Public does not mean clear
A blockchain is a complete record that anyone can read, which gives on-chain data an air of objectivity. The record itself is accurate: a confirmed transaction happened as written. The trouble starts when that record is converted into statements about people, such as “long-term holders are accumulating” or “whales are selling”. Those statements need interpretation, and interpretation introduces error.
This article lists the most common sources of that error. It complements the overview in the introduction to on-chain analysis by focusing on what goes wrong.
Labels and entity clustering
The ledger shows addresses, not owners. To describe behaviour by group, such as exchanges, miners, funds or individuals, analytics firms assign labels.
A widely used approach is clustering based on the common-input heuristic: if several addresses are spent together as inputs to one transaction, they are presumed to be controlled by the same party. This often works. It can also fail, for example when users combine coins in joint transactions designed to break the assumption, or when a service pools funds from many customers. A wrong merge can combine two unrelated owners into one apparent giant, while a missed link can split one owner into many small ones.
Other labels come from sources such as deposit tests, public announcements, leaked lists or inference from behaviour. These vary in quality, they are rarely published in full, and they go stale. An address labelled as belonging to an exchange last year might have changed hands since. Two providers can disagree about the same address. For this reason, any series built on labels, such as exchange flows or whale wallets, deserves extra caution.
Custody shifts
Coins can change their on-chain appearance without anyone’s economic position changing. A fund might move its holdings from one custodian to another. An exchange might move customer funds into new wallets. A person might consolidate dozens of small outputs into a single one.
In each case the chain records large movements. Metrics that depend on coin age may show older coins “waking up”, and balance-based metrics may show one group shrinking while another grows. The change is real on the ledger, but the story that it signals buying or selling can be wrong. Moves between wallet types, for example from self-custody to a regulated custodian, can alter measured supply in categories while underlying ownership stays the same.
The off-chain blind spot
Much of what moves prices happens off the chain. Trades inside an exchange are settled in the exchange’s own books. Futures, options and perpetual contracts, the subject of crypto derivatives, are priced and margined elsewhere. Funds that hold coins through a custodian may buy and sell exposure without a visible transfer. On-chain data therefore represents a biased sample of activity: the portion that touches the ledger.
A strong reading drawn from that sample can mislead. For example, quiet on-chain activity might coincide with very heavy trading on exchanges.
Noise and artefacts
Not every transaction reflects genuine economic activity. Common sources of noise include:
- Self-transfers and change outputs. Counted as activity, though no value changes owner.
- Dust and spam. Tiny amounts sent to many addresses, inflating counts.
- Batching. One transaction paying many recipients can look like many separate events or a single one, depending on how the provider counts.
- Automated scripts. Bots can create large numbers of transactions cheaply.
- Data revisions. Providers sometimes restate history when they improve labels or fix bugs, so a chart saved last month may not match today’s version.
Noise tends to be worse at daily resolution. Longer windows smooth some of it, but can also hide real changes.
Bias in how results are chosen
Even a perfect dataset can mislead through the way analysts select and present it.
Survivorship and hindsight. Metrics that “called” a past turning point are celebrated; those that did not are forgotten. With hundreds of series and many thresholds, some will fit any past price chart by chance.
Small samples. Bitcoin has experienced only a few large cycles. Conclusions drawn from them carry little statistical weight, and the same warning appears in the article on evaluating predictions.
Confirmation bias. It is easy to read an ambiguous chart as agreeing with an existing opinion. A rise in exchange balances can be framed as bearish or bullish with equal ease.
Changing meaning. Fee habits, wallet defaults, scaling layers and new transaction types evolve. A metric calibrated on one era may not carry over.
Commercial incentives. Dashboards and newsletters compete for attention, and dramatic interpretations spread more easily than cautious ones. Treat confident predictions from any source with scepticism.
A short checklist
Before leaning on an on-chain claim, consider asking:
- Is this a direct ledger observation or an inferred label?
- Who produced it, how is it defined, and can the method be inspected?
- Could a custody move, internal transfer or batching event explain the same picture?
- What share of relevant activity is off-chain?
- How many observations does the pattern rest on?
- What would a sceptic say about it?
The goal is not to dismiss on-chain data. It is to hold conclusions loosely. Cryptocurrency prices are volatile, past patterns do not predict future results, and people can lose everything they invest. Nothing here is financial advice.