
- Pass0
- Modify0
- Hold1
One decision per turn that ran a review. This is the number that says whether the product is finding anything: a service that only ever passes is either checking very safe orders or not checking.
When it could not check
Every review so far ran to a decision. Nothing was turned away for want of a label or a patient fact.
Rising refusals mean the corpus or the rule set is too narrow for what is being asked. Only clinical refusals are counted: a turn stopped before any checking — the daily ceiling, a busy process, an unpaid subscription — is not one of these, because a busy afternoon must not make the corpus look narrower than it is.
How long an answer took
Measured from the moment the request arrived to the moment the answer was ready, so it is the wait a pharmacist actually had. Failed turns and uploads are excluded from the timings: a failure’s duration is how long the failure took, and an upload is not an answer.
Most-asked medications
- metformin1
The resolved ingredient, never what anyone typed. A line the resolver could not place against the published labelling is counted here as nothing at all rather than under its own text — which is what keeps a prescription out of a statistic.
Where from
- United States71
- Not reported16
- Egypt3
- Australia1
- Kenya1
These come from the metering counter, which has been running longer than the figures above and counts every device that made a request rather than every device that asked a question. Its total is therefore the larger of the two, and neither is wrong.
Devices are not people. A phone and a laptop count twice, a private window counts again, and clearing site data starts over. Anonymous device counts read high; account counts do not.
An account here is one that has signed in and used the app. An account that registered and has never asked anything is not in this figure — the accounts table is scoped per tenant by the database, and reading across every tenant to count rows would need a privilege this page has no business holding.
Country comes from the browser’s own time zone, not its IP address — no address is stored or sent anywhere, which also means a traveller or a VPN is counted where their clock says they are.
Nothing here is per-person or per-request. What is stored is a salted hash of the device id, a salted hash of the account, a date, a duration, a decision and a resolved ingredient name. No raw identifier, no address, no timestamp finer than the day, and nothing about a patient.