OnlyTheBestStuff

OnlyTheBestStuff is a UK product-discovery site built on a deceptively simple promise: for any category, show only the three best options and nothing else. Where most buying guides bury the answer under affiliate-padded top-ten lists, it resolves each category to a Best Overall, a Runner-up and a Best Value pick — each with an aggregate score, the sources behind it, and current UK pricing. The homepage says it plainly: the three best, nothing else.

The problem it solves is decision fatigue. A shopper researching an air fryer, a robot vacuum or a TV faces dozens of review sites that disagree, each with its own scale, sample size and commercial incentive. OnlyTheBestStuff does the reconciliation: it aggregates ratings from across the review ecosystem, normalizes them onto a common 0–100 scale, and weights them so that independent lab tests, larger samples, more recent reviews and reviews from the shopper's own region count for more — with thin, low-confidence scores pulled back toward the category average rather than allowed to top the chart.

Under the hood is a transparent six-step pipeline the site publishes openly: it gathers candidates from retailer APIs, affiliate feeds and product databases, resolves duplicates into one canonical record using GTIN/ASIN/MPN identifiers plus fuzzy brand and model matching, normalizes and weights the scores, then hands the ranking to a human editor who can override the order — with both the algorithm's score and any override recorded and shown. Rankings refresh weekly across 219 categories in 13 groups. Show your working, not a black box.

The commercial model is handled with unusual candor: the site earns commission on some outbound buy links, including as an Amazon Associate, but is explicit that this never influences rankings — the commercial layer sits strictly downstream of the algorithm. That separation is the whole brand; trust is the product. It is a lean, server-rendered Next.js build with a restrained navy-and-blue identity, whose real sophistication lives in the data layer — identity resolution across messy retailer feeds, defensible score normalization, and a weighting scheme that behaves under small samples — all hidden behind three confident answers per category.