How the Doorprice Score works
Star averages are a bad summary. Ratings pile up at five stars, sellers solicit and sometimes buy reviews, and a launch-week review can't tell you what happens in month ten. The Doorprice Score treats each source as evidence of a product's real quality, weighs it by how much that evidence should move us, and reports how sure it is.
1. Read every source on the same 0 to 100 scale
- Lab and expert tests: the reviewer's score, rescaled.
- Long-term owner reports (Reddit threads, forums): the share of owners who are positive, with a light prior so 3 of 3 doesn't read as 100%.
- Video reviews: each reviewer's verdict, from creators on a trust list that is earned, not bought.
- Retailer stars: read against what is normal for that store in that category. A 4.5 on a marketplace where 4.4 is typical is barely above average.
- Return and defect rates: compared with the category average.
2. Weigh the evidence
weight = reliability(type) × trust(source) × (1 − e−n/n₀) × 0.5age/half-life × (1 − authenticity risk) × sponsor factor × ownership boost
- Sponsored videos count at 35% weight.
- Long-term reviews and owners with a year or more of use get a boost (×1.25 to ×1.3).
- Authenticity risk comes from review patterns: share of five-star reviews above the category norm, unverified purchases, bursts of reviews in a short window, near-duplicate text.
3. Combine with a prior, and say how sure we are
We start from the category average (σ = 15 points) and update with each signal as a noisy reading of true quality (σ = 7 points at unit weight). The score is the posterior mean; the "likely range" is an 80% interval, widened when sources disagree. Confidence is high below a 4.5-point standard deviation and low above 7.5.
4. Show the disagreements
When reviewers and long-term owners differ by 12 points or more, the product gets a warning. That gap is the most useful thing in a review: it is how you find headphones whose hinges crack in month ten, or air fryers whose coating peels.
Source types
| Type | Reliability | n₀ | Half-life |
|---|---|---|---|
| Lab and expert tests | 1 | 1 | 24 mo |
| Long-term owner reports | 0.9 | 60 | 30 mo |
| Video reviews | 0.7 | 1 | 18 mo |
| Retailer star ratings | 0.45 | 400 | 36 mo |
| Return and defect rates | 0.9 | 1 | 24 mo |
| Community discussion | 0.6 | 30 | 18 mo |
n₀ is how much data a source needs before its weight saturates (one lab test, ~60 owner reports, ~400 store ratings).
How trust gets earned
In production, each source's trust is learned, not assigned: we backtest how well a reviewer's early verdicts predicted long-term owner satisfaction and return rates. Reviewers who called it right gain weight; those who rate everything 9/10 lose it.
Where the data comes from (production plan)
- YouTube Data API for videos; sponsorship from creator disclosures and transcripts.
- Reddit through a commercial data agreement (Reddit requires approval and a contract for commercial use).
- Store ratings from retailer APIs and affiliate feeds.
- Expert sites through licensing or summaries with links, never copied text.
Rules we follow
We never write or buy reviews, never hide negative ones, and label every sponsored source. The FTC's rule on fake reviews has been in force since October 2024.