- Shipped
- August 27, 2026 at 3:34 PM UTC
- Author
- Kamo
- Commit
- 19460fc
Prescribing ran three checks the practice's own data can answer and no drug-drug interaction check at all. Declining to build one was right -- interaction knowledge is licensed, and a hand-rolled table would cover a few dozen pairs out of tens of thousands while being trusted as though it covered all of them. What was wrong was leaving the omission invisible. A review that returns no warnings looks identical whether interactions were checked and found clean or never checked at all, so a prescriber saw a safety check run, saw it come back clean, and reasonably concluded the drug had been compared against the patient's other medications. Nothing had. That is the same "an empty result is not an established fact" error the exchange transport refuses to make, at the highest stakes in the product -- and worse in one way: a failed fetch is intermittent and eventually noticed, whereas a capability the platform does not have is absent from every prescription forever and would never be reported as a bug. So the safety service now returns what was found AND what was never examined. DrugInteractionSource is the seam a licensed database drops into; **************** throws rather than returning an empty list, because an empty list here asserts that two drugs do not interact. The gap rides on the 409 so the override screen repeats it, and lands on the prescription's own log line so a later review knows what the system could check at the time. When a practice licenses a database the disclosure disappears on its own -- it is derived from isConfigured(), not written into the client.