How we count what owners say

Every number on this site can be traced to a source. This page explains how the numbers are produced, and — just as importantly — what we deliberately do not do.

What we don't do

We do not buy, test, measure or handle products. We take no product samples. Nothing here is a review in the sense a reviewer would mean it, and if you want someone who has held the thing, read a reviewer.

What we do instead is count. A reviewer spends a week with a product and forms one opinion. We count what thousands of owners said after living with it — which is a different question, and the one that turns out to be harder to answer anywhere else.

Where the specifications come from

Every specification is copied from the manufacturer's own product page or a named publication, and the source is linked beneath each product. Figures are reproduced exactly as stated, including their conditions. Where a manufacturer quotes 44 hours of battery with noise cancellation off and 28 hours with it on, we print both, because the headline number on its own misleads.

Where a figure is not published, the cell is left blank. We never estimate a specification.

Where the owner sentiment comes from

Public comments on YouTube review videos for the product in question. A typical product yields 2,000 to 3,000 comments, of which roughly 200 to 900 mention a specific attribute.

Comments are excluded when they:

• name a different product — comment sections are full of talk about rivals and previous models
• ask a question rather than report an experience
• express a wish rather than an outcome

Comments in Hindi and Hinglish are counted, not discarded. They are roughly 9 percent of the corpus and disproportionately come from Indian buyers rather than the international audience English review channels attract.

The evidence threshold

An attribute must draw at least 12 opinionated mentions, and one polarity must hold at least 60 percent of them, before we describe it at all. Below that we print that the evidence is insufficient rather than characterise a product on a handful of remarks.

This is why some products show verdicts on two attributes and others on none. A page with fewer conclusions is not a less thorough page — it is a more honest one.

Every claim is audited

Before any verdict is published, each one is checked against the comments behind it: do the quotes genuinely say what the count claims, are they about this product, and is the sample large enough? Claims that fail are dropped rather than softened. In a typical run, most proposed verdicts do not survive this step.

How current it is

Owner sentiment is re-counted at least every 30 days. Sentiment older than that stops displaying automatically, so a page cannot quietly go stale while still looking authoritative.

Corrections

If a figure here is wrong, we want to know — every specification links to its source, so errors are checkable. Write to contact@thestacksmiths.com and we will correct it or explain the source.