Methodology

Updated September 20, 2026

Two separate outputs

The consensus engine produces the score, tier, verdict, and confidence from stored reviewer mentions. The editorial pipeline extracts what each reviewer said, synthesizes useful themes across opinions, and rewrites those themes into buyer editorial. Editorial prose does not assign or change the grade.

When a grade can be shown

A public grade requires at least three contributing creators with product-focused evidence. It also needs either ten product-focused mentions or thirty total mentions. A mention is a stored piece of reviewer evidence; ten mentions do not mean ten different videos.

Until that gate is met, the public score, tier, verdict, and confidence are withheld and the site says “Building consensus.” Useful attributed commentary and source links can still be shown. Attention, such as views, is displayed separately from product quality.

Inspect a product's trace

Open a product from the catalogand look for “Under the hood.” Where a trace matching the stored result is available, it shows the engine's inputs, weights, intermediate values, and evidence checks. A missing or stale trace is not replaced with an invented calculation.

For an example, open the ASICS Megablast computation trace. Its source reviews link back to the cited videos and, where available, the relevant moments. Trace availability can change as the evidence is updated.

What the editorial means

Reviewer opinions are subjective and can disagree. We aim to preserve their meaning and trace claims to their sources. Fabrication, material distortion, and broken traceability are reasons to reject or correct a claim. A small evidence set or an imbalance of positive and negative opinions does not by itself make an attributed statement false.

AI assists the editorial process and can make mistakes. Check the underlying sources and send us a correction when something is wrong.