Token holder concentration: wallets are not owners

Token holder concentration measures how much of one token's defined supply is held by its largest addresses or attributed entities. It does not directly measure unique owners: one custodian can represent many customers, while one owner can control many addresses. Report the raw ranking, the exact supply denominator and any evidenced exclusions together. A custody-adjusted address share still leaves beneficial ownership uncertain.

A headline such as “the top five hold 77%” leaves several questions unanswered. Five addresses or five people? Of recorded supply, circulating supply or a filtered subset? Does the list include a burn sink? Answer those questions before interpreting the percentage as ownership concentration.

Define the snapshot before reading the ranking

Start with one chain, one token contract and one block. A ticker symbol alone does not identify the ledger being measured. Record the balance units, decimal conversion, extraction time and whether the source supplies a complete distribution or only its largest rows.

For example, Etherscan's top-holder endpoint returns ERC-20 holder addresses, quantities and address types. That is an address-balance dataset. Its documented request limit of up to 1,000 holders also matters: a largest-holder extract is not necessarily a complete ledger from which to derive the denominator. Its documented request parameters do not include a block number. Use a block-aware source for a historical comparison, and do not pair a latest-holder list with an earlier supply count.

Choose the ranking rule explicitly. “Top ten addresses” selects a fixed count; “top 1% of addresses” selects a changing count as the holder population changes. Coin Metrics documents separate top-percentage and richest-100-address measures. Its native-asset methodology also warns that one individual can control many addresses, so the distribution does not directly reveal individuals' holdings. Do not silently apply one provider's definitions to another provider's token chart.

For a declared included universe, divide the sum of its N largest balances by its total included balance. After exclusions, re-rank the remaining addresses and recalculate both parts. Removing rows from the numerator while keeping an unexplained denominator produces a different statistic.

Recalculate one complete holder ledger

This original example is hypothetical. A fictional token has 100 million recorded tokens at one invented block. Its complete ledger contains seventeen addresses. All quantities below are millions of tokens; the ten small addresses each hold exactly one million.

The fixture stipulates that E is a pooled custodian, T is a treasury and S is a verified inaccessible sink. S still has a balance included in this fictional contract's reported total supply: its earlier receipt did not execute a supply-reducing burn. A through D and the ten small addresses remain unattributed. These are supplied facts of the example, not conclusions inferred from transaction patterns.

Keep three views of the same balances. The second removes only S. The third removes S and E. Treasury T stays included in all three: being labeled does not itself justify an exclusion.

The raw top three are E, S and T: 24 + 20 + 15 = 59 million. Once S is excluded, the top three become E, T and A: 24 + 15 + 10 = 49 million. Their share is 49 / 80 × 100 = 61.25%.

Removing both E and S leaves T, A and B as the top three. Their combined 33 million is 33 / 56 × 100 = 58.93%. The five remaining large addresses hold 46 million, or 82.14% of that selected universe. Calculate from unrounded balances, then round the displayed percentages.

Interpret the adjustment without losing the excluded holdings

The raw 59% does not establish that three investors own 59% of economically usable tokens. One row is inaccessible under the fixture; another aggregates unknown customers. Nevertheless, the address balances themselves are valid observations.

Excluding the sink reduces the top-three balance from 59 to 49 million, yet its share rises to 61.25%. The denominator shrinks faster. A higher adjusted percentage does not require an acquisition by a large holder.

Excluding custody creates a narrower question: how concentrated are balances outside this identified custodian and sink? It does not reveal ownership of the whole token. Keep E's 24 million visible alongside the adjusted result. That is 30% of the non-sink supply, and its customers might include holders who also control addresses elsewhere in the ledger.

Call the 56 million denominator “outside the identified sink and custodian.” Calling it circulating supply or free float would add assumptions about treasury restrictions and other holdings that this example has not established. Likewise, an unattributed address is not automatically an independent investor, a team wallet or a readily available seller.

The treasury's inclusion preserves a material known balance. If a separate question requires excluding it, publish another explicitly named view and its exclusion ledger. A cleaner-looking percentage is not a reason to change the universe.

Verify custody and burns separately

Custody attribution concerns who operates an address and what its balance represents. Coinbase's Prime Custody documentation gives a concrete example: that portfolio structure uses shared operational wallets without per-client on-chain segregation. This supports the possibility of many customers behind a visible balance; it does not establish that every custody product uses the same arrangement.

Save the evidence behind each label. An official address disclosure, verified contract documentation and a third-party tag have different provenance. Glassnode's exchange-data notice explains that its attribution combines verified addresses, external labels and clustering, and that exchange coverage can remain incomplete. A provider's known-exchange filter therefore does not guarantee that every remaining address is outside custody.

Burn treatment requires a separate supply reconciliation. In the documented OpenZeppelin ERC-20 implementation, _burn destroys tokens and lowers total supply. A transfer that merely leaves tokens at an address can have different accounting. Inspect the actual token implementation and reported balances before deciding whether an adjustment is needed.

In the hypothetical fixture, S's 20 million remains in the 100 million supply, so the non-sink denominator is 80 million. If a subsequent actual burn destroys five million from another address and lowers reported supply to 95 million, the unchanged sink still contributes 20 million. The new non-sink denominator is 75 million. Subtracting the five-million burn again would incorrectly produce 70 million.

A “burn” nickname or unfamiliar address is insufficient evidence for removal. Record why funds are considered inaccessible, whether supply already reflects their destruction and what uncertainty remains. If that evidence is missing, retain the row and show an explicitly conditional alternative rather than treating the exclusion as settled.

Stress-test hidden common control

Now introduce a separate hypothetical assumption: A and C share one controller. Their combined balance is 10 + 7 = 17 million. Within the 56 million universe, that controller's share becomes 30.36%, exceeding T's 26.79%.

After merging A and C, leaving every other address separate and re-ranking, the three largest rows hold 17 + 15 + 8 = 40 million, or 71.43%. The address-based top-three figure was 58.93%. The balances and supply did not change; the assumed control mapping changed. Other ownership relationships remain unknown.

This is a sensitivity calculation, not evidence that A and C are linked. Similar transfer timing, a common distributor or funding from the same exchange can suggest further investigation without proving common ownership. Preserve the observation separately from the proposed identity relationship.

Glassnode describes its entity clusters as estimates based on heuristics and proprietary methods, with recent data potentially revised as clustering improves. Treat an entity-adjusted reading as a documented model output. It still cannot automatically split a custodian into customers or identify legal beneficial owners.

The active-addresses versus users guide explains the broader identity problem. Here the practical consequence is specific: neither seventeen address rows nor fifteen rows after exclusions establishes the number of independent owners.

Build a review record that another reader can reproduce

Use this worksheet before comparing tokens or reporting a change over time. A percentage without these choices is difficult to audit.

For a time comparison, apply the same classification version to both snapshots. If an address is newly identified as a custodian, separate that label change from actual token movements. Otherwise an apparent reduction in concentration may be a research update rather than a redistribution.

Keep holder concentration separate from portfolio concentration measured with HHI. Portfolio HHI squares one investor's weights across assets. A holder HHI would square one token's shares across addresses or attributed entities. The arithmetic can match while the objects being measured differ; splitting an owner's balance into addresses can change the latter without changing that owner's assets.

Concentration also cannot estimate an executable sale by itself. The market-cap versus liquidity comparison addresses the depth needed to absorb an order. A large balance does not reveal its owner's intention, and excluding a pool or custodian does not erase the associated trading or custody dependency.

Nor does an evenly spread holder table validate reported turnover. Use the separate wash-trading volume checks when the claim concerns trading activity, retaining the same distinction between a suspicious pattern and proof of common ownership.

For the next holder chart you use, save the raw balances, one justified exclusion ledger and the recomputed shares. Write the conclusion in its actual unit: “These addresses hold this percentage of this defined supply at this block.” Add an owner-level claim only when you can show the attribution evidence that supports it.