Social intelligence

Social data entered crypto analysis on a simple premise: communities discuss assets before those assets move, so measuring the discussion gives an early view. The premise is plausible and it is why Moonboard collects social data at all. It is also considerably less settled than the confidence with which it is usually asserted.

This page sets out what the platform measures, what each measurement can support, and the structural problems that no amount of processing removes.

What is measured

These feed SMI, the social component of the Moonboard Score, and the retail leg of WSD.

Volume is not the same as information

The central difficulty is that a mention count treats every post identically. A considered thread from someone with domain expertise and a one-word reply both increment social volume by one.

Interactions partially correct this, on the assumption that posts worth reading get engaged with. That assumption holds better on some platforms than others, and it fails specifically in the cases most worth detecting: engagement can be bought, automated, or reciprocally farmed, and a deliberately promoted asset is precisely the one where interaction counts least reflect genuine interest.

The result is a metric that works well in the ordinary case and degrades exactly when something unusual is happening — which is when analysts look at it most closely.

Attention has no direction

This is the limitation that most often produces confused readings. A protocol exploit, a lawsuit, an exchange delisting and a major partnership all raise social volume sharply. The measurement is symmetric; the events are not.

Sentiment classification is supposed to resolve this, and partially does. But sentiment is itself a model output on text that is unusually hostile to classification, and during a genuine crisis crypto communities produce large volumes of defiant, ironic and rallying language that classifiers frequently read as positive. The correction is weakest precisely when it matters most.

In practice: a social spike is a prompt to find out what happened, not a finding in itself. Moonboard's news feed and trending topics exist to make that lookup fast.

Ticker collisions

Assets are tracked by symbol, and crypto symbols are short, unregulated and frequently collide with ordinary words. Any asset whose ticker is a common English word, an abbreviation in wide use, or the name of an unrelated company accumulates mentions that have nothing to do with it.

Filtering helps and cannot be complete: the same string genuinely refers to different things in different contexts, and resolving which requires understanding the post. For affected assets, social volume carries a persistent additive bias, and no cross-sectional ranking removes it because it is specific to the asset rather than to the universe.

The practical guidance is narrow but real: when an asset with a generic ticker shows unusual social readings, verify against the underlying conversation before treating the reading as a signal.

Platform coverage defines the universe

Social metrics measure conversation on the platforms that are collected. Communities concentrated elsewhere — private Telegram groups, Discord servers, regional platforms, closed forums — are invisible.

This produces a systematic bias that is easy to miss because it looks like data rather than absence: an asset with a large, active community on an uncollected platform reports low social volume, and that low number is indistinguishable from genuine obscurity. It affects non-English-speaking communities disproportionately.

Moonboard treats missing social values as zero. That choice keeps the metrics computable and is documented in the methodology, but it means every social metric should be read as "no evidence of activity" rather than "evidence of no activity".

Reflexivity

Social metrics are published, and published metrics get acted on. Once appearing in a social ranking attracts attention, the ranking is partly measuring its own effect. Anyone wanting an asset to appear prominent has a clear, cheap and well-understood way to arrange it.

This is not hypothetical and it is not solvable by better counting. It is a structural property of any published attention metric, and it is a strong argument for treating social measures as one input among several rather than as a primary ranking — which is why the Moonboard Score deliberately divides social share by market share instead of ranking social volume directly.

Open questions

Stated as questions because they are not answered here. Each would need a study designed against look-ahead bias and a defensible sample before anything could be claimed:

• Does a rise in social volume precede price movement more often than it follows one, once the direction of surprise is controlled for?
• Is engagement rate a better predictor than raw volume, or does it mainly measure how one viral post distributed itself?
• How much of measured social volume for mid-cap assets is organic? See AI agents and synthetic activity.
• Does social dominance carry information beyond what market dominance already implies?

Research notes describes the format answers would have to take.