How we work — and why we don't buy products
Our evidence model: review-data research, reproducible aggregates, and no hands-on testing. What that means for what we can and cannot tell you.
The short version
We do not buy, borrow or physically test products. Every number on this site comes from aggregated Amazon reviews (2023 vintage), computed offline and published with its sample size. We are a data-research publication, not a test kitchen.
That is a deliberate limit, not a temporary gap. It shapes what we publish and how we word it.
What we do
- Aggregate the review corpus. 18.1M reviews across 1,186,539 products, processed into category, problem, price-band and lifecycle aggregates.
- Publish investigations, not product descriptions. Each investigation combines several findings, states the sample, and separates what the data shows from what it does not.
- Show the absolute rate, not just the ratio. A 7.8× ratio built on a 0.04% baseline is not a 7.8× failure probability. We say so on the page.
- Correct errors in public. Corrections are logged, not silently edited; see Corrections .
- Disclose the money. Some outbound links are Amazon affiliate links; they do not change a score or a finding. See Affiliate disclosure .
What we do not do
- No hands-on reviews. We do not write “we tested”, “lab tested”, “our test kitchen” or “after 30 days of use”.
- No invented experience. We do not invent a tester, a photo, a measurement or a purchase date.
- No sponsored rankings. A brand cannot pay to change a score, a ranking or a finding.
- No raw review republication. We publish aggregates, rates and short representative quotes only.
Why this is a limit, not a shortcut
A review corpus answers a specific question: what do a lot of owners report over time? It is good at finding where complaints concentrate, how ratings drift and which product types fail in similar ways. It is not a lab. It cannot measure a specific pan’s coating thickness, thermal shock or whether the unit in your cart matches the unit in the sample.
We would rather be precise about that boundary than pretend to a kind of evidence we do not have. It also keeps the work reproducible: another person can re-run the aggregation and check the number, which is not true of a private test.
How to read a page
- Check the byline and date. Every investigation is dated; a 2023 snapshot is not a current price or availability claim.
- Check the sample. A category rate with 229,181 reviews and a product rate with 12 reviews are different kinds of evidence.
- Read “what this does not tell you”. It is part of the page, not a disclaimer bolted on at the end.
- Prefer the absolute rate. The ratio tells you where a complaint concentrates; the absolute rate tells you how common it is.
What would change this
If we ever add hands-on testing, it will require a new public decision, a named tester, raw measurements and a clear “Tested by” label on the page. Until then, no page on this site claims first-hand testing.