Where every number comes from, feed by feed

Six families of feed produce every figure on this site, and they differ enormously in how far an outside party can verify them. Chain data can be recomputed by anyone willing to run software; venue data cannot be audited at all; regulated-product disclosures are slow but legally accountable. Knowing which family a number belongs to is more useful than knowing the name of the vendor behind it, because the family determines what kind of error the number is exposed to.

Chain data, read from nodes and indexers

The foundational feed is a full node for each covered network, plus an indexer that decodes raw blocks into queryable records. This produces block counts, transaction counts, active addresses, fees paid, issuance, burns, contract balances, validator sets, reachable node counts and, on proof-of-work networks, difficulty and the hash rate estimated from it. Hash rate deserves a note: it is never observed directly, because miners do not report their work. It is inferred from how quickly blocks arrive relative to the difficulty target, which makes it a statistical estimate that is noisy over hours and reliable over weeks.

Everything in this family is anchored to block height rather than clock time, and read at a fixed distance behind the chain tip. Recent blocks can be discarded in a chain reorganization, and networks reach finality by different rules, so the lag differs per chain and is documented on the data sources page. Figures describing the unconfirmed queue, such as mempool transactions, are the exception: the mempool is local to each node and never globally consistent, so that figure is presented as a single observation point rather than as a network-wide fact.

Market data, aggregated from trading venues

Price and volume come from spot market venues over their public interfaces. The reference price is a volume-weighted composite across a screened set of venues rather than a quote from any single one, because no venue is authoritative and any one of them can be unavailable, halted or dislocated. Screening exists because reported volume is self-published and has a long history of inflation through wash trading; venues that fail the screen are excluded from the composite, and the criteria are stated rather than left to judgment.

This family is the least verifiable on the site. A centralized exchange keeps its order book internally, and no outside party can confirm that a printed trade occurred. Every figure derived from volume — turnover, the volume side of any ratio — inherits that limitation, which is why the site does not build risk conclusions on it.

Protocol accounting, decoded from contract events

Fee and deposit figures for applications are produced by decoding specific contract events and converting token amounts to dollars at the price prevailing when the event occurred. 30-day fees, decentralized exchange volume, total value locked and stablecoin supply all originate here. The mechanical part is exact; the judgment is in the boundaries. Which contracts belong to a protocol, whether a deposit receipt redeposited elsewhere is counted twice, and whether a wrapped token is counted on both chains are decisions, not observations, and they are written out on the methodology page.

Pricing the tokens inside a contract introduces a second dependency. Thinly traded collateral has no reliable price, and the value assigned to it comes from the same composite used elsewhere or, in some protocols, from an oracle whose own feed can lag or fail. Total value locked therefore moves when prices move even if no user does anything, which is a property of the metric rather than a defect in the feed.

Consensus and staking data

Validator and staking figures come from the consensus layer of each network directly: the active set, the staked balance, participation, and penalty events. Staking ratio and staked supply are counted, not modeled. Nominal staking yield is computed from the network's own issuance formula given current participation, and real staking yield subtracts supply growth from it. Both are current-conditions arithmetic rather than a rate anyone has promised, and neither accounts for the operator's commission, for slashing penalties, or for the unbonding delay. Where liquid staking is significant, the share routed through it is shown separately, because the same underlying stake would otherwise be counted once as stake and again as collateral.

Regulated disclosures, the slowest and most accountable feed

Figures about products and companies come from documents filed with regulators or published by issuers under a disclosure obligation. Coins held and assets under management for each exchange-traded product come from the issuer's daily statement. Corporate treasury holdings come from company filings, and where an institution reports positions in listed products, from the 13F filing regime. The corporate treasury policy and digital asset treasury company entries explain what those filings do and do not commit a company to.

This family carries a reporting lag measured in days for products and in weeks or quarters for companies, and the site timestamps it to the publication date rather than to the position date. In exchange for the delay it is the only family where a false figure has a named party attached to it, which is why it is used in preference to any inference whenever the two are available for the same quantity.

Curated records, conflicts and revisions

Sector classifications, entries in the incident log, dated items on the calendar and the written assessments on the risk section are compiled by people from primary sources. Each entry cites the source it rests on, and each records what was not established as well as what was.

Conflicts are resolved by a fixed precedence rather than case by case. Chain observation outranks everything for quantities that exist on a chain. A named party's filing outranks any inference about that party. A venue composite outranks a single venue. A modeled figure from an external data provider is used only where nothing above it is available, and is labeled as modeled wherever it appears. A quantity that survives none of these tests is not published, which is the subject of the previous article in this track.

Numbers also change after publication. A decoding fix, a corrected venue series or an amended filing produces a revision, recorded rather than silently applied, and a backfill that extends history is flagged so that new coverage is not mistaken for new activity. Across every series on this site, the observation time shown with each figure is what allows a reader to distinguish the two.

The current feed inventory, with per-network lag and coverage gaps, is on the data sources page; the definition and formula behind any individual figure is in the metric catalog; and the judgment calls behind boundaries and exclusions are on the methodology page.