On-chain data limitations: seven blind spots every beginner should know

Why do on-chain metrics need a warning label?#

On-chain metrics look precise: a chart with values to several decimal places, calculated from a public ledger. The underlying records really are exact. But most useful metrics need extra steps — grouping addresses, labelling exchanges, removing change, assuming that a movement is a trade — and each step adds judgement. This guide walks through the seven most common blind spots, so you can read any on-chain analysis with the right amount of scepticism.

From certain to unknown

1. Why is an address not the same as a person?#

Bitcoin was designed so that addresses are cheap and disposable. The whitepaper recommends a new key pair for each transaction so that payments are not linked to a common owner ^[1], and the developer guide encourages new addresses for change as well ^[3]. One careful user can therefore create dozens of addresses, while one exchange address can hold coins for thousands of customers.

The address count understates the people here by a factor of about 90, and on another day it could overstate them. Coin Metrics warns that where address creation and transactions are cheap or free, active addresses can be trivially forged ^[4].

2. How reliable is grouping addresses into owners?#

To get closer to people, providers cluster addresses into entities. The oldest clue comes from the whitepaper: a transaction that spends several inputs necessarily reveals that those inputs had the same owner ^[1]. Providers add further heuristics and machine-learning algorithms. Glassnode’s long- and short-term holder metrics, for example, are entity-adjusted using heuristics and clustering ^[5].

Heuristics can be fooled — for example by transactions that deliberately combine several people’s inputs, a technique the Bitcoin developer guide mentions as a way to make tracking difficult ^[3] — and they change as new links are discovered. Glassnode lists clustering for entity adjustment as one reason historical values can change after the fact ^[6]. Its API reference adds a nuance for entity-based series: “its established history is stable, but recent data points may revise as clustering improves” ^[7].

3. Why are exchange figures only lower bounds?#

Nothing on-chain marks an address as an exchange’s. Providers rely on addresses exchanges disclose, public tags and clustering, and Glassnode says its exchange balances can largely be considered lower bounds of the true balance ^[2]. Coverage depends on what each exchange chooses to disclose, and internal reshuffles can briefly look like large inflows or outflows. Read more in how to read exchange flows.

4. How do lost coins distort the picture?#

Coins whose keys are lost never move again, but nothing on-chain says they are lost. Coin Metrics notes that a significant fraction of Bitcoin has not moved since 2010 and can be presumed lost ^[8], and Glassnode’s HODL waves guide says coins older than five years are very rarely spent and are often assumed to be lost or out of circulation ^[9]. These coins keep counting in supply, keep ageing in long-term holder metrics, and keep their ancient prices in realized price.

Nobody can see which old coins are truly lost, so every provider has to choose between counting them and guessing. See realized price and cost basis.

5. What happens off-chain?#

A great deal of crypto activity never reaches the blockchain. Trades between customers on the same exchange are entries in the exchange’s own database; Glassnode explicitly keeps off-chain spot and futures volume out of its on-chain exchange metrics ^[2]. Payment channels such as the Lightning Network move value off-blockchain ^[10]; only channel openings, closings and disputes settle on the Bitcoin chain ^[11]. A quiet on-chain chart can sit alongside very busy off-chain markets.

6. Why do past values sometimes change?#

Many people assume that blockchain history, once written, means metric history is fixed too. It is not. The latest data point can change while late blocks are still being included; Glassnode recommends waiting about two hours for a Bitcoin data point to fully settle ^[12]. Older values can change when new exchange addresses or entity links are found ^[6].

Any chart you saved last month would now disagree with the provider’s current history. Point-in-time datasets exist to preserve values as they were first published ^[6].

7. Do the cleaning rules change the answer?#

Yes. Coin Metrics’ adjusted transfer value, for instance, drops outputs spent within an hour, discounts outputs returning to an input address, and ignores an output when its precision suggests it is change ^[13]. These are sensible rules, but they are rules of thumb. Another provider choosing different rules will publish a different number for what sounds like the same metric — and chains with full privacy, such as Monero, are not covered by address metrics at all ^[4].

What mistakes do beginners make here?#

• Treating chart precision as accuracy

  A value shown to six digits can still rest on estimated labels and clustering.

• Assuming history never changes

  Providers revise past values when they discover new exchange addresses or entity links.

• Forgetting the coins nobody can spend

  Lost coins still sit in supply, holder cohorts and cost-basis calculations.

• Reading silence as inactivity

  Low on-chain activity can coincide with heavy trading inside exchanges or on payment channels.

Frequently asked questions#

The bottom line#

On-chain data is exact at the bottom layer and increasingly uncertain as analysts add labels, clusters and assumptions on top. None of that makes it useless; it makes it something to read with method notes open, with trends preferred over single values, and without pretending to know motives.

Put these limits to work on real metrics: start with active addresses and exchange netflow, and see how coin age is used in long-term vs short-term holders.

Sources#