Guide
How Fyndra scores products
Every recommendation comes from a deterministic scoring engine — the same inputs always produce the same output. A language model never picks products and never changes scores; it can only rewrite the explanation text, and only from the verified record.
Relevance
35%How closely the product's keywords, use cases and description overlap with what you described. Partial word matches earn half credit, so synonyms still surface.
Budget
25%How comfortably the price fits your budget. Way under budget scores fine; eating most of it scores lower; exceeding it scores zero.
Features
15%Whether the product's specs and stated features match the preferences you typed — wireless, quiet, compact and so on.
Data quality
10%How complete the product's record is: specs filled in, pros and cons listed, availability known. Thin records score lower so we don't over-recommend what we can't back up.
Price freshness
5%How recently the price was verified against the merchant. Stale prices quietly lose points.
Category fit
10%Whether the product sits in a category that plausibly answers your problem.
What this means for you
You can always see the breakdown: every recommendation shows its scores per axis. If a pick looks wrong, the numbers tell you why — and the alternatives show you the trade-offs we considered.