The pitch for on-chain analytics is seductive. Every Bitcoin transaction that has ever settled is sitting in a public ledger you can download and verify yourself. No filings, no quarterly delay, no company deciding what to disclose. When a chart says "whales are accumulating," it feels less like an opinion and more like a fact read off an instrument.
It is worth separating two things that get sold as one. The ledger is genuinely public and genuinely verifiable. The dashboard sitting on top of it is a stack of assumptions about who owns what, and the assumptions are doing more work than most people realize.
The chain records addresses, not people
A blockchain does not know what an exchange is. It records that value moved from one address to another. Every "exchange inflow," "whale wallet," and "long-term holder" label is the output of a clustering step that guesses which addresses belong to the same entity, followed by an attribution step that guesses which entity that is.
The workhorse guess is the common-input-ownership heuristic: if several addresses appear as inputs to the same transaction, assume one party controls all of them. That is usually true and specifically false for CoinJoin transactions, which are built by multiple independent parties on purpose. The second guess is change-address detection, which assumes the freshly created output in a two-output transaction is the sender's change. When that guess misfires, the error does not stay local. A wrongly linked change output merges two clusters, and the label on one silently propagates across the other. Academic work on forensic clustering keeps finding the same failure mode, and commercial tools inherit it.
Exchange netflow has a plainer version of the problem. A billion dollars moving into a known exchange cluster reads as "holders are preparing to sell." It also reads as an exchange rotating funds between hot and cold wallets, a custodian migrating clients, or an ETF authorized participant doing a creation. Same on-chain footprint, opposite meaning.
The thresholds came from a different market
MVRV is the cleanest example of a metric whose math is sound and whose folklore is stale. It divides market cap by realized cap, where realized cap values every coin at the price it last moved rather than today's price. That makes it a rough aggregate cost basis for the whole network.
The arithmetic is simple:
market cap = circulating supply × current price
realized cap = Σ (each UTXO × price when it last moved)
MVRV = market cap / realized capIf a network holds 20 million coins with an aggregate cost basis of $800 billion and a market cap of $1.6 trillion, MVRV is 2.0, meaning the average coin is sitting at roughly a 100% unrealized gain. That is a real, useful statement about positioning.
What is not real is the received wisdom that MVRV above 3.5 marks a top. That number is an empirical artifact of two prior cycles, and the market structure underneath it changed. Spot Bitcoin ETFs, approved in January 2024, introduced mechanically rebalancing allocators that the indicator was never calibrated on. This cycle's MVRV peak came in around 2.52 in January 2025, well short of the historical trigger. Anyone waiting for 3.5 was waiting on a threshold from a market that had already been replaced.
The failure runs the other direction too. MVRV sat below 1.0 from June to November 2022, five months during which price fell roughly another 35%. "Undervalued" is not a timing signal.
What the data is actually good for
There is a category of on-chain fact that requires no clustering and no calibration: total supply, issuance schedule, fees paid, block space consumed, the state of a smart contract, the collateral sitting in a lending protocol. These are read directly off the ledger and they are as close to ground truth as finance gets. Nobody has to trust a labeling vendor to know how much stablecoin supply exists or what a DeFi pool's reserves are right now.
The useful mental split is between measurement and interpretation. Measurement is auditable. Interpretation is a model with a vendor's assumptions baked in, and different vendors will hand you different numbers for the same "exchange balance" because they cluster differently.
If you use this data, the practical habits are unglamorous. Find out whether a metric is measured or inferred before you weight it. Check whether the thresholds attached to it come from three cycles or from thirty years. Require several independent signals to agree rather than acting on one crossing a line. And treat a metric that has been publicly famous for years as one that has already been traded against.
If you want to go deeper, the most useful reading is not another metrics guide. It is the academic work on clustering itself, like the 2025 USENIX paper Ghost Clusters on how illicit-service attribution goes wrong, alongside CryptoQuant's own writeup of the heuristics their product depends on. Both are more candid about the error bars than any dashboard will be.
On-chain data is the most transparent dataset in finance. That transparency is a property of the raw ledger, not of the confident label somebody painted on top of it.