Key Takeaways
- What it is: The count of addresses appearing on a chain for the first time in a period, plus the narrower count of those that received a nonzero balance.
- What it actually measures: The cost of creating an address, more than the arrival of people. Creating one requires no permission, no identity, and on most chains no more than a transaction fee.
- Mechanical inflation: Hierarchical deterministic wallets produce a fresh receiving address per payment by design, and UTXO change outputs create an address on every spend.
- Deliberate inflation: Airdrop and points farming manufacture first appearances as the explicit objective, and dust distribution manufactures them at the price of a fee.
- Why funded exists: The funded subset filters addresses that were merely referenced. The gap between new and new-funded is itself the diagnostic.
- Practical rule: Never compare new-address counts across chains without conditioning on what a transaction costs on each.
Who This Guide Is For
Read this page if you have seen new addresses presented as user growth and want to know what a first appearance on a chain actually costs to produce. The answer, on most chains, is one transaction fee, and everything else about interpreting the metric follows from that.
For the separate question of whether an address is a person at all, see the Swoopr Investment guide to active addresses, which covers the address-to-user mapping problem this page assumes rather than repeats.
Educational content. Not individualized financial advice. Every figure below is hypothetical arithmetic used to demonstrate a method.
What Does New Addresses Measure?
New addresses measures how many address strings appeared on a chain for the first time during a period. That is a statement about the chain's history of references, not about population, and the distinction matters because producing a first appearance is close to free.
Creating an address requires nothing
An address is derived from a key pair generated offline: no registration, no fee, no limit, and none of them exist on any chain until something references them. The chain-side step is equally cheap. On an account chain, an address enters the chain's history the moment a transaction names it, so anyone can create somebody else's first appearance by sending a trivial amount to a string they picked. Producing a new address in the metric does not require controlling the address, and that has no workaround at the data level.
Why the funded subset exists
Providers publish a narrower series counting addresses that received a nonzero balance. Funded is a real improvement: it excludes addresses that were only referenced, only pre-generated, or only mentioned in a failed interaction. It does not exclude an address funded with a fraction of a cent, which is why it is a first filter rather than the answer.
The cumulative version means even less
New addresses is the first difference of the cumulative set of distinct addresses ever seen, so it can never be negative: addresses are never removed. A chain adding 100,000 new addresses a day while losing every one of them still accumulates 36,500,000 addresses over a year, and its total-addresses figure keeps rising throughout. A cumulative address total describes an accounting identity rather than adoption.
How Is New Addresses Constructed?
Construction is a set-difference operation: take the distinct addresses observed in the interval and subtract every address observed in any earlier interval. The result is entirely determined by what counts as an observation, and the ledger model changes that answer.
Account chains
An address is a persistent account, entering the chain's history when a transaction first references it as sender or recipient. The sender case requires the key holder to act. The recipient case does not, which is the path spam takes. Some chains require a minimum balance or rent deposit before an account is instantiated at all, which collapses the new and funded series into effectively one number.
UTXO chains
An address is a spending condition attached to an output, and wallet software works against address reuse deliberately. Two mechanisms manufacture addresses. Receiving wallets present a fresh address per payment so that observers cannot link incoming payments to one another. Less obviously, spending creates one too.
Inputs are consumed whole. A wallet spending a 1.0-unit output to pay 0.3, with a fee of 0.001, must return the remaining 0.699 to itself, and it returns it to a newly derived change address rather than the original one. The mechanics are set out in the Bitcoin Developer Guide: Transactions. A single payment between two existing users therefore produces two new addresses, one for the recipient and one for the sender's change, and adds zero people to the network. A hypothetical merchant receiving 200 payments in a year and spending 200 times generates up to 400 addresses while remaining one participant.
What the pipeline cannot recover
No filter applied later can distinguish a fresh receiving address from a fresh person, because the two are identical at the data level. Clustering heuristics operate on spending patterns that a single-use address has not yet produced.
Formula and Measurement Logic
New addresses in period t = |A_t| minus |A_t intersected with (A_1 union ... union A_(t-1))|, where A_t is the set of addresses observed in period t
The expression is exact and the arithmetic is never in dispute. Everything contentious sits in the definition of observed, and the useful series are built by tightening it in stages.
| Filter stage | Definition of observed | What it removes |
|---|---|---|
| New | Referenced in any transaction | Nothing. Includes addresses nobody controls. |
| New funded | Received a nonzero balance | Pure references, pre-generation, failed interactions |
| New funded above a fee floor | Received more than the cost of moving it | Dust distribution, which the funded flag does not catch |
| New and retained | Funded, then active again in a later window | Single-use farming, which passes every filter above |
The value floor deserves a note. Setting it in fiat terms makes the series incomparable across chains with different fee levels; setting it as a multiple of the prevailing transaction fee makes it self-adjusting and comparable, at the cost of moving with the fee market. Either way the choice must be published, because it is the difference between catching dust distribution and not.
How Should New Addresses Be Interpreted?
Interpret it as an upper bound on first-time participation, read together with the filters below it. The single number carries almost no information; the shape of the funnel carries most of it.
The gap is the signal
Each filter catches a different failure, and no filter catches all of them. Reading the four series together identifies which pattern is driving a spike.
| Pattern | New addresses | Funded share | Above-fee share | Retention |
|---|---|---|---|---|
| Genuine onboarding | Rises | High | High | High |
| Wallet address rotation | Rises with payment volume | High | Varies | Near zero by design |
| Dust distribution | Rises sharply | High, at trivial value | Near zero | Near zero |
| Airdrop or points farming | Rises sharply | High | High | Near zero |
Note the last row. Farming passes the funded test and the value-floor test, because a farmer must fund each address with enough to transact. Only retention separates it.
What it never supports
It does not support a headcount of new users, because address creation and person arrival are different events with no fixed ratio between them. It does not support cross-chain adoption comparison on the raw count. And it does not support reading a decline as user loss, since a fall on a base layer is a routine consequence of activity moving to a rollup.
Step-by-Step Workflow
- Pull the new and funded series for the same interval and chain, from the same provider.
- Compute the funded share. A sharp fall during a spike points at addresses being referenced rather than used.
- Apply a value floor expressed as a multiple of the prevailing transaction fee, and compute the above-floor share.
- Check whether an airdrop, points program, or incentive campaign was running in the window, since farming survives both filters above.
- Follow the day-0 cohort forward and measure how many are active again at 30 and 90 days.
- Identify the funder of each new funded address, and check how concentrated those funders are.
- If comparing chains, record what a simple transfer costs on each first.
- Report the retained figure alongside the headline, and describe the headline as an upper bound.
Worked Hypothetical Scenario
The following arithmetic is hypothetical and is constructed to show how much a headline new-address figure survives each filter. Take a single day's cohort of 100,000 addresses appearing for the first time.
| Stage | Count | Share of new | Share of previous stage |
|---|---|---|---|
| New addresses | 100,000 | 100.0% | Not applicable |
| Funded, any amount | 41,000 | 41.0% | 41.0% |
| Funded above one transaction fee | 12,000 | 12.0% | 29.3% |
| Active again within 30 days | 3,500 | 3.5% | 29.2% |
| Active again in days 61 to 90 | 1,900 | 1.9% | 54.3% |
The headline was 100,000 and the number of addresses that received something worth moving and came back within a month was 3,500, which is 3.5 percent of it. The 59,000 addresses that were never funded at all cost their creator nothing but a share of a batched transaction. Note also the last row: 1,900 of the 3,500 thirty-day survivors were still active two months later, a 54.3 percent survival rate at that stage, which is far healthier than anything above it. Attrition is concentrated at the very front of the funnel, which is exactly where the headline number is measured.
Why the cross-chain comparison inverts
Now compare two chains on the same day. Chain L has very low transaction fees. Chain H has high ones. Nothing else about them is stipulated.
| Measure | Chain L | Chain H | Ratio |
|---|---|---|---|
| New addresses | 100,000 | 14,000 | 7.1 to 1 |
| Funded share | 41% | 88% | Not applicable |
| Funded count | 41,000 | 12,320 | 3.3 to 1 |
| Above-fee share | 12% | 79% | Not applicable |
| Above-fee count | 12,000 | 11,060 | 1.08 to 1 |
Chain L looks seven times larger on the headline. After the funded filter the gap is 3.3 to 1, and after the fee floor it is 1.08 to 1, roughly an 8 percent difference. Almost the entire headline gap was the price of creating an address, not a difference in arrivals. Chain H's higher shares at every stage are the same fact seen from the other side: when creating an address costs something, fewer worthless ones get created.
The cost of manufacturing the metric
Put a price on it. Producing 1,000,000 first appearances by sending a trivial amount to a million generated strings costs about $1,000 where a transfer costs a tenth of a cent, and about $2,000,000 where a transfer costs two dollars. That is a factor of 2,000 for the identical distortion, and it is the same factor that makes the raw counts incomparable in the first place.
What Can Make the Interpretation Wrong?
- Reading it as new users. The metric counts first appearances of strings. Creating one does not require controlling it, so an attacker can manufacture a million new addresses without holding a single key.
- Ignoring wallet mechanics. Fresh receiving addresses per payment and change outputs on every spend generate first appearances continuously from existing participants. This is correct wallet behavior, not an anomaly, and it never stops.
- Stopping at the funded series. Funded excludes bare references but not dust. An address funded with a fraction of a cent passes the test and can never economically move what it holds.
- Missing farming because the funded share looks healthy. Farmers fund their addresses properly, because they have to in order to transact. Farming clears the funded and value-floor filters and is only visible in retention.
- Comparing chains without conditioning on fees. Fee level sets the unit price of a manufactured address and dominates raw cross-chain differences.
- Quoting cumulative addresses. The cumulative set never shrinks, so the total rises on a chain losing every participant it gains.
- Attributing a base-layer decline to user loss. Addresses created on a rollup do not appear in base-layer counts, so migration reads as contraction.
- Assuming clustering will fix it later. Entity clustering works from spending patterns, and a single-use address has produced none. This metric is the one least improved by entity resolution.
Cross-Network and Provider Comparison
The transaction fee is the confounder in every cross-chain new-address comparison. The metric's numerator is an activity anyone can generate, and the fee is the price of generating it, so the count on any chain reflects that price at least as strongly as it reflects demand. The worked example above collapsed a 7.1 to 1 headline gap to 1.08 to 1 with one fee-relative filter.
Ledger model is the second axis. UTXO chains manufacture addresses mechanically through fresh receiving addresses and change outputs, so they generate more first appearances for the same human activity. On account chains reuse is normal, so they generate fewer. Where a minimum balance or rent deposit is required for an account to exist, the new and funded series converge and the raw count sits structurally lower again. See the Swoopr Investment guide to UTXO and account models.
Provider disagreement reduces to two questions. First, whether a first appearance requires receiving value or merely being referenced, which can change the figure by more than half. Second, whether contract addresses created by deployment are counted, which matters enormously on chains with active factory contracts. See Coin Metrics: New Addresses for one provider's choices.
Advanced Analytical Methods
First-funder concentration
Every newly funded address received its first inbound transfer from somewhere, and that sender is a clustering key requiring no heuristic. If 30,000 of a day's 41,000 newly funded addresses were first funded by twelve addresses, that is 73.2 percent of the cohort traced to a dozen sources: a farm, not an onboarding wave. It works where general clustering fails, because it uses the one transaction a single-use address is guaranteed to have.
Fee-relative value floors
Express the funding threshold as a multiple of the prevailing transaction fee rather than a fixed currency amount. The resulting series is comparable across chains and self-adjusting as fee markets move.
Cohort survival curves
Track each day's cohort forward rather than reading a single day. Steep early attrition followed by a stable plateau is normal; a cohort that vanishes entirely within a month indicates single-use creation. The method is developed further in the Swoopr Investment guide to on-chain user retention.
Deployment separation
Split contract addresses out of the count. On a chain with active factories, deployed contracts move the series for reasons unrelated to participation, and the split is cheap because deployment is a distinguishable event.
Incentive-window annotation
Mark airdrop snapshots, points programs, and campaign deadlines on the series before interpreting any spike. Farming clears every value-based filter, so knowing when a program ran substitutes for a filter that does not exist.
Practical Checklist
- I pulled the new and funded series together, not the headline alone.
- I computed the funded share and watched it during any spike.
- I applied a value floor expressed relative to the transaction fee.
- I checked for an incentive program running in the window.
- I followed the cohort forward to at least 30 days.
- I checked how concentrated the first funders were.
- I recorded what a transfer costs on each chain before comparing counts.
- I know whether contract deployments are counted as new addresses.
- I accounted for change outputs and receiving-address rotation.
- I did not quote a cumulative address total as adoption.
- I considered rollup migration before reading a decline.
Frequently Asked Questions
Does a new address mean a new user?
No, and the gap is not small. Creating an address requires no permission and no fee, and putting one on the chain requires only that some transaction references it. Because you do not need to control an address to make it appear, someone can manufacture a million new addresses without holding a single key to any of them.
Why does airdrop farming survive the funded filter?
Because a farmer has to fund each address properly for it to transact and qualify. Farming therefore clears both the funded test and a fee-relative value floor, and it shows up only in retention, where the cohort goes quiet once the program ends. This is the strongest argument for carrying a retention series alongside the headline.
Why can new-address counts not be compared across chains?
Because the transaction fee sets the price of manufacturing the metric, and fees differ by orders of magnitude. In the hypothetical comparison on this page, one chain looked 7.1 times larger on the raw count, 3.3 times larger after the funded filter, and only 1.08 times larger once the funding threshold was set relative to each chain's own fee. Almost the whole headline gap was cost, not arrivals.
Is the gap between new and funded addresses useful on its own?
Yes, as a diagnostic rather than a level. A stable funded share suggests the composition of first appearances is unchanged. A funded share that collapses during a spike says the spike is addresses being referenced rather than used, which is the signature of bare address generation or a batched distribution.
What makes an address exist on a chain in the first place?
The answer differs by data model, which is why new-address definitions are not portable. On an account-model chain, an address becomes part of the state when it first receives a transaction or is otherwise touched, so existence is a recorded event with a timestamp. On an unspent output model, addresses are not stored as accounts at all; they appear only as conditions attached to outputs, so a first appearance is inferred from output history rather than read directly. The same phrase describes two different underlying facts.
Does one user create many addresses by design?
Routinely. Modern wallets derive a fresh address for each receipt from a single seed, which is the recommended practice for privacy and is invisible to an observer. A user who receives ten payments therefore generates ten new addresses. This is not evasion or manipulation; it is the default behaviour of common software. It means a new-address count reflects a mixture of genuine new participants and existing participants receiving payments, in a proportion that no external observer can determine.
Why does an exchange onboarding users spike the new-address count?
Because platforms that assign a unique deposit address per customer create addresses in bulk, sometimes in advance of demand. A single business decision can therefore add a large number of addresses in a short period with no corresponding growth in participants, and the reverse is true when a platform switches to a shared-address model. Because these events are large and undisclosed, an unexplained step in a new-address series is worth checking against exchange address sets before it is treated as growth.
What is address churn, and how does it distort a growth series?
Churn is the constant creation of addresses that are used briefly and abandoned, whether by wallets rotating receive addresses, by automated systems, or by incentive-seeking activity. A raw new-address count treats each of these as an arrival while never recording a departure, so the series measures gross creation against nothing. Pairing new addresses with a measure of how many remain active some period later converts a one-sided count into something closer to net growth.
Is a decline in new addresses evidence that growth stopped?
Not on its own. The count can fall because fewer participants arrived, because existing participants moved their activity to a rollup or a custodial platform where no base-layer address is created, because fees made small first transactions uneconomic, or because a large address-creating business changed how it operates. Each of these produces the same visible decline. Establishing which one applies requires evidence from outside the series, which is why this metric supports a conclusion rather than producing one.
References
These sources should be reviewed during editorial verification. They support data structures and methods, not the hypothetical conclusion. Provider formulas, chain rules, and APIs can change. Confirm current documentation before publication.
- Coin Metrics: New Addresses: One provider's published distinction between new and funded addresses.
- Bitcoin Developer Guide: Transactions: Inputs, outputs, and the change outputs that create addresses on every spend.
- Coin Metrics: Network Data Glossary: What an address is under UTXO and account ledger models.
- Ethereum.org: Technical Introduction: Accounts and state, and the conditions under which an account first exists.
- Ethereum.org: Gas and Fees: The fee mechanics that set the cost of putting an address on a chain.
- Dune: Address Labels: Labelling used to identify funders, factories, and campaign-related clusters.