Top holder concentration
A top-ten percentage is only meaningful once you have classified each address. Pool accounts, program vaults and exchange wallets all sit in that list and none of them mean the same thing.
Read the pageSignals: what a pattern in the data can and cannot support
A signal is a pattern several people would notice in the same dataset. It is not a verdict. The failure mode in this section is not bad arithmetic, it is confident inference: reading a distribution or a flow imbalance as intent when a mundane mechanical explanation fits the same data.
Each page here works the same way. It states what the pattern looks like, lists the explanations consistent with it, and then gives the discriminating test that separates them. Where no discriminating test exists, the page says so rather than picking the more dramatic story.
These pages assume you already know what the underlying figures count. If a definition is unfamiliar, start in metrics and come back.
3 pages in this section
Distribution, flow direction and wallet counts are pattern data, not verdicts. These pages set out what each pattern can support, what it cannot, and the alternative explanation you must rule out first.
Buy/sell ratio, read carefully
The same ratio can be computed from trade counts, from volume, or from net flow, and the three answers routinely disagree. Which one your dashboard shows, and what each version can prove.
Read the pageUnique wallets and what they prove
A unique-wallet count measures addresses, and addresses are free. What the metric still supports, what funding-graph analysis adds, and the derived ratios that survive wallet inflation.
Read the pageBefore and after a signal
A signal that survives an attempt to explain it away is worth recording. Recording it properly, with the evidence attached and the checks you skipped noted, is what the verification section is for.
Metrics
Each page takes a single figure from a token dashboard, states its formula, names the inputs it depends on, and shows what the same figure looks like when the inputs are manipulated.
Verification
Procedures for reproducing a dashboard figure from primary sources, recording what you checked, and scoring a token on evidence you gathered yourself rather than on a badge someone else applied.