Scope
Ethereum cohort sample only
The report aggregates the holder cohorts WalletWall’s scheduled analytics already cover (active, whale, dormant cohorts). It is a cohort sample, not full-market coverage, and says so in the artifact itself.
Zero new data collection
Editions are computed from the existing Dune auto-run snapshot the app already caches. Publishing an edition never executes a query and never adds analytics cost.
How figures are computed
All aggregation is a pure, deterministic function over the cached cohort snapshot:1
Wallets observed
Count of cohort leaves in the snapshot.
2
Exposed-signature share
Wallets flagged with observed signature exposure, as a share of wallets observed, plus their share of observed cohort value.
3
Dormant share
Wallets in the dormant cohort or carrying dormancy flags.
4
Value concentration (top-5 / top-10 / top-25)
The largest observed balances as a share of total observed cohort value, at three depths so readers can inspect how concentrated the sample is.
5
Observation floor
Below 10 observed wallets an edition reports insufficient data with no figures. Small samples are not findings.
6
Deltas
Compared only against a prior edition that itself had real figures; a trend is never fabricated from an insufficient edition.
Coupled signals — one query, not three
Signature-exposure, dormancy, and vault-candidate detection currently run over the same dormant-wallet query. When a figure is fully covered by the dormant set it is the same wallets counted again, not an independent finding. The report detects this coupling deterministically and, when present, presents the three as one coupled signal — never as three independent or corroborating findings, and never as a Venn of independent sets. The limitation is stated in the edition’s own boundaries and methodology.Daily interactive editions
Each edition is generated after the existing 02:00 / 03:00 UTC Dune cache-warmer produces a coherent, fresh Redis snapshot — the report never runs its own schedule and never reads Dune itself.1
Coherence gate
All required cohorts must be present, non-empty, warmed on the same UTC day, and within the freshness window. A mixed-day or partial batch does not publish.
2
Deterministic evidence
Exact metrics become approved, formatted tokens; Claude may cite nothing else.
3
Constrained narrative
Claude writes only text fields (headline, summary, findings). A deterministic validator rejects any unsourced number, banned claim, or unsafe content before an edition can publish.
4
Immutable edition
A valid edition is published to a stable URL (
/exposure-report/<date>); a failure keeps the prior edition live.observed, observed-coupled, or
insufficient).