Composite Overperformance Ranking (COR)
Five metrics standardised into z-scores, averaged, and mapped from a fixed plus-or-minus three sigma range onto 0 to 100.
- Scale
- centred Centred scale — 50 is the neutral point; distance from 50 carries the signal.
- Formula
- mean(z) over 5 metrics, then (z̄ + 3) / 6 × 100
- Inputs
- sentiment
- interactions24h
- socialDominance
- volume24h / marketCap
- 1 − volatility
Purpose
COR is the platform's answer to a problem visible in SMI and OPS: when quantities of wildly different magnitude are summed, the largest one silently takes over. COR avoids that by standardising every input first, so each metric enters the average on identical statistical footing regardless of its native units.
It is also the only Moonboard metric that preserves magnitude. A percentile tells you an asset is ahead; a z-score tells you by how much. Two assets can occupy adjacent percentile ranks while being separated by a full standard deviation, and COR is the metric that shows the difference.
Inputs
Five quantities, each standardised across the universe independently:
| # | Quantity | Direction |
|---|---|---|
| 1 | sentiment | Higher is better |
| 2 | interactions24h | Higher is better |
| 3 | socialDominance | Higher is better |
| 4 | volume24h / marketCap | Higher is better |
| 5 | 1 − min(volatility, 1) | Higher is calmer |
The fifth term is inverted before standardisation so that all five point the same way — otherwise averaging them would cancel signal rather than accumulate it. Note that the Moonboard Score itself is not among the inputs, which distinguishes COR from OPS: the two multi-factor metrics do not share their heaviest term.
Formula
The final line maps the interval [−3, +3] onto [0, 100]. An asset exactly at the universe average on every input scores 50. Each full standard deviation is worth 16.67 points.
The fixed range compresses the scale severely
The ±3σ window is applied to the average of five z-scores, not to a single one. Averaging shrinks dispersion: if the five inputs were uncorrelated, the average of five standard normals would have a standard deviation of about 0.45, not 1. Reaching z̄ = 3 would then require being roughly 6.7 standard deviations out on the average — which effectively never happens.
The practical consequence:
| Mean z-score | COR | How often |
|---|---|---|
| −1.0 | 33.3 | Uncommon |
| −0.5 | 41.7 | Regular |
| 0.0 | 50.0 | The bulk of the universe |
| +0.5 | 58.3 | Regular |
| +1.0 | 66.7 | Uncommon |
| +2.0 | 83.3 | Rare |
| +3.0 | 100.0 | Essentially unreachable |
So COR values cluster tightly around 50, and most of the 0–100 range is never used. A COR of 65 is not a middling result — it is a strong one. Reading COR on the same mental scale as the Moonboard Score will systematically understate what it is saying.
Three of the five are social measures, and interactions and social dominance move together closely. The effective number of independent dimensions is smaller than five, which widens the spread of z̄ somewhat relative to the uncorrelated case — but not enough to change the conclusion that the upper reaches of the scale go unused.
Normalisation
The mean and standard deviation are recomputed over the current universe on every refresh. COR is therefore doubly relative: not only the ranking but the scale itself is defined by the peer set. If the universe becomes more homogeneous, σ shrinks and every asset's z-score inflates without any change in the underlying data.
Where σ is zero — every asset identical on a metric — that metric contributes a z-score of zero rather than a division by zero.
Interpretation
| COR | Reading |
|---|---|
| ~33 | About one standard deviation below average across the five inputs. Consistently weak rather than weak on one axis. |
| ~50 | Average on the composite. The most common region by a wide margin. |
| ~67 | A full standard deviation above average — strong, and rarer than the number looks. |
| ~83 | Two standard deviations up. Very few assets reach this in a normal market. |
Because COR averages rather than ranks, an asset can reach a high value two ways: by being moderately above average everywhere, or by being extreme on one input and average elsewhere. The composite does not distinguish them. Inspecting the underlying fields is the only way to tell which case you are looking at.
Worked example
A clearly above-average asset lands at 55.4 — only five points above neutral. That is the compression at work, and it is the single most important thing to internalise before reading COR alongside the other six metrics.
Limitations
- Most of the scale is unused. Real values cluster in roughly 35–70.
- Z-scores assume a well-behaved distribution. Interactions and turnover are heavy-tailed, so μ and σ are pulled by extremes, and the resulting z-scores are not the tidy standard-normal quantities the formula implies.
- Extreme assets distort everyone's scale, because they inflate σ for the whole universe.
- Three of five inputs are social, so the composite leans social despite equal nominal weighting.
- Averaging hides the shape. Uniformly good and wildly uneven profiles can produce the same COR.
- Doubly universe-dependent. Both position and scale shift when the scanned set changes.
Data quality
Missing or non-numeric values are treated as zero before standardisation, which for most inputs places the asset below average and drags its z-score down. Metrics that fail to produce a finite z-score are skipped, and the average is taken over however many succeeded — so an asset with sparse data is scored on fewer dimensions than one with complete data, without that difference being visible in the output.
Related
- What is a z-score?Standardisation from first principles
- Z-score calculatorCompute z and the COR mapping yourself
- OPSThe other multi-factor metric, weighted rather than standardised
- Percentile rankingWhy COR and the percentile metrics disagree