BlockShield/Metrics/Bitcoin Exchange Reserves

Bitcoin Exchange Reserves

How BlockShield builds an exchange's Bitcoin address set by common-input clustering, why confidence is published, and why an all-failed sweep is refused.

Drawn on the Smart Money tabAPI /api/v1/market/liquidity/exchange-flowsmempool.space and blockstream.info (keyless)the curated exchange label seed set

What it measures

The Bitcoin balance held across the address set an exchange controls, and how that balance changed over a window — the difference between 'a whale moved 500 BTC' and 'this exchange's reserve fell by 9,000 BTC over a month'.

Those are different questions. Per-transaction monitoring cannot answer the second one: it needs the SET of addresses an entity controls and a balance series over that set.

How it is built

The address set is built by common-input-ownership clustering. When a transaction spends outputs from several addresses at once, whoever signed it held the keys to all of them, so starting from curated exchange labels the walk collects co-spent addresses outward.

That is the standard heuristic and it IS a heuristic. Transactions that deliberately combine inputs from unrelated parties break it; the obvious cases are skipped, the residual risk is that a cluster over-collects, and a confidence figure is published for that reason. The curated seed set stays authoritative.

NULL is not zero at the write layer. During a two-host provider outage an empty sum was once stored as three exchanges' reserves: a completeness flag recorded the failure, but the VALUE beside it was a lie no later reader could tell from a real zero, and a downstream exclusion figure collapsed from hundreds of thousands of BTC to a fraction of one. An all-failed sweep is now refused outright — stale and saying so beats wrong and silent.

Consumers count negligible clusters OUT by name rather than summing them, because 'minus 0 BTC' is an exclusion caption that excludes nothing.

What to watch

A reserve is a balance, not an intention. Coins leaving an exchange are as consistent with a custody migration as with accumulation.

Cluster coverage differs per exchange, so comparing two exchanges' reserves compares two different degrees of completeness. The tracked-cluster count travels with the figure.

This platform tracks a small number of clusters. Subtracting them from a chain-wide aggregate is captioned as exactly that, and never as excluding every exchange.

Open the live Smart Money tab →

Related metrics

Entity Portfolio Value
The total on-chain value held by a named entity — an exchange, a treasury company, a fund — summed across every address claim t….
Bitcoin Whale Band Holdings
The holder structure of the whole Bitcoin chain, banded by balance size: how many addresses sit in each band and how much BTC t….
Stablecoin Net Flows
The change in total stablecoin supply over a period — net issuance minus net redemption.
Coinbase Premium Index
The premium is the gap between what Bitcoin trades for on a US dollar venue and what it trades for on the largest offshore USDT….
Every rule above is enforced in code and pinned by a test. The doctrine they all follow — NULL is not zero, partial cannot erase, an unfinished period is not evidence — is published on the methodology page.
Descriptive, not advice. Every score, band and verdict on this platform is an automated heuristic over public data and can be wrong in both directions. All metrics · Methodology · About · Dashboard