Define a useful question.
We start with a cooking or preparation task and the person’s practical constraints: quantity, working space, handling, cleaning, storage and budget. A product should answer that question rather than merely occupy a category in a shopping list.
Keep the evidence attached.
For a research shortlist, we retain the exact product name, manufacturer source, relevant specifications and the date checked. We identify missing information. Performance, durability, usability and long-term value need stronger evidence than a specification table.
Use the right label.
Research-based means we compared available information. Hands-on means the exact model was used and documented. A recipe test means the stated version was cooked and the method and result recorded. These labels cannot be earned by generating a photograph or repeating a retailer’s claim.
Show why it might be wrong for someone.
Every recommendation should retain meaningful limitations: size, weight, maintenance, region or a capability it does not provide. Include useful non-affiliate alternatives where they fit. We do not invent a single universal score to hide differences between tasks.
Review when something changes.
A model revision, discontinued part, changed merchant destination or corrected specification can change a recommendation. Keep the original decision record and update the affected page rather than quietly presenting old evidence as new testing.