Research Score methodology

The Research Score v1 ranks tracked crypto assets from 0 to 100 by research priority. This page documents the full method so you can understand, and challenge, every score you see.

A score of 90 is not a 90% chance of profit. It means the asset ranks highly against its peers on the components we can currently measure.

1. Universe

The score is computed across a universe of roughly the top 250 crypto assets by market capitalization, refreshed with market data. Stablecoins and wrapped assets are excluded from the universe because their prices are designed to track another asset.

2. Hard exclusions

An asset receives no score, rather than a low score, if any of the following apply:

• 24-hour trading volume below $500,000.
• Market cap missing.
• Input data is stale (see freshness states on Data sources).

3. Components and raw metrics

4. Normalization

Each raw metric is converted to a percentile rank from 0 to 100 across all eligible assets in the universe. For metrics where lower is better (volatility, drawdown depth), the rank is inverted so that 100 always means “best in universe”.

pct(x_i) = 100 × (rank of x_i among N eligible assets − 1) / (N − 1)
Inverted: pct'(x_i) = 100 − pct(x_i)
Ties receive the average of their ranks.

A component’s percentile is the average of its metric percentiles:

C_k = mean of pct(m) for metrics m in component k with data

5. Aggregation and missing data

Let A be the set of components with data for the asset. The score is the weighted average over A, with weights re-normalized to sum to one:

Score = Σ_{k ∈ A} w_k · C_k  /  Σ_{k ∈ A} w_k
Coverage = Σ_{k ∈ A} w_k   (weights expressed as percentages of 100)

With Trend, Momentum, Liquidity, Tokenomics and Risk resilience available, coverage is 18 + 12 + 15 + 15 + 10 = 70%. Assets with shorter histories, for example fewer than 200 days for the long moving average, may have lower coverage. Coverage is always shown next to the score.

6. Data-quality grade

Each score carries a grade from A to D summarizing the completeness and freshness of its inputs. Higher coverage and fresher inputs earn better grades. Use the grade as a confidence label: a D-grade score should be treated as provisional.

7. Versioning

The weights and metric definitions above are versioned as score_v1. Any change to weights, metrics, universe rules or exclusions produces a new version identifier. Historical scores retain the version they were computed with, so a change in score can always be attributed to market movement or to a method change.

8. Interpretation

• The score is relative. In a falling market, top-ranked assets may still be falling.
• The score describes current conditions; it is not a return forecast.
• Component breakdowns often matter more than the headline number.
• Until per-asset on-chain and catalyst data are connected, about 20% of the intended weight is not represented.

9. Limitations

• Percentile ranks discard the size of differences: the gap between the 50th and 60th percentile may be tiny in raw terms.
• Supply data from aggregators can lag token-schedule changes, which affects the tokenomics proxy.
• Trading volume can include wash trading on low-quality venues.
• Momentum and trend signals can reverse abruptly.