1. We find the label. Every field starts from the manufacturer's own product page, label PDF, or guaranteed-analysis panel -- never from an Amazon listing, which can carry stale or third-party-edited text. We record the exact source URL per field group. As we acquire physical units, we add our own label photo to the product page; until then, the page says so rather than implying a photo exists.
2. We transcribe, not summarize. Organism names, CFU counts, prebiotics and storage wording are copied as printed. If a number isn't stated, the field says "unstated" -- we never calculate or guess a value the label doesn't give.
3. We match strains to studies. Each organism links to a strain page listing the canine trials we could find on PubMed/PMC, by PMID. Cat, human and rodent studies are shown separately and never count toward a "canine evidence" badge.
Evidence badge rule: 0 dog studies found shows a grey "0 found" badge. 1-2 dog studies shows a grey "N studies" badge. 3 or more canine RCTs shows an amber "N RCTs" badge. The badge counts studies on that specific numbered strain, not the species in general.
"Stronger value" rule on compare pages: more strain-numbered organisms counts as stronger; a CFU guarantee through the printed expiry date counts as stronger than one stated only at manufacture; more canine RCTs counts as stronger. A higher raw CFU count, on its own, is never marked as the stronger value -- there is no evidence that more CFU performs better independent of strain and dose.
What we don't do: we don't show prices, price history, or star ratings; we don't accept payment for placement or for a better position in any table; and we don't write a review we couldn't source from the label, a study, or a dated manufacturer page.
Keeping it current: labels are re-checked on a quarterly cycle. Each product card shows the date its label was last verified; if a formula changes, the card updates and the old numbers are replaced, not hidden.