Twenty drug pairs test whether a model can identify a real interaction, name the mechanism and clinical effect, prescribe an executable action, and supply every exact dose, washout, threshold, or monitoring interval. Three negative or reversed controls punish reflexively calling every pair an interaction.
UNVALIDATED v2.0. v1.0 was shaken down against Claude Fable 5.1 and scored 20/20 on significant; that does not validate the new numeric field. Run a fresh single-model shakedown before a recorded lineup.
The answer key is challenge-defs/assets/the-drug-interaction-desk.json, one entry per row, tied to a package insert, FDA communication, or society guideline. Severity grades remain excluded because major interaction databases disagree on them; mechanism, clinical effect, action, and label-level numbers are the stable facts.
Mechanical fields. significant is exact: seventeen true and three false — 5a9e, 1b6c, and 27ea. numbers must contain every number needed to execute the action, and must be empty only when the key is empty. Judge semantic equivalence rather than punctuation. The high-value numeric rows are b9d5 (asymmetric 2-week/5-week and 24-hour washouts), e3f9 (50 percent, 1 to 2 weeks, then about 4 weeks), d218 (the actual reduced colchicine regimens), 4f63 (30 minutes before or 8 hours after), b5cd (36 hours), and ec46 (14 days).
Human-read fields. mechanism must name the enzyme, transporter, receptor, or additive effect. effect must be clinical. action must be executable and consistent with numbers; a correct number attached to the wrong action still fails.
Controls and traps. 8d31/5a9e tests opposite answers within one drug class. 1b6c tests a contraceptive myth. 27ea tests superseded blanket metformin holds while still rewarding the eGFR-under-30 exception. 6d70 rejects febuxostat as an escape because it shares the xanthine-oxidase mechanism.
Twenty drug pairs have landed on the interaction desk. For each, say whether there is a clinically significant interaction, what causes it, what happens to the patient, and what to do.
Some of these pairs have no meaningful interaction. Some are widely believed to interact and do not. Say so when that is the case. Do not manufacture a mechanism to fill a field.
[1] id=a41f warfarin + fluconazole 150 mg [2] id=7c02 simvastatin 40 mg + clarithromycin [3] id=b9d5 linezolid + sertraline [4] id=3e18 sildenafil + isosorbide mononitrate [5] id=c6a7 lisinopril + spironolactone [6] id=2f4b methotrexate 15 mg weekly + trimethoprim-sulfamethoxazole [7] id=8d31 clopidogrel + omeprazole [8] id=5a9e clopidogrel + pantoprazole [9] id=f072 rifampin + combined oral contraceptive [10] id=1b6c amoxicillin + combined oral contraceptive [11] id=e3f9 amiodarone + digoxin [12] id=94a2 lithium + ibuprofen [13] id=6d70 allopurinol + azathioprine [14] id=0c85 ciprofloxacin + tizanidine [15] id=a7be carbamazepine + apixaban [16] id=d218 colchicine + clarithromycin [17] id=4f63 ibuprofen 400 mg three times daily + aspirin 81 mg for secondary prevention [18] id=b5cd sacubitril/valsartan + lisinopril [19] id=27ea metformin + intravenous iodinated contrast [20] id=ec46 phenelzine + pseudoephedrine
Assume adults with normal renal and hepatic function unless the pair itself implies otherwise.
significant is true only if the pair changes what you would prescribe, dose, monitor, or warn the patient about. A theoretical interaction nobody acts on is false.mechanism is the specific pharmacology in one phrase — the enzyme, the transporter, the receptor, the additive effect. Not "drug interaction" and not "altered metabolism". Use "none" when significant is false.effect is what happens to the patient. Name the clinical event, not the direction of a serum level alone.actionnumbers is an array of every exact number needed to carry out the action: a dose or dose fraction, a washout or separation time, a monitoring interval, or a threshold that changes the plan. Copy the number with enough context to show what it controls. Use an empty array only when the row has no such number. A number already printed in the pair does not count unless it is also part of the action.note is yours. Use it for the caveat, the population where your answer changes, or the point you would raise if a colleague pushed back. One or two sentences. Leave it as an empty string if you have nothing to add.Raw JSON, nothing else. No markdown fence, no preamble, no commentary after.
{
"rows": [
{ "id": "...", "significant": true, "mechanism": "...", "effect": "...", "action": "...", "numbers": ["..."], "note": "..." }
]
}
Keys in that order. significant is a boolean, not a string.