How we rate
Every rating platform has a model. Most hide it. We publish ours. No pay-to-play. No self-reported data. Everything below is exactly how your score is calculated.
CapyScore
0-100 points · pure algorithmic
Five qualitative dimensions, each scored 0-20 by AI analysis of real product content. No weights, no hidden coefficients — just a straight sum. A product that scores 15 in every dimension gets 75. One that's exceptional in depth but closed-source gets a different profile.
| Dimension | 0-5 | 6-10 | 11-15 | 16-20 |
|---|---|---|---|---|
| AI Depth | Thin API call | — | Fine-tuning, RAG | Proprietary model, breakthrough |
| Innovation | Direct copy | — | New approach in market | Created a new category |
| Utility | Limited feature | — | Solves real problem for many | Indispensable for its users |
| Accessibility | Enterprise-only | — | Freemium with trial | Free / open-source |
| Ecosystem | Isolated product | — | API + plugins + community | Platform with marketplace |
Examples
Claude Code
18+17+18+16+15
Cursor
16+15+17+14+13
Copy.ai
5+4+12+14+8
AI Stamp
Defensibility classification
The question every founder dreads: “Is this a wrapper or the real thing?” AI Stamp answers it. Each product is classified based on its actual technology, not its marketing.
Proprietary model, proprietary data, defensible technology.
e.g. Midjourney, Runway, Anthropic
Deep ecosystem, integrations, open-source community, or marketplace.
e.g. Hugging Face, Cursor, LangChain
AI is a significant feature but not the core product.
e.g. Notion AI, Canva AI
Thin layer over LLM APIs, easily replicable.
e.g. Generic AI writers
AI Proxy in a crowded category with 10+ alternatives.
e.g. Copy.ai in a sea of AI writers
Not yet evaluated or insufficient data.
e.g. Newly submitted products
AI Stamp is a label, not a ranking factor.
It doesn't affect CapyRank numerically. AI Depth in the CapyScore already penalizes thin wrappers (5/20 vs 18/20). A stamp multiplier would be double-counting.
Community
Sentiment + Matchups
Two independent signals that capture what the public thinks. Quick reactions (votes) and deep judgments (head-to-head matchups).
Sentiment
Two reactions per product. Mind Blown = “this is amazing”. Flying Kick = “this doesn't deserve the hype”. One vote per user per product.
Matchups
Side-by-side comparisons powered by vector similarity. Products are paired based on similar features (not category), and scored via TrueSkill — the same system Xbox uses for skill rating.
Diverse pairs maximize information per comparison. Bayesian math means early matchups barely move the needle.
Bayesian smoothing
1 vote barely moves the ranking. 100 votes move it significantly. This is by design. We use Bayesian averaging to prevent a single enthusiastic user (or bot) from manipulating scores. The more data a product has, the more community input matters.
CapyRank
The composite ranking
CapyRank = CapyScore × Community Multiplier
The formula is public. Always will be.
Two views, one product. You can toggle between CapyScore (pure algorithmic) and CapyRank (algorithmic + community). CapyScore shows what the data says. CapyRank shows what happens when people weigh in.
No stamp multiplier. AI Stamp is for filtering and visual context — not for penalizing products numerically. AI Depth in the score already differentiates wrappers from originals.
Community starts neutral. With zero votes and zero matchups, the Community Multiplier is 1.0. CapyRank equals CapyScore until real community data builds up.
Evolution is natural. As votes accumulate and matchups happen, community signal grows. Products that the community loves will rise. Products that don't will fall.
What we don't do
No paid placement
No score or ranking can be purchased. Ever.
No self-reported data
We analyze real product pages, not questionnaires.
No 'verified' badges for sale
Every product is evaluated the same way.
No hidden weights
The formula above is the real formula. Not a simplification.