moonboard.ioResearch & Data
Methodology

Whale-Social Divergence (WSD)

The difference between an asset's turnover rank and its retail-attention rank, centred on 50.

Scale
centred Centred scale — 50 is the neutral point; distance from 50 carries the signal.
Formula
50 + (turnover percentile − retail percentile) / 2
Inputs
  • volume24h
  • marketCap
  • sentiment
  • socialVolume24h
  • interactions24h
There is no on-chain data in this metric

"Whale" is a label for the first term, not a measurement of large-holder behaviour. WSD contains no wallet data, no transaction sizes, no exchange flows and no holder distribution. Its first leg is reported trading volume divided by market capitalisation — the same turnover figure that drives LMR. Turnover rises when anyone trades actively, whether that is one large participant or ten thousand small ones. Read WSD as "trading activity versus social attention", which is what it computes.

Purpose

Trading activity and social attention usually move together. WSD isolates the cases where they do not, by ranking each separately and subtracting the ranks:

  • Turnover rank above attention rank — the asset is being traded more actively than it is being discussed.
  • Attention rank above turnover rank — the asset is being discussed more than its trading activity reflects.

Both are descriptions of a present mismatch between two observable quantities. Neither implies who is doing the trading, nor what happens next.

Formula

— Leg 1: trading activity — turnover = volume24h / marketCap × 100 smf = log₁₀(turnover + 1) smfPct = percentile of smf across the universe — Leg 2: retail attention — activity = socialVolume24h + interactions24h normSocial = min(activity / 50 000, 1) × 100 retail = sentiment × 0.6 + normSocial × 0.4 retailPct = percentile of retail across the universe — Combine — WSD = clamp( 50 + (smfPct − retailPct) / 2 , 0 , 100 )

Since both percentiles lie in [0, 100], their difference lies in [−100, 100], halved to [−50, 50] and shifted to [0, 100]. The clamp is a guard that never binds in practice. Dividing by two is what makes 50 the neutral point and compresses the output into half the nominal range: a WSD of 75 means a 50-point percentile gap, which is a very large divergence.

The social half of the retail index saturates almost immediately

normSocial reaches its maximum at a combined social activity of 50 000. Because interactions24h alone routinely runs into the hundreds of thousands or millions, that ceiling is cleared by most assets with any meaningful following. For all of them the retail index collapses to:

retail = sentiment × 0.6 + 40

which is a monotonic function of sentiment — so retailPct becomes, in effect, a pure sentiment ranking across most of the universe. The social-activity component only discriminates among assets below the 50 000 threshold. In practice WSD is therefore closer to "turnover rank minus sentiment rank" than the formula's structure suggests.

Interpretation

The application presents WSD in four regions. They are descriptive classifications of a present state, and they are described that way here — not as trade instructions.

WSDObserved condition
> 70Turnover rank well above attention rank. The asset trades actively while sentiment and social presence rank low. Sometimes framed as quiet accumulation; equally consistent with distribution, forced selling, or a market-making arrangement. The metric cannot tell these apart.
50–70Turnover somewhat ahead of attention, or both ranks high together — the pattern of a broadly active asset.
~50The two ranks agree. Either both high, or both low; the metric does not distinguish an actively traded, widely discussed asset from a neglected one. Check the underlying fields.
< 30Attention rank well above turnover rank. Positive conversation without matching trading activity.
The buy and sell framing does not survive scrutiny

A divergence of this kind is sometimes read as informed participants acting ahead of the crowd. That reading requires an assumption the data does not support: that turnover identifies who is trading. It does not. High turnover with low sentiment is exactly what a distressed asset looks like while holders exit, and exactly what an accumulating asset looks like — the two are indistinguishable in this metric. Nothing here should be treated as a buy or sell signal, and Moonboard does not present it as one.

Worked example

Asset A turnover 6.0 % → smf = log₁₀(7.0) = 0.845 → smfPct = 88 sentiment 41, activity 120 000 → normSocial = 100 retail = 41 × 0.6 + 40 = 64.6 → retailPct = 24 WSD = 50 + (88 − 24) / 2 = 82.0 Actively traded, sentiment near the bottom of the universe. Asset B turnover 0.4 % → smf = log₁₀(1.4) = 0.146 → smfPct = 19 sentiment 84, activity 310 000 → normSocial = 100 retail = 84 × 0.6 + 40 = 90.4 → retailPct = 91 WSD = 50 + (19 − 91) / 2 = 14.0 Warmly discussed, comparatively little trading.

Note that in both cases normSocial is pinned at 100, so the retail leg reduced to sentiment alone. That is the saturation described above, and it is the normal case rather than the exception.

Limitations

  • No large-holder data. Turnover is not a measure of whale activity, and the metric's name should not be read as a claim that it is.
  • The retail leg is effectively sentiment for most assets, because its social component saturates at a low threshold.
  • The mid-range is ambiguous. A WSD near 50 covers both "high on both legs" and "low on both legs" — opposite situations with the same output.
  • Volume quality is unverified. Wash trading inflates the first leg directly; no filter is applied.
  • Social quality is unverified. Coordinated posting inflates the second leg. See AI agents and synthetic engagement.
  • Doubly universe-dependent. Both legs are percentiles, so WSD can move when neither of the asset's own inputs changed.

Data quality

A zero or missing market cap returns the neutral value of 50 rather than an undefined ratio. Missing social fields are read as zero, which lowers the retail leg and mechanically raises WSD — so an asset with a social collection gap can appear in the high band for no reason other than absent data. When a WSD reading looks striking, checking that the social fields are populated is the first thing worth doing.

Related