How to Read Bitcoin On-Chain Accumulation Data: Holder Cohorts, Trend Scores and Exchange Reserves (2026 Guide)

In the first week of July 2026, two datasets told opposite stories about bitcoin. Spot ETFs had just closed their worst month on record, with $4.51 billion in June net outflows. At the same time, Glassnode reported that long-term holders — wallets that have held coins for at least 155 days — had flipped from distribution back to net accumulation, per CoinDesk's July 2 coverage. Price sat near a 21-month low while the oldest hands in the market were quietly buying. If you don't read on-chain data, weeks like that are just noise. This guide teaches you to read it.

Updated July 2026, this is an evergreen walkthrough of the four on-chain concepts that do most of the analytical work — holder cohorts, accumulation trend scores, exchange reserves and whale segmentation — plus the mistakes that trip up beginners and the caveats that trip up everyone else.

Concept 1: Long-term vs. short-term holders — the 155-day line

On-chain analytics firms split the bitcoin supply by the age of the coins, not the identity of the owner. The convention, popularized by Glassnode, is 155 days: coins that have not moved in 155 days or more belong to long-term holders (LTHs), everything younger to short-term holders (STHs). The threshold isn't arbitrary — statistically, coins that survive 155 days without moving become far less likely to ever be spent in a panic. LTHs are the market's ballast; STHs are its churn.

Why it matters: LTH behavior is one of the most reliable cycle signals in the dataset. LTHs historically distribute (sell into strength) near cycle tops and accumulate through capitulations. As of early July 2026, LTHs control roughly 78% of circulating supply, and their flip back to accumulation during a record ETF outflow month is exactly the kind of divergence this metric exists to surface: forced sellers exiting through the ETF door while patient capital absorbs the coins.

Concept 2: The Accumulation Trend Score — one number, many wallets

Glassnode's Accumulation Trend Score compresses aggregate wallet behavior into a value between 0 and 1. Readings near 1 mean the market's larger entities have been meaningfully adding to their positions over the past month; readings near 0 mean they have been distributing. The score is weighted by both entity size and the intensity of balance change, so a handful of mega-wallets adding aggressively can move it as much as a broad retail bid.

The real power comes from splitting the score by cohort. In early July 2026, the readings looked like this, per Glassnode data cited by BeInCrypto and CoinDesk: wallets holding under 1 BTC scored roughly 0.8–0.9 (near-maximum accumulation), entities in the 100–1,000 BTC band read similarly strong, while the largest cohort — wallets above 10,000 BTC — sat near a neutral 0.4–0.5. Translation: retail and mid-sized players were dip-buying with conviction, but the very largest entities had not yet committed. Analysts on the same data warned it was 'too early to call a full accumulation regime' — a nuance a single blended number would have hidden.

Concept 3: Exchange reserves and netflows — the supply on the shelf

Exchange reserves measure how many coins sit in wallets attributed to trading venues. Coins on exchanges are inventory available for immediate sale; coins withdrawn to self-custody are, statistically, being shelved. Sustained reserve declines mean supply is migrating from the liquid shelf to cold storage — a structural tailwind that plays out over months, not days.

As of July 2026, exchange reserves sit near six-year lows, and roughly 74% of circulating supply is classified as illiquid — held by entities that historically almost never sell. The companion metric, exchange netflow, gives the daily pulse: net inflows to exchanges flag potential sell pressure (coins moving toward the market), net outflows flag accumulation. Spikes matter more than levels — a single day of 30,000+ BTC flowing in has historically preceded volatility far more reliably than any slow drift.

Concept 4: Whale cohorts — and why 'whale' is the most abused word in crypto

Cohort analysis buckets entities by balance: shrimp (under 1 BTC), crab (1–10), fish (10–100), shark (100–1,000), whale (1,000–10,000) and humpback (over 10,000). Divergences between buckets are informative precisely because the cohorts have different information, time horizons and constraints. Mid-sized cohorts accumulating while mega-wallets stay neutral — July 2026's exact setup — historically marks late-capitulation phases, though it is no guarantee of a bottom.

Now the caveats, which matter more as the market institutionalizes. First, entity adjustment: naive wallet counts are meaningless because one entity can control thousands of addresses; serious providers cluster addresses into entities, but clustering is probabilistic. Second — and this is the big one in the ETF era — custodial distortion. When a spot ETF buys bitcoin, the coins land in an exchange-linked custodian wallet (often classified under Coinbase Custody). A retail investor buying IBIT shows up on-chain as a 'whale-sized' custodial movement, and ETF redemptions can masquerade as whale distribution. Always cross-check apparent whale activity against the daily ETF flow prints before drawing conclusions. Third, lost coins: a meaningful slice of 'LTH supply' is simply unspendable, inflating the ballast.

The supporting cast: cost-basis metrics

Two more tools round out the kit, both built on the idea of cost basis. Realized price is the average price at which every coin in circulation last moved — effectively the market's aggregate cost basis. When spot trades below the realized price of a cohort, that cohort is underwater in aggregate, and history shows capitulation risk concentrates there. Analysts track the short-term holder realized price especially closely: in downtrends it tends to act as overhead resistance, because recent buyers are eager to exit at break-even, while in uptrends it flips to support as dip-buyers defend their basis.

The companion ratio, MVRV (market value to realized value), divides market cap by realized cap to express how far the average coin sits in profit or loss. Extreme highs have historically flagged euphoria and cycle tops; readings below 1 mean the average holder is underwater — territory that has coincided with every major accumulation zone in bitcoin's history. Neither metric times entries, but both tell you where you are in the emotional cycle, which is most of what a long-horizon investor needs.

A worked example: reading July 2026 like an analyst

Put the four concepts together on this month's data and a coherent picture emerges. ETF outflows (a record $4.51 billion in June) represent price-insensitive redemption selling. LTHs flipped to accumulation — patient capital absorbing that supply. Exchange reserves at six-year lows say the absorbed coins are leaving the liquid shelf. And the cohort split — small and mid-sized wallets near maximum accumulation, humpbacks neutral — tells you who is doing the absorbing and who is still waiting. The honest synthesis: constructive but unconfirmed. The missing signal is the largest cohort joining the bid, and a sustained flip in ETF flows themselves.

That 'unconfirmed' verdict is the discipline on-chain data enforces. It rarely gives you a green light; it tells you which side of the market is under pressure and which is patient, and it forces you to wait for confluence instead of narrative.

Notice also what the worked example did not use: price predictions, chart patterns or sentiment surveys. That is deliberate. On-chain analysis answers a narrower question than technical analysis — not 'where is price going' but 'who owns the supply, at what cost basis, and what are they doing with it.' Kept in that lane, it is arguably the most falsifiable tool in crypto; dragged out of it, it becomes another narrative machine. The practitioners worth reading are the ones who state the cohort, the metric, the reading and the historical base rate — then stop.

Common mistakes to avoid

Four errors account for most bad on-chain analysis. One: single-metric storytelling — every metric has false positives, so demand confluence across at least three independent families (holder behavior, exchange flows, cohort trends). Two: treating address counts as people — always use entity-adjusted data. Three: ignoring the ETF layer — in 2026, roughly 1.3 million BTC sits inside U.S. spot ETFs, so custody wallets dominate raw flow charts; strip them out or be misled. Four: timeframe mismatch — reserves and LTH supply are monthly-to-quarterly signals; trading them on a daily chart is a category error.

Where to find the data

Glassnode Studio is the reference for holder cohorts, trend scores and liquid supply, with free tiers covering the basics. CryptoQuant and CoinGlass cover exchange reserves and derivatives context. For ETF flows — your custodial cross-check — the issuers publish daily creations and redemptions, aggregated by trackers like SoSoValue and The Block's data dashboards. Start with three charts and check them weekly: LTH supply, the cohort-split Accumulation Trend Score, and aggregate exchange netflow. That trio, cross-referenced with the daily ETF print, covers 80% of what professionals actually look at.

On-chain data will not tell you where bitcoin trades next week. What it will tell you — reliably, and years before it becomes consensus — is whether the asset is migrating toward patient hands or away from them. In July 2026, the ledger says the migration toward patience has resumed. Whether price agrees is the part no dashboard can promise.