Upcoming Features

We aim to provide first-principles, testable metrics for on-chain launches. The sections below give the mathematics and how each piece connects to the Analysis Panel.

In Development

Activity Pattern Score

APS combines heterogeneity across funder types, wallet ages, acquisition, and SOL tiers with penalties for concentration and coordination patterns that appear in the top-holder table. The result is bounded and comparable across tokens.

Mathematical foundation and construction

A. Evenness on categorical facets

For a categorical distribution $p=(p_1,\dots,p_S)$ , Shannon entropy is

H(p)=-\sum_{i=1}^{S}p_i\log p_i

Normalize by the maximum $\log S$ to obtain Pielou evenness

J(p)=\dfrac{H(p)}{\log S}\in[0,1]

Facets taken from the Analysis Panel

funder types ⇒ $J_\text{fund}$
wallet age buckets ⇒ $J_\text{age}$
acquisition categories for example BOUGHT and Free ⇒ $J_\text{acq}$
SOL balance tiers for example <1, 1–2, 2–5, 5–10, ≥10 ⇒ $J_\text{sol}$

For small samples, a bias-reduced entropy estimator such as Miller–Madow is used before normalizing.

B. Concentration of top holders

Let $s=(s_1,\dots,s_L)$ be the normalized share vector of the displayed top holders. The Herfindahl–Hirschman measure is

\text{HHI}(s)=\sum_{i=1}^{L}s_i^2

The normalized version ensures comparability across different numbers of top holders:

\text{NHHI}(s)=\dfrac{\text{HHI}-1/L}{1-1/L}\in[0,1]

The concentration metric is then defined as $M_\text{conc}=1-\text{NHHI}$ , where higher values indicate more diffuse ownership.

C. Coordination density from the SOL proximity graph

Create a graph on the displayed holders. Connect two nodes when their SOL balances differ by at most ten percent and they share either the same wallet age in days or the same initial funder. If $m$ is the number of nodes and $E$ the edge set then the density is

\rho=\dfrac{2\,|E|}{m\,(m-1)}\in[0,1]

The coordination metric is set to $M_\text{coord}=1-\rho$ , so higher values indicate weaker clustering under this rule.

D. Additional signals from the table

whale dominance share $W\in[0,1]$ ⇒ $M_\text{whale}=1-W$
Free acquisition fraction $F_\text{free}\in[0,1]$ ⇒ $M_\text{free}=1-F_\text{free}$
.sol domain presence ratio $D\in[0,1]$ ⇒ $M_\text{domain}=D$

E. Composite

\text{APS}=100\sum_{j}w_jM_j\quad\text{with}\quad M_j\in[0,1],\ \sum_j w_j=1,\ w_j\ge 0

bounded in $[0,100]$ by convexity
permutation invariant for entropy parts
monotone in each factor by construction

In Development

Wallet Intelligence Suite

Influence is summarized by a PageRank-style reputation. Discipline is summarized by a conviction score that blends a conservative success bound with a risk-adjusted return built from log returns.

Reputation and conviction

Reputation

A directed weighted graph is built on wallets, where an edge from $v$ to $w$ signifies funding or material influence. The graph's transition matrix, $P$ , is made column-stochastic after normalizing outflows. With a damping factor $d\in(0,1)$ and a teleport vector $\pi$ , the reputation vector $R$ is the solution to:

R=(1-d)\,\pi+d\,P\,R

The Google matrix is positive and stochastic and by Perron–Frobenius the stationary vector is unique. Power iteration converges.

Conviction

Reliability uses the Wilson lower bound for the success rate at ninety five percent confidence

\underline p=\frac{\hat p+\frac{z^2}{2n}-z\sqrt{\tfrac{\hat p(1-\hat p)}{n}+\tfrac{z^2}{4n^2}}}{1+\tfrac{z^2}{n}},\quad \hat p=\tfrac{s}{n},\ z=1.96

Counts are EWMA-weighted across launches to favor recent behavior.

Log returns, defined as

x_i=\ln(1+r_i)

, are used to estimate the EWMA mean

\mu

and volatility

\sigma

. These are combined into a stabilized quality ratio:

Q=\max(\mu,0)/(\sigma+\epsilon)

\text{BCS}=\lambda\,\underline p+(1-\lambda)\,\mathrm{logistic}(kQ),\quad \lambda\in[0,1]

Both terms lie in $[0,1]$ so the blend is bounded.

Experimental models

While these constructions are mathematically well posed, markets remain stochastic. The resulting signals should therefore be interpreted with careful judgment.

Analysis Panel

FAQ