Home › How we check
How the content is checked
Most clinical apps say they take accuracy seriously and give you a feedback button. Here is what we did instead.
Every score, row by row
Each calculator was broken into individual claims — every criterion, threshold, band and citation. Each was checked against the primary source and the source wording quoted alongside it, so the comparison can be read rather than trusted. That produced 114 rows across ten scores. A practising doctor then ruled on every row where the app and the source disagreed.
What that turned up
A scoring bug. The Glasgow-Blatchford score asked for haemoglobin twice, once for men and once for women. Answering both scored it twice — a maximum of 29 against a true maximum of 23.
A renal calculation reading too favourably. Our gentamicin calculator fed adjusted body weight into Cockcroft-Gault where UK guidance specifies ideal. In a 140 kg patient that showed 97 ml/min where it should show 69 — every part of the error pointing toward more drug.
Two wrong guidelines. Both upper GI bleed scores cited NICE NG12, which is the suspected-cancer referral guideline. The right one is CG141.
None of those would have been found by a feedback button, because none of them look wrong on the screen.
Deliberate differences are recorded as deliberate
Where we knowingly differ from a published figure, the reason sits in the code beside it so nobody later "corrects" it back. The CIWA-Ar severe threshold is one of these: published cut-offs vary between 15 and 19, and ours is a considered choice rather than an oversight.
What this does not mean
It does not mean the app replaces your trust's guidelines, and it does not mean it cannot be wrong. Guidelines change, and a checked page goes stale like any other. It means the numbers had to earn their place, and that where we differ from a source, we know we do and why.
Pathways are gated, not just reviewed
A calculator can be checked once. A pathway is prose, and prose drifts — so the review is wired into the build rather than left as a promise.
A new pathway cannot ship unread. Its id goes on an unreviewed list the moment it is written, and the build refuses while that list has anything on it. Emptying it is a deliberate act that says a doctor has read the pathway end to end.
The cases are gated with it. Every condition carries four worked cases, and they live in a file the app does not import until they are signed off. The two halves are coupled: signing off a pathway while its cases are still staged fails the build, so they can only land together.
Rejected wording cannot come back. When a sentence is read and rejected, it goes into a list of retired phrasings with what it used to say, what it says now, and who decided — and the build fails if any of them reappears anywhere, including inside a case that teaches the same point a second time.
The most recent round
Fifteen new conditions went through this in September 2026, across two readings. The first produced 62 amendments. Almost none were wrong concepts: they were crisp teaching sentences that had shaded into absolutes — "a D-dimer is useless in an inpatient", "check the ABPI before compressing anyone", "corticosteroids do not work" — each true enough to sound right and too absolute to be safe.
Four pathways were then held back rather than shipped, and rewritten against guidance published this year: the MHRA Drug Safety Update of 16 June 2026 on bradykinin-mediated angioedema, and UKHSA scabies guidance updated on 29 April 2026, which had changed what the app was saying about isolation. One condition — hypoalbuminaemia — was withdrawn on review and returned to a short background note, because it did not generate a decision anyone makes at 3am.
Calculator verification completed August 2026. Pathway review: 136 of 136 read and signed off, most recently 15 September 2026.
All of that was done between shifts, and the app is free with no ads. If it has been useful: