An address is not a person: reading active addresses

Active addresses counts how many distinct addresses appeared in transactions on a network during a period. It is the closest thing on-chain data has to a user count, and it is not a user count. One person can control thousands of addresses, one address can safeguard funds for millions of people, and a large share of all activity is generated by software that has no user at all.

What an address is

An address is a string derived from a public key, or from the code of a smart contract. Creating one requires no permission, no identity check and no interaction with the network: it is a local computation, and a person can generate a million of them on a laptop in an afternoon. Nothing links an address to a legal person unless that person chooses to link it, so the count is of identifiers, not of people.

One person, many addresses

Fragmentation is normal and often automatic. Wallets following the common derivation standards create a fresh address for each receipt by default, for privacy. Chains using the UTXO model, Bitcoin among them, generate a change address on nearly every spend, so a single payment can involve several addresses belonging to one owner. Anyone using multiple wallets, a cold wallet for savings and a hot one for spending, or a separate address per application, multiplies their footprint further.

Address clustering attempts to group addresses under one controlling entity using heuristics — common spending inputs, change patterns, timing. The heuristics are useful and imperfect; they produce estimates with error bars, and privacy techniques degrade them deliberately. Any figure claiming to count entities rather than addresses depends on such a model, and the model's assumptions belong with the number.

Many people, one address

The relationship also runs the other way. A centralized exchange holding assets for millions of customers may settle their trades internally and touch the chain only when moving funds between its own wallets. Those millions of people generate a handful of active addresses. A custody provider, a payment processor or a rollup batching thousands of user transactions into one settlement posting has the same compressing effect.

This matters most when comparing networks. A chain where most users control their own keys will show many more active addresses for the same number of humans than a chain where most users sit behind custodians, and neither arrangement is a measure of the other. Comparing address counts across chains with different custody cultures compares the plumbing, not the population.

Machines, and addresses created to be counted

Automated participants — arbitrage bots, market makers, MEV searchers, liquidators, indexing services — transact continuously and account for a substantial share of activity on active chains. They are legitimate participants performing necessary functions, and they are not users in the sense the metric is usually read as implying.

Some activity is created specifically to be counted. Where a network, an application or an anticipated airdrop rewards participation, the cheapest way to qualify repeatedly is to control many addresses, and address counts rise accordingly. Dusting attacks, which send tiny amounts to large numbers of addresses, make those addresses appear active without their owners doing anything. Both effects inflate the figure without any change in the number of people involved.

The complement to activity is creation. A new address count records identifiers appearing for the first time, and it rises both when a network gains participants and when existing participants spread their balances across more identifiers. Sustained growth in new addresses alongside flat activity per address is consistent with either reading, and no on-chain evidence separates them; the defensible statement is that the identifier population grew.

Counting rules, and what the metric is still good for

Providers differ on the details: whether an address counts when it only receives, whether contract addresses count as participants, whether failed transactions count, whether internal calls count. Those choices can change a figure materially, so a number without a definition is not comparable to another number without a definition. The rules used here are set out in methodology, and they are applied identically across networks so that a comparison at least holds the definition constant.

Within a consistent definition the metric does real work. Active addresses 24h and active addresses 7d compared against their own history show whether a network is being used more or less than it was, and the 30-day change makes the direction explicit. Addresses with balance adds a slower-moving stock series that is harder to inflate cheaply, because leaving a balance costs something. Reading either alongside daily transactions distinguishes more participants from more activity per participant.

One derived figure deserves a warning. Market cap per active address divides a valuation by a count of identifiers, and it inherits every distortion above while adding the market's own figure on top. It can be compared with its own history; treating it as a per-user valuation asserts a relationship between addresses and people that the data does not contain.

A genuine user count would need something the ledger does not contain: an identity layer, or a dependable mapping from identifiers to entities. Part of that mapping exists off-chain, held by exchanges and custodians under know your customer obligations, and it is not published. Figures describing users rather than addresses therefore rest on clustering heuristics with unstated error, on survey data, or on numbers a company chose to disclose about itself. Each is a different kind of evidence from a ledger count, and keeping the difference visible matters most when the two appear in the same sentence.

The network activity section shows these series with their definitions attached, and the glossary covers the address types involved. The next lesson turns from counting activity to judging it, by asking what makes one revenue stream sturdier than another.