Rankquant
MethodologyAbout⌕ Search

We fixed reviews

Normalized review scores for things you love

Rankquant is better: we aggregate a product's rating based on each users relative score.

See our ratings

WineFilm and TVCruisesHotels

The process

Four deterministic steps from review to rank

STEP 01

Per-reviewer z-score

Subtract the reviewer's personal mean, divide by their personal stddev. Every reviewer now speaks the same dimensionless language.

STEP 02

Three lenses

R1 unweighted (headline). R2 source-weighted by credibility. R3 broadens to constant-rater reviewers via imputed σ.

STEP 03

90% CI floor

Ẑ − 1.645·SE penalizes thin samples. A 4-reviewer +2.1 mean loses to an 80-reviewer +1.6 mean.

STEP 04

ECDF percentile

Rank floor against the universe → 0–100. Cohort percentile re-ranks within same-category, ±20%-price peers.

Read the full methodology →

Why us

What we do that Google and ChatGPT can't

vs Google Search
Popular ≠ good

Search ranks by links and click behavior. We rank by what bias-corrected reviewers actually thought. The most-clicked wine isn't the highest-quality wine.

vs ChatGPT / Claude / Perplexity
No hallucinated scores

AI summarizes whatever it scraped — usually flat averages or vibes. Our methodology is published, deterministic, and reproducible. Every score has a paper trail.

vs Amazon / Yelp / Vivino
Inflation removed

Native platforms have rating clusters: 4.5 means nothing when every product gets a 4.5. We strip the cluster effect by re-centering each reviewer on their own scale.