Active addresses: define the activity before counting
Count active addresses only after defining the action that makes an address active. An address that sends a transaction, receives a token, calls an application, or signs an offchain message belongs to a different measurement universe. Address growth should not be casually relabeled as growth in people or paying customers.
Ethereum’s block explorer documentation describes the kinds of chain information explorers present. Explorer observations can support activity checks, while the analyst still needs to define the population, action, deduplication rule, and observation interval.
Write the activity contract
Choose one primary action tied to the research question. For a hypothetical application adoption study, successful user-initiated interactions with specified contracts may be more relevant than every incoming transfer. Record included contracts, transaction status, network, and whether indirect interactions through routers count.
Define the unit before building the chart. Unique sending addresses per day, unique addresses over a month, and the sum of daily unique addresses are not interchangeable. A single address appearing on thirty days belongs once in a monthly unique count but thirty times in the sum of daily counts.
| Definition item | Decision to write down |
|---|---|
| Population | Network and included application contracts |
| Qualifying action | Successful call, transfer, or another explicit event |
| Identity unit | Address, proposed cluster, or independently verified user |
| Window | Exact boundaries and timezone |
| Exclusions | System contracts, known operational activity, test events |
| Sensitivity | Result with and without uncertain activity |
Avoid silently applying a filtering rule because it makes the growth chart smoother. The rule must follow the research question and be disclosed.
Hypothetical campaign cohort
Assume a fictional application has 1,000 qualifying addresses in week one and 1,800 in week two. Of the second week’s addresses, 900 participate only in a one-time incentive claim. Under an all-qualifying-actions definition, growth is real at the address level. Under a recurring application-use question, the claim campaign may be a separate cohort.
Split the analysis into returning addresses, new noncampaign addresses, and campaign-only addresses using explicit rules. Do not declare the campaign-only group bots without evidence. One person may control several addresses, and several people may use a shared service address. Neither relationship can be resolved merely by counting identifiers.
A useful follow-up asks how many week-two addresses perform the specified action again in a later defined window. This measures address-level recurrence, not customer retention unless a verified user mapping exists. All counts in this example are hypothetical.
Check denominators and coverage
Reproduce one day’s count from underlying events before relying on an aggregate series. Check whether failed transactions are included and whether contract-generated activity is attributed to a user or to an intermediary. Inspect the largest apparent bursts for contract coverage changes and repeated operational patterns.
Then compare a narrow and a broad definition. A narrow series might include only a specified successful action; a broad series might include all application interactions. If the conclusion reverses between them, the note should highlight the dependence rather than selecting one series without explanation.
For cohort analysis, preserve first-observed dates within the available history and mark left-censoring. An address first seen in the research dataset may have used the application before the dataset begins. Calling every such address a new user would overstate what the evidence establishes.
Limits and common mistakes
Address counts cannot independently establish unique people, economic value, satisfaction, or genuine organic demand. Filtering suspected automation is especially uncertain without a disclosed and tested method. A public research note should describe patterns and sensitivity rather than claiming definitive identity classification.
The method is unsuitable for precise person-level adoption claims when no identity mapping exists. Common mistakes include adding daily uniques into a monthly unique total, counting unsolicited recipients as engaged users, treating all campaign activity as organic, and assuming every unusual pattern is fraudulent. The final record should retain both the count and the definition that produced it.
FAQ
Are active addresses the same as active users?
No. They are identifiers satisfying a stated activity rule. Mapping them to people or customers requires additional evidence.
Should incentive-related addresses be removed from the analysis?
Usually preserve them as a separately defined cohort. Exclusion may be appropriate for a particular question, but disclose the rule and show its effect.