The 72% Mirage: What Bitcoin's Long-Term Holder Realized Profit Number Actually Measures
A single number moved through crypto research feeds this week: Bitcoin's long-term holders are realizing profits at roughly 72%, down sharply from a cycle peak of 350%. The framing writes itself. Gentle profit-taking. No distribution top in sight. The patient cohort still patient.
I spent two hours pulling that number apart, and I still cannot tell you precisely what it measures. That is not a rhetorical flourish. It is a forensic finding. Depending on which of three defensible definitions the author used, "72%" describes three different markets: a modest average markup on coins that were spent, a ratio of realized profit to realized loss, or the share of long-term-holder supply currently sitting in unrealized gain. These are not near-synonyms. One of them supports the "we are mid-cycle" reading. Another supports the "we are much closer to the top than the chart admits" reading. A third is close to worthless as a timing signal because it barely moves.
The headline number is precise. The precision is the problem.
Context: The Mechanics Behind a Cohort Metric
To understand why 72% is ambiguous, you have to understand what a realized-profit metric actually is at the protocol level.
Bitcoin's ledger does not store balances. It stores unspent transaction outputs β UTXOs. Every coin that exists is an output created at some block height with some value. When that output is later consumed as an input to a new transaction, the protocol records that it was spent. The difference between the price at which the output was created and the price at which it was spent is the realized profit or loss on that specific coin.
This is the foundation of the entire on-chain profitability family. SOPR β spent output profit ratio β is the ratio of the value of spent outputs to the value at which those outputs were created. When SOPR is above 1, the average spent coin moved at a profit. MVRV β market value to realized value β compares the aggregate market cap to the aggregate cost basis of all coins, effectively the network's average paper gain or loss. Realized profit, as a raw quantity, sums the dollar gains across all coins spent in a window.
Now bring in the cohort. Long-term holders, by the industry-standard convention popularized by Glassnode, are addresses whose coins have not moved for more than 155 days. The number is arbitrary but deliberately chosen: it sits at a local optimum that separates coins that have survived one full short-term rotation from coins that have not. It is a heuristic, not a law of nature. Short-term holders are everything below that line.
The analyst here is Darkfost, a name active on the CryptoQuant author platform. The claim, as circulated, is that LTH realized profit has cooled from a 350% cycle peak β anchored by the author to December 2024, when Bitcoin traded near six figures β down to roughly 72% today. The interpretation: holders are taking profits gently, similar in intensity to a prior bear-market reading, and therefore are more likely to keep holding than to dump.
That is a coherent story. It is also a story built on a number whose denominator is invisible. And in on-chain analysis, the denominator is where errors live.
Core: Three Definitions, Three Different Markets
Zero knowledge isn't magic; it's math you can verify. The same discipline applies here. If you cannot verify what a percentage is a percentage of, you cannot verify the conclusion drawn from it.
Consider the possibilities for "72%" as an LTH realized-profit figure.
Definition A β average markup on spent coins. Here 72% means the average LTH coin that was spent during the measurement window moved at 1.72x its acquisition price. If that is the true reading, the average cost basis of recently spent LTH coins is roughly 58% of the current spot price. That tells you something concrete: long-term holders who are selling are doing so at a healthy but not spectacular gain. It says nothing directly about how many are selling, only at what margin.
Definition B β realized profit relative to some normalized scale. Here 72% is a point on a bounded oscillator, where 100% might represent a historical extreme or a full reset of the ratio between realized profit and realized loss. This is the most likely reading, and it is the most dangerous to quote without the axis labels.
Definition C β share of LTH supply in profit. Here 72% means 72% of long-term-holder coins are above their cost basis. This metric is almost always higher than 90% in bull markets and almost always above 70% even in deep drawdowns. A reading of 72% would actually be a bearish signal, implying 28% of the most patient supply is underwater β something you typically see only near capitulation. Framed this way, the same number flips from reassuring to alarming.
Three definitions. One cheerful headline. Opposite conclusions at the edges.
The tell is the comparison point: 350%. No mainstream profitability metric trades at 350% as a percentage in the conventional sense, because probabilities and shares are bounded at 100%. A 350% figure only makes sense as a ratio that can exceed unity β realized profit divided by realized loss, or a normalized MVRV-style oscillator expressed in percentage points rather than multiples. This is the fingerprint of a custom or semi-custom metric, almost certainly the author's own construction or a platform-specific variant.
And the AMM model hides its truth in the invariant. So does this: the metric hides its truth in the normalizer. Darkfost's number is a ratio, and ratios can be moved by either the numerator or the denominator. If realized losses spiked β say, from a capitulation flush β the ratio would collapse even if realized profits barely changed. A falling ratio would then be read as "calmer selling," when the mechanical truth is "more pain on the other side of the ledger." That is a catastrophic misread, and it is invisible in a single percentage.
The lag problem
There is a second structural issue that no definition fixes. Realized profit is a behavioral derivative. It describes spending that already happened. A UTXO must be consumed before its profit is realized and recorded. By construction, the metric is backward-looking. It tells you what holders did, not what they will do.
I have modeled this class of indicator before. In a Python simulation of cohort spending behavior, I found that realized-profit oscillators lag price tops by a variable interval that ranges from a few days to several weeks depending on the aggressiveness of the distribution. The lag is not constant. In fast tops, holders spend into strength and the metric peaks slightly after price. In slow, grinding tops, the metric can roll over before price does. Either way, you cannot treat a single reading as a forecast. You treat it as a recency-weighted description.
When a research note presents a cooling metric next to a bullish conclusion, it is quietly converting a description into a prediction. That conversion is the analytical sin.
The December 2024 anchor and why it matters
The author chose December 2024 as the 350% reference. That choice is not neutral. It implies December 2024 was the local peak of LTH distribution intensity β a period when long-term holders were aggressively turning paper gains into realized ones. If that is the peak, then today's 72% is roughly one-fifth of peak intensity. That ratio β one-fifth β is the actual claim. Not "72% is low," but "today is one-fifth as intense as the recent high."
Now stress-test the anchor. Suppose the real distribution peak was not December 2024 but some earlier window, or suppose the metric's normalizer changed between then and now. The whole comparative structure collapses. Articles that pick a single comparison point and do not test alternatives are committing a methodological error that is easy to hide and expensive to inherit. If you swapped the anchor for the same phase of the previous cycle, the ratio could read completely differently β higher or lower. The author did not run that test. That is a gap, not a verdict.
The deeper issue is scale. Conventional MVRV-style oscillators peak in the three-to-four multiple range β call it 300% to 400% in percentage terms β which is why 350% is superficially plausible. But that plausibility is exactly what makes it dangerous. A reader sees a number that matches the historical shape of a known indicator family and assumes the methodology is known too. It is not. The resemblance is doing the persuasive work, not the math.
Data provenance: the missing audit trail
The single most important omission is data sourcing. The note does not state whether the underlying chain data comes from Glassnode, CryptoQuant, or a self-built indexer. It does not disclose the exact LTH threshold used (155 days is standard, but it is not universal), nor the exchange-adjustment methodology, nor how it handles coins moved internally by custodians.
I do not trust a number I cannot reproduce. Give me the vendor, the threshold, the adjustment logic, and I can rebuild the pipeline and check the output. Without those four things, the number is an assertion wearing the costume of a measurement. This is the same discipline I applied compiling Solidity on a local testnet years ago: if a claim about behavior cannot be reproduced from the inputs, it is not evidence. It is a rumor with a decimal point.
The ETF blind spot
Here is where the on-chain methodology collides with the structural change of this cycle. Long-term holder metrics track coins on the Bitcoin ledger. They do not track Bitcoin held inside a spot ETF's custody wallet that never moves on-chain in a way that registers as an age-cohort spend, and they certainly do not track the beneficial owners behind those shares.
When an institution buys ETF shares, the underlying BTC is often cold-stored and rarely rebalanced. From the chain's perspective, that supply looks like the most patient coin in existence β effectively a permanent long-term holder that never sells and never registers a realized profit. Meanwhile, the humans who actually control that exposure can exit by selling shares on a stock exchange, an event that produces zero on-chain footprint.
This means the traditional LTH cohort is now a blend: genuine self-custodied long-term holders, plus a growing mass of custodied institutional supply that is inert by design. A cooling LTH realized-profit reading may simply reflect that more of the long-term supply has migrated into custodial wallets that mechanically cannot realize profit on-chain. The metric would register calm. The market could be selling. Two different things, one misleading number.
I flagged a version of this risk when I compared institutional custody architectures against open-source multisig alternatives during the 2024 ETF due-diligence cycle. The centralization was legible. What was less legible β and what this metric exposes β is the analytical blind spot: as more supply moves behind custodial walls, on-chain cohort indicators lose resolution precisely when the market most wants them.
What the supply structure actually implies
Strip away the ambiguity and one structural fact survives. Bitcoin's supply is rigid. Twenty-one million, halving every four years, block rewards down to 3.125 BTC since April 2024. There is no emission schedule to dilute holders, no unlock cliff, no governance token vesting. Whatever selling pressure exists comes from humans choosing to spend coins they already own.
That changes the meaning of "gentle profit-taking." In an inflationary token, a cooling distribution metric still sits on top of a structural subsidy that continuously pressures price. In Bitcoin, a cooling distribution metric sits on top of fixed supply. The pressure is purely behavioral and therefore reversible in either direction within weeks. This is why I read the 72% figure as a temperature reading, not a pressure gauge. Temperature can drop for a day and spike the next. Pressure builds and releases with structure.
And here is the asymmetry most readers miss. A high LTH supply share is often celebrated as bullish β coins locked away, float shrinking. But low float is a coiled spring. The same coins that refuse to sell today become the overhead supply that floods the market the moment a psychological trigger is hit. Suppressed selling is not absent selling. It is deferred selling. A metric that reads "calm" is, in the strictest reading, a measure of how much future selling has been postponed.
The Contrarian Angle: The Comfort of a Single Number Isn't the Number's Fault. It's Yours.
The consensus reading of this data point is that it validates the bull case β holders holding, distribution muted, cycle intact. I want to invert that.
The strongest argument against the bullish interpretation is not that 72% might be the wrong definition. It is that the narrative "gentle selling means we are not at the top" is pro-cyclical by construction. It is exactly the story a market wants to hear during a high-level consolidation, and it is exactly the story that has historically preceded the sharpest reversals. Long-term holder distribution does not announce itself. It begins with metrics that read calm, then accelerates over two to three weeks into a print that reads climactic. By the time the cohort metric screams, the top is already history.
There is a further irony buried in the source framing. It compares today's intensity to "levels similar to the previous bear market." Read that twice. A bear-market intensity reading on a distribution metric means selling pressure is as low as it was during a period of apathy and capitulation. You can file that under bullish β coiled spring, quiet accumulation β or you can file it under warning β the market has gone quiet right before a directional resolution whose direction is unknown. The source picked the bullish filing. The number does not tell you which filing is correct.
I also want to name the incentive. Ask who needs this narrative and when. A single-author, single-source on-chain indicator circulating during a consolidation phase performs a specific function: it supplies certainty to a market that is starved for it. That does not make the author dishonest. Darkfost may be entirely rigorous. But the demand-side pull toward reassuring, precisely-numbered narratives is real, and it should raise your discount rate on any reading that arrives exactly when reassurance is most wanted. The propagation timing of a data point is itself a signal.
The final contrarian point concerns metric decay. If ETF and custodial migration continue, the top signal embedded in every LTH cohort indicator will dull. The cycles will still turn. The indicators will turn later, or faintly, or not at all. A tool that worked because it captured the behavior of self-custodied holders is being progressively starved of the exact behavior it was built to measure. Traders who keep trusting it without adjusting for this structural drift will be reading yesterday's map on tomorrow's terrain.
Takeaway: Treat It as One Input, Downgrade It Until Verified
The correct use of this data point is embarrassingly modest. It is one line item inside a cross-verification framework, not a standalone thesis. Put it beside Glassnode's and CryptoQuant's official LTH metrics, beside funding rates, beside exchange net-flows, beside ETF creation data, beside short-term-holder supply share. If all of them agree, you have a signal. If only one of them speaks, you have a headline.
Until the definition, the vendor, the threshold, and the exchange-adjustment logic are disclosed, the honest move is to downgrade the entire reading by one confidence tier and treat the 72% as a direction, not a value. Direction says selling is not currently aggressive. Value would require proof, and there is no proof on the page.
The question worth carrying forward is not whether 72% is high or low. It is whether the indicator family that has organized cycle-top debates for three cycles can survive the migration of its subject matter off-chain. If the most patient coins are increasingly the ones that never move, then the metric designed to watch patience will be watching an emptying room. The number looked reassuring this week. The silence behind it is what I am watching next.