On-chain metrics · Exchange reserve data

Bitcoin exchange reserves with exchange balance data and charts

Bitcoin exchange reserves is the total BTC held in addresses believed to belong to exchanges. Every published reserve figure is an address list somebody assembled by inference, with simple arithmetic over the chain on top, and the arithmetic was never the uncertain part.

or try it below ↓

REST API · SQL · dashboards · address history from genesis

Query Console

BTC address or sample query

btc

try:

Hit Run to query the Bitcoin blockchain.

Live from the chain for .

BTC

Balance trail

informational on-chain data · not financial advice

REST API · SQL · dashboards, one indexed dataset. Reading the chain ...

The chain records balances exactly and nobody disagrees about them. The address list is a judgment call, it is rarely published, and it is where the differences between providers come from. We keep re-measuring the same set to show what that means in practice. On 16 August 2026, at block 962,803, five widely reported exchange cold addresses held 671,622.49 BTC between them. We measured the identical five addresses again on 24 August 2026 at block 963,839, mined at 09:16:34 UTC. They held 689,765.55 BTC. That is 18,143.06 more BTC in eight days, an increase of 2.70 percent, the sort of move that gets written up as coins flowing onto exchanges.

Four of the five addresses did not move. Between them they changed by less than four thousandths of a single bitcoin, and three of them recorded no transactions at all. The entire 18,143.06 BTC came from one address, in four transactions, and one of those four was a 546 satoshi dust output. Three real transfers produced the whole eight day move.

Then it gets more specific. All three transfers were sweeps out of the same single address, a hot wallet with 2,308,660 transactions on it that is not on the reserve list at all. Each one pulled a pile of native SegWit outputs from that wallet, paid a round number into the cold address we track, and sent the change straight back to the source. All three paid an identical 5.03 satoshis per virtual byte, which is what an automated treasury script looks like rather than a market. No bitcoin entered anyone's control during those eight days. One operator moved coins from an address that was not being counted into an address that was, and a reserve series recorded it as a 2.70 percent inflow.

We read the same five addresses a third time on 29 August 2026, at block 964,592. The total was 689,765.55 BTC, unchanged. In the 753 blocks since the 24 August reading, five transactions touched the set and every one of them was inbound dust: two receives of 546 satoshis, one of 3,148, and a single spray that dropped 330 satoshis into two of the addresses at once. Total received across all five addresses over five days: 5,446 satoshis. Total spent: zero. We read them a fourth time on 2 September 2026 and the total was still 689,765.55 BTC, and a fifth time on 3 September 2026 at block 965,290, when it had risen to 694,600.03 BTC because one transaction that morning swept 4,834.484 BTC out of that same off-list hot wallet at the same 5.03 satoshis per virtual byte. We read them a sixth time on 4 September 2026 at block 965,562, a seventh on 6 September at block 965,791, an eighth on 8 September at block 966,164 and a ninth on 10 September at block 966,451, and on all four occasions the total was 69,460,003,430,648 satoshis, identical to the satoshi across 889 blocks and six days. Then on 16 September 2026, at block 967,238, it moved: 700,223.17132286 BTC, up 5,623.13701638 BTC, and one transaction in block 967,201 accounts for all but 1,092 satoshis of it, even though the off-list wallet that funds the set pushed 58,861.16 BTC of gross flow through 1,677 transactions during the most recent of those windows alone, while its own balance climbed past 13,598.75 BTC. The eleventh reading, on 21 September 2026 at block 968,071, came to 710,260.04947352 BTC, up another 10,036.87815066 BTC, and it decomposes exactly: two sweeps out of that same wallet, 5,151.775 BTC in block 967,458 and 4,885.103 BTC in block 967,898, plus 15,066 satoshis of dust. So the series reads plus 2.70 percent, then ten days of nothing, then plus 0.70 percent in a single block, then nothing again, and the nothing is as informative as either jump. A metric that moved 18,143 BTC in eight days and 0.00005446 BTC in the next five is not measuring a continuous flow, it is recording the occasional moment when one operator moves coins across the boundary of somebody address list. Note also what a transaction counter would have said: these cold wallets were active five times in five days. That is the structural problem with reserve data, and it is not fixable by better arithmetic. A fixed address list turns internal treasury housekeeping into apparent flow, and it misses real deposits into any address nobody has attributed yet. BitcoinDatabase gives you the layer underneath: exact balances for any address at any height, the transactions that changed them, and the ability to define your own address set rather than inherit somebody else's. Both explorers we checked agreed to the satoshi on all five addresses, so the disagreement was never in the numbers. This is informational on-chain data and analytics, not financial, investment or trading advice.

REST API SQL DASHBOARDS WEBHOOKS CSV EXPORT

History from genesis queryable in seconds

On-chain data not financial advice

Why it works

What you get with Exchange reserve data

Tenth reading where the sweep landed at 5.028 satoshis per virtual byte

We read the same five addresses a tenth time on 16 September 2026 at block 967,238, 787 blocks after the ninth reading, and this time the number moved. The total is 70,022,317,132,286 satoshis, which is 700,223.17132286 BTC, up 5,623.13701638 BTC from the five byte-identical readings before it. Two independent explorers agreed on every address to the satoshi, as they have at all ten readings. We can account for the entire change in three transactions. One of them is the sweep: 5,623.137 BTC into 3M219KR5 from the off-list funding wallet, in block 967,201. The other two are 546 satoshi dust receives, one into 3M219KR5 in block 966,534 and one into 34xp4vRo, and a fourth dust payment of the same size was sitting unconfirmed in the mempool while we read. That is the whole of a 5,623 BTC move: one treasury transfer and some spam, and a reserve dashboard counts all four as inflows. Meanwhile the funding wallet fell from 13,598.75 BTC to 11,238.84 BTC, down 2,359.92 BTC, across 4,329 new transactions. We traced that window independently rather than inferring it: those 4,329 transactions took in 90,839.14 BTC and paid out 93,199.06 BTC, which is 184,038.20 BTC of gross flow, and the net of the two matches the separately read balance change to the satoshi. So the reserve series records a 0.81 percent inflow, and what actually happened underneath it was 184,000 BTC of churn through an address the list does not contain, of which one transaction happened to cross the boundary.

Ninth reading where the funder crossed the threshold and the metric still did not move

We read the same five addresses a ninth time on 10 September 2026 at block 966,451, 287 blocks after the eighth reading. The total was 69,460,003,430,648 satoshis. It was that same figure on 4, 6 and 8 September, so four consecutive readings across 889 blocks and six days have now returned a number identical to the satoshi, confirmed by two independent explorers on every address at every reading. The wallet underneath kept rebuilding, and fast. The off-list hot wallet that funded every sweep on record went from 11,353.48 BTC to 13,598.75 BTC, up 2,245.27 BTC, and it now carries 2,324,375 transactions. We traced the window rather than inferring it: across those 287 blocks that one address confirmed 1,677 transactions, took in 30,553.21 BTC and paid out 28,307.95 BTC, which is 58,861.16 BTC of gross flow, and the net of those two figures matches the independently read balance change to the satoshi. Fifty-eight thousand bitcoin moved through one address in about two days and the reserve metric drew a flat line. The flat line was accurate. It was also the least informative thing on the chart.

Eighth reading where the funder turned and the metric still did not move

We read the same five addresses an eighth time on 8 September 2026 at block 966,164, 373 blocks after the seventh reading. The total was 69,460,003,430,648 satoshis. It was 69,460,003,430,648 satoshis on 4 September and on 6 September too, so three consecutive readings across nine days and 602 blocks have now returned a byte-identical figure, confirmed by two independent explorers on every address. What changed is the wallet underneath. The off-list hot wallet that funded every sweep on record had been draining for a week, and in this window it turned: it went from 8,844.99 BTC to 11,353.48 BTC, up 2,508.50 BTC. We traced the window rather than inferring it. Across those 373 blocks that one wallet confirmed 2,225 transactions, took in 32,494.30 BTC and paid out 29,985.80 BTC, which is 62,480.10 BTC of gross flow, and the net of those two figures matches the balance change to the satoshi. So an operator pushed sixty-two thousand bitcoin through a single address over two and a half days and the reserve metric reported a flat line. Flat does not mean quiet. It means the activity happened at an address the list does not contain.

Seventh reading where we predicted flat and got flat to the satoshi

On 4 September 2026 this page published a dated prediction with its falsifier stated up front: the five address total would stay within a few thousand satoshis of 694,600.034 BTC until the hot wallet funding it rebuilt. We read the same five addresses again on 6 September 2026 at block 965,791, 229 blocks later. The total was 69,460,003,430,648 satoshis. The 4 September total was 69,460,003,430,648 satoshis. Not close, identical, and two independent explorers agreed on every address. The prediction resolved correctly and tighter than we claimed, and the reason it did is the mechanism rather than luck: the hot wallet has not rebuilt. It held 8,844.99 BTC at this reading against 8,921.21 BTC on 4 September, so it is still draining rather than accumulating, and with nothing staged there is nothing to sweep. We grade these in public whether they land or not. The previous one was right about the mechanism and wrong about the magnitude.

Zero satoshis moved while the same operator pushed 27,829 BTC

This is the clearest illustration we have found of what an address-list reserve metric hides. Across the identical 229 block window in which the tracked total moved exactly zero satoshis, the hot wallet feeding these cold addresses confirmed 1,221 transactions: 13,952.73 BTC out and 13,876.51 BTC in, which is 27,829.24 BTC of gross flow and a net drawdown of 76.22 BTC. A dashboard reading the five addresses would have drawn a perfectly flat line for two days and been completely accurate. The operator moved almost 28,000 BTC in that time. At the moment we took the reading the wallet also had one unconfirmed transaction in the mempool worth 22.27 BTC, which no reserve chart would show at all. Flat does not mean quiet. It means the activity happened at addresses your list does not contain.

Sixth reading where the reserve sat still while its funder shed 1,942 BTC

We read the same five addresses again on 4 September 2026 at block 965,562, 272 blocks after the sweep. The total was 694,600.03430648 BTC against 694,600.03418335 BTC the day before, a change of 12,313 satoshis, and we can name the single transaction that caused it: 3eb2c7a8eba726e4 in block 965,394, one inbound payment of 12,313 satoshis to 34xp4vRo. The other four addresses recorded no transactions at all. In the same window the hot wallet that funds this cold address fell from 10,863.60 BTC to 8,921.20673631 BTC, down 1,942.39426033 BTC, and none of it arrived at any tracked address. It went out in thousands of small customer withdrawals instead. So the reserve metric reads perfectly flat across two days in which the operator behind it moved nearly two thousand bitcoin. A reserve chart shows you the boundary of an address list, not the movement of an exchange.

The next prediction, on the record and dated

Prediction five resolved on 17 September 2026. On 16 September this page said the next step in the five address total would be a single transaction out of bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h into a tracked cold address, paying 4 to 6 satoshis per virtual byte, once that wallet was back above roughly 13,000 BTC. What happened: transaction 060a8293811331c8cdd6bc641564bf06bbc2ab4e6ad30563bceb55252e5ebfd4 confirmed in block 967,458 at 19:51:08 UTC. One transaction, 44 inputs, all from that wallet. It paid 5,151.775 BTC into 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6 with the change back to the source, at 5.027 satoshis per virtual byte, with the wallet holding 15,426.53 BTC just before it. A second sweep followed in block 967,898 on 20 September: 70 inputs, 4,885.103 BTC into the same address, 5.029 satoshis per virtual byte, from a wallet at 13,703.98 BTC. None of the three falsifiers fired. The wallet never passed 16,000 BTC (its high was 15,452.87), the total never stepped while it was under 11,000 BTC, and both fee rates sat inside the band. What we got wrong: the date. We said around 19 September, because the wallet had been gaining 4.15 BTC per block. It gained 19.03 BTC per block instead and crossed 13,000 BTC at block 967,347, so the sweep came two days early. The fee rate is the sharper result. The median fee actually paid across recent blocks was 0.348 satoshis per virtual byte on 17 September and about 1.05 on 21 September, so a sweep at 5.03 pays roughly five to fourteen times the going rate. Every sweep we have caught pays the same number, which points to a fixed setting in the operator treasury script rather than a fee estimator. We traced every one of the 5,094 transactions on that wallet between blocks 967,239 and 968,069: 199,166.42 BTC of gross flow, a net of minus 3,032.34 BTC, reconciling to its independently read balance of 8,206.49769473 BTC to the satoshi. Prediction six, dated 21 September 2026: the next step will again be a single transaction out of that wallet into a tracked cold address at 4 to 6 satoshis per virtual byte, once it is back above roughly 13,000 BTC. We are not dating it this time, because the wallet has swung from gaining 19 BTC per block to losing 3.6 BTC per block since the last sweep, and a date built on either rate would be a guess. Same three falsifiers: wrong if the total steps while the wallet is under 11,000 BTC, wrong if it passes 16,000 BTC with no sweep, wrong if the fee rate falls outside 4 to 6. The twelfth reading, on 22 September 2026 at block 968,125, found the total unchanged to the satoshi at 71,026,004,947,352 on both explorers, while the funding wallet drained further, to 7,304.93538561 BTC. We traced all 237 of its transactions between blocks 968,072 and 968,125, exactly the change in its on-chain transaction count, for 5,590.03 BTC of gross flow and a net of minus 901.56230912 BTC, again reconciling to the satoshi. That is about 16.7 BTC leaving per block, so the wallet is moving away from the threshold, not toward it, and no falsifier has fired. Prediction six stays open. One more thing sat in the mempool at that reading: a 546 satoshi dust payment to 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6 from a new sender, paying 0.162 satoshis per virtual byte, a rate that only relays since Bitcoin Core lowered its default floor to 0.1. Earlier sprays at this set paid about 1. Six predictions on this page: four right, one right on mechanism and wrong on size, one open.

Thirteenth reading where prediction six went wrong because of a fixed address list

On 26 September 2026 at block 968,700, mempool.space and Blockstream both returned 68,543,154,334,626 satoshis for the five tracked addresses, 685,431.54334626 BTC, down 24,828.50612726 BTC from the twelfth reading. The drop is exactly two transactions, both spending from 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6. Transaction 20ea3fdd4656f5a9411541c80652e88043d8f5e1ab5d3c13ae9837ac386174cb in block 968,173, at 19:37:24 UTC on 22 September, sent 4,000 BTC to the funding wallet bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h and 2,306.9040038 BTC to a new address, 33TDwJRsFJbtSEwkXGwtJkj3Ey3Gman3TV. Transaction a89534d502b350f908ff9335394234b99f45607a6635984db47c108150e7bf5e in block 968,175, with 45 inputs, sent 10,000 BTC to 1Pzaqw98PeRfyHypfqyEgg5yycJRsENrE7 and 8,521.60204577 BTC to another new address, 32z4XkTCBMp4QfcoobxaXtJkWngyHqGjHp. Both paid 3.005 satoshis per virtual byte, not the 5.03 every inbound sweep paid. Prediction six said the next step would be a sweep into cold storage, and that it was wrong if the total stepped while the funding wallet was under 11,000 BTC. The total stepped at block 968,173 with that wallet near 7,300 BTC. The falsifier fired, so prediction six is graded wrong. The model only allowed coins to flow from the hot wallet into cold storage. Here cold storage refilled the hot wallet. Where the coins went is the useful part. 4,000 BTC landed in the funding wallet directly. 1Pzaqw98PeRfyHypfqyEgg5yycJRsENrE7 then forwarded 2,000 BTC in block 968,280, 4,000 in block 968,481 and 4,000 in block 968,556 to the same funding wallet, and now holds 0.49999939 BTC. The two new addresses received the non-round remainders, 10,828.50604957 BTC together, and have not spent. Both transactions also spent 546 satoshi dust inputs sitting at 3E97AjYaCq9QYnfFMtBCYiCEsN956Rvpj2 and 3PXBET2GrTwCamkeDzKCx8DeGDyrbuGKoc, so the same signer controls those addresses too. Read with the standard common-input and change heuristics, of the 24,828.5 BTC a reserve chart shows leaving, 14,000 BTC went to the operator's own hot wallet and 10,828.5 BTC went to its fresh change addresses: none of it left. Add the two change addresses and the set holds 696,260.05 BTC. Heuristics are evidence, not proof, and we label them that way. The funding wallet held 9,703.23809133 BTC across 2,337,746 transactions at this reading. The 0.162 satoshi per virtual byte dust payment seen at the twelfth reading is still unconfirmed more than 570 blocks later. The ledger now stands at six predictions: four right, one right on mechanism and wrong on size, one wrong. There is no seventh yet. A model that just missed a whole direction of flow should account for it before it predicts again.

Fourteenth reading where the total held still to the satoshi for 359 blocks

On 28 September 2026 at block 969,059, mempool.space and Blockstream again agreed on every digit: 68,543,154,334,626 satoshis, 685,431.54334626 BTC, for the five tracked addresses. That is identical to the satoshi to the thirteenth reading at block 968,700, 359 blocks earlier. The two change addresses from the 22 September move, 33TDwJRsFJbtSEwkXGwtJkj3Ey3Gman3TV with 2,306.9040038 BTC and 32z4XkTCBMp4QfcoobxaXtJkWngyHqGjHp with 8,521.60204577 BTC, each still have exactly one transaction and have spent nothing. The 0.162 satoshi per virtual byte dust payment of 546 satoshis to 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6, transaction 9ac092529d872e7bc9f2271765365845ebd312b2dc67151cd371fc7909e1afbf, is still unconfirmed, now more than 900 blocks after we first saw it, so a reserve feed that counts pending receives has carried it as an inflow for most of a week. The funding hot wallet bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h rose from 9,703.23809133 to 11,659.65108144 BTC across 2,339,834 transactions, an average of about 5.45 BTC a block, with five pending transactions netting minus 1.88619813 BTC. Underneath that net figure the gross traffic is far larger: in the 33 blocks from 969,027 to 969,059 alone, 42 sweeps pulled 1,709 customer deposits worth 2,191.47198638 BTC into the wallet, and the median deposit had waited only 2.2 hours at its deposit address. A flat reserve line and a wallet taking in 2,191 BTC in 33 blocks are both true at once. We are not issuing a seventh prediction yet: the cold to hot direction that broke prediction six is still unmodelled.

Sixteenth reading shows the hot wallet passed 13,000 BTC and did not sweep

On 30 September 2026 at block 969,296 the five tracked addresses totaled 68,543,154,334,626 satoshis for a third reading in a row, identical to the satoshi across 123 more blocks, with mempool.space and Blockstream agreeing on every digit. The 546 satoshi dust payment to 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6 is still waiting in the mempool, and both change addresses from 22 September are untouched at 2,306.9040038 and 8,521.60204577 BTC. The funding hot wallet bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h is where the answer is. It went from 13,376.12623317 to 12,811.73616948 BTC. We traced all 723 of its transactions in the window: it received 14,069.16043446 BTC and sent 14,633.55049815 BTC, a net of minus 564.39006369 BTC that reconciles to the two balance readings to the satoshi. Not one of those payments went to a tracked cold address. The largest single outflow was 129.1 BTC, and every large spend looks like a customer withdrawal batch. So the wallet sat above the roughly 13,000 BTC level where every September sweep fired, and no sweep came. Together with prediction six, that is two readings in a row where the September pattern did not hold, which is why we are still not publishing a seventh prediction. It is also a reminder of scale: 28,702.71 BTC of gross flow in about 20 hours sat behind a reserve total that did not move by one satoshi.

Fifteenth reading stayed flat while the hot wallet passed 13,000 BTC

On 29 September 2026 at block 969,173 the five tracked addresses again totaled 68,543,154,334,626 satoshis, 685,431.54334626 BTC, identical to the satoshi to the fourteenth reading 114 blocks earlier, with mempool.space and Blockstream agreeing on every digit. The 546 satoshi dust payment 9ac092529d872e7b to 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6 is still unconfirmed. The two change addresses from the 22 September move still hold 2,306.9040038 and 8,521.60204577 BTC in one transaction each. The funding hot wallet bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h rose from 11,659.65108144 to 13,376.12623317 BTC across 2,340,451 transactions, about 15.06 BTC a block, with five pending transactions netting minus 77.8172784 BTC. That puts it back above the roughly 13,000 BTC level at which every September sweep into cold storage happened, each paying about 5.03 satoshis per virtual byte. We are still not issuing a seventh prediction, because the cold to hot move that broke prediction six remains unmodelled, but the next reading will show whether the old pattern resumes. We also checked a third party daily feed against this set on the same day: it refused both P2WSH cold addresses, which hold 226,942.49 BTC, and read the others 3 to 21 satoshis low, which is why every figure here comes from two explorers at a named height.

We predicted the step, then caught it in the act

On 2 September 2026 this page recorded a flat reserve total and an explanation for it. The five tracked cold addresses had received 1,400 satoshis and spent nothing across 594 blocks, while the hot wallet that funded every August sweep quietly grew from 8,524.23 BTC to 14,932.62 BTC. We wrote that if that operator swept, the metric would step in a single block and every chart would render nine days of accumulation as a sudden inflow event. It swept the following morning. In block 965,274, at 07:07:41 UTC on 3 September 2026, one transaction moved 4,834.484 BTC from that hot wallet into the largest tracked address, across 56 inputs, at exactly 5.03 satoshis per virtual byte, the same fee rate as the 17, 21 and 23 August transfers. Our fifth reading, taken at block 965,290, puts the total at 694,600.03 BTC. We were right about the mechanism and wrong about the magnitude: 0.70 percent rather than the roughly two percent we guessed, because the operator swept about a third of its balance rather than all of it, and still holds 10,863.60 BTC that no reserve figure counts. Nothing about the exchange changed that morning. What changed was that coins crossed a line somebody drew on a list. A reserve series is a step function sampled from a continuous process, and the steps are bookkeeping, not behavior.

Your address set, not an inherited one

Reserve totals are decided by which addresses go on the list, and most providers never publish theirs. Define your own set, version it, and see exactly how the total responds when you add or remove an address. The single largest address in the set above is 36 percent of it, so one attribution decision moves the headline by roughly 248,000 BTC.

Balances at any height, not just now

Every balance is available as of any block, so a reserve series is a real historical measurement rather than a chart stitched together from whatever was recorded that day. That is what makes an eight day delta auditable: you can recompute both endpoints from the chain and get the same numbers we did.

The transactions behind every move

A change in the total always ties back to specific transactions, with their inputs, outputs, fee rates and counterparties. That is the only way to tell a genuine deposit from an internal sweep, and it is the difference between reporting an inflow and understanding one.

Side by side

The same five addresses, read nine times, measured first-hand

Every balance below was read from the indexed chain and cross-checked against a second independent explorer, which agreed to the satoshi on all five, at every reading. The 16 August figures were taken at block 962,803, the 3 September figures at block 965,290 and the 10 September figures at block 966,451. Readings from 24 and 29 August and from 2 and 4 September are omitted only because each was identical to the reading beside it, which is itself the point this table is making. Nothing here is estimated or modeled.

Tracked cold address	16 Aug 2026	3 Sep 2026	10 Sep 2026	What changed since 8 Sep
3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6	196,082.40 BTC	219,059.94 BTC	219,059.94 BTC	Nothing. Still 597 transactions
34xp4vRoCGJym3xR7yCVPFHoCNxv4Twseo	248,597.59 BTC	248,597.59 BTC	248,597.59 BTC	Nothing
bc1qgdjqv0av3q56jvd82tkdjpy7gdp9ut8tlqmgrp...	130,010.08 BTC	130,010.08 BTC	130,010.08 BTC	Nothing
bc1qjasf9z3h7w3jspkhtgatgpyvvzgpa2wwd2lr0e...	96,932.41 BTC	96,932.41 BTC	96,932.41 BTC	Nothing
1FzWLkAahHooV3kzTgyx6qsswXJ6sCXkSR	0.0139 BTC	0.0139 BTC	0.0139 BTC	Nothing
All five combined	671,622.49 BTC	694,600.03 BTC	694,600.03 BTC	Zero satoshis across 287 blocks

The two bech32 addresses are shortened above only so the table stays readable. In full they are bc1qgdjqv0av3q56jvd82tkdjpy7gdp9ut8tlqmgrpmv24sq90ecnvqqjwvw97 and bc1qjasf9z3h7w3jspkhtgatgpyvvzgpa2wwd2lr0eh5tx44reyn2k7sfc27a4, so you can verify every figure yourself. The three transfers that produced the whole move landed on 17 August at 21:11:32 UTC (4,735.68 BTC from 37 inputs), 21 August at 11:47:40 UTC (5,754.215 BTC from 97 inputs) and 23 August at 21:46:03 UTC (7,653.163 BTC from 121 inputs). Every input in all three came from one address, bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h, which held 8,524.23 BTC across 2,308,660 transactions when we first traced it on 24 August and appears on no reserve list we have seen. Each transfer returned its change to that same source address and paid exactly 5.03 satoshis per virtual byte. Read together, that is one operator rebalancing between two of its own wallets, recorded by the metric as a 2.70 percent inflow. The fourth reading, on 2 September 2026 at block 965,186, found the five tracked addresses perfectly still: 1,400 satoshis received and nothing spent across 594 blocks, with the published total unchanged at 689,765.55 BTC. Meanwhile that same source wallet had grown to 14,932.62 BTC across 2,316,989 transactions, from 8,524.23 BTC on 24 August, so roughly 6,408 BTC had accumulated where no reserve figure counts it. On the strength of that we wrote on this page that if the operator swept, the metric would step in a single block and every chart would render nine quiet days as a sudden inflow. It swept the next morning. At 07:07:41 UTC on 3 September 2026, in block 965,274, transaction 508857ebe7978c0e moved 4,834.484 BTC from bc1qm34lsc65zpw79lxes69zkqmk6ee3ewf0j77s3h into 3M219KR5vEneNb47ewrPfWyb5jQ2DjxRP6 across 56 inputs, paying 19,480 satoshis at exactly 5.03 satoshis per virtual byte, the identical fee rate as the August sweeps. The prediction was right about the mechanism and wrong about the size: the total rose 0.70 percent rather than the two percent we guessed, because the operator moved only about a third of what it was holding and kept the rest as working balance. It still holds 10,863.60 BTC that no reserve total counts.

What it handles

The indexed Bitcoin chain, queryable your way

Look up an address, a transaction, a UTXO, the rich list or an on-chain metric, by REST API, SQL or dashboard. The same authoritative data, reconciled block-by-block against the canonical chain, without running a node.

Track exchange balance totals across any address set you define
Read exact balances as of any block height, not just the current tip
See the transactions that moved a reserve total, with fee rates and counterparties
Separate internal treasury sweeps from genuine deposits and withdrawals
Build inflow and outflow series from the underlying transfers
Version an attribution list and measure how much each address decides
Rebuild reserve history in SQL over the indexed chain
Join exchange balances to coin age, address type and price in one query

GET /v1/address/{addr} query result

200 · JSON

{
  "address": "bc1qxy2k…l0wdv8",
  "balance_btc": 68432.10,
  "balance_usd": 4612165420,
  "tx_count": 1284,
  "unspent_outputs": 37,
  "first_seen": "2014-02-09"
}

indexed from genesis · to the satoshi ✓ reconciled block-by-block

Why BitcoinDatabase

One platform, queryable three ways

Not a raw node to sync, not an indexer to build, and not five vendors to stitch together. The fully-indexed Bitcoin blockchain, available as a REST API, as SQL, and as dashboards, on one authoritative dataset.

REST API

Typed JSON for addresses, transactions, balances, UTXOs and metrics. Drop it into apps, wallets, explorers and agents with curl, Python or any HTTP client.

SQL access

Run SQL directly against the indexed Bitcoin dataset for ad-hoc analysis, cohorts and exports, the same data the API and dashboards read from.

Compliance-first

Informational on-chain data and analytics only. Entity labels and flow tracing are framed as tooling to support a regulated team's own review, not accusations.

Good questions

Questions about Exchange reserve data

Because the chart counts coins leaving a fixed list of addresses, not coins leaving the exchange. On 22 September 2026 our tracked set fell 24,828.5 BTC in two transactions. Tracing them, 14,000 BTC went to the same exchange hot wallet and 10,828.5 BTC went to fresh change addresses. By standard heuristics, none of it left.

Sometimes, if you watch the funding wallet instead of the reserve total. On 8 September 2026 we published a prediction that the next step would be one transaction out of a specific off-list wallet into a specific cold address at about 5.03 satoshis per virtual byte, once that wallet passed 13,000 BTC. It confirmed on 15 September in block 967,201 as a single 77 input transaction paying 5.028 satoshis per virtual byte. The reserve figure is a lagging indicator of a treasury process that is visible in advance.

Because one operator moved coins across the boundary of an address list. Between 10 and 16 September 2026 the five address set we track rose 5,623.13701638 BTC, and a single transaction in block 967,201 accounts for 5,623.137 of that. It swept 77 outputs out of a hot wallet that is not on the list into a cold address that is. No bitcoin entered or left anyone control. The remaining 1,092 satoshis were dust payments.

Yes, and it is a known contaminant of these metrics. In the 787 blocks to 16 September 2026, three separate 546 satoshi payments landed on addresses in our tracked set, with a fourth pending in the mempool at 0.162 satoshis per virtual byte. Each one is an unsolicited payment that an address list counts as an inbound transfer. The amounts are trivial, the transaction counts are not, so any metric that reports exchange wallet activity rather than value will be distorted by them.

Exchange reserves is the total amount of BTC sitting in addresses attributed to cryptocurrency exchanges. It is used as a rough gauge of how much supply is held in places it could be sold from. The chain records balances exactly, but it never records who owns an address, so the attribution behind every reserve figure is inference.

Someone assembles a list of addresses believed to belong to exchanges, then sums the unspent outputs held by those addresses at a given block. The arithmetic is trivial and reproducible. The list is the hard part, and it is the part that almost no provider publishes, which is why identical chain data yields different published totals.

Any specific number depends entirely on whose address list you use. To show the scale of that, we measured five widely reported exchange cold addresses on 24 August 2026 and found 689,765.55 BTC between them. A list with one more or one fewer large address would report a total tens or hundreds of thousands of BTC away from that.

Because they are measuring different address sets, not different chains. Attribution is inference drawn from clustering and observed behavior, and reasonable analysts disagree. In the five address set we track, the largest single address is 36 percent of the total, so including or excluding it alone shifts the answer by about 248,000 BTC.

In practice they are used interchangeably, and both mean the BTC held at addresses attributed to an exchange. Reserves is the more common phrasing for an industry-wide total, while exchange balance more often refers to a single named venue. Neither term implies any audit, attestation or claim about customer liabilities.

No, and the gap matters. Exchange reserves is an outside estimate of what addresses appear to hold. Proof of reserves is an exercise the exchange itself runs, usually signing messages from its addresses and publishing a commitment to customer balances. Reserves measures assets only. Proof of reserves is an attempt to compare assets against liabilities.

Mechanically it means BTC left the addresses on somebody list, which can be a withdrawal, an internal transfer to an address nobody has attributed, or a change to the list itself. BitcoinDatabase does not offer a market reading of that, because the same movement can have several causes. This is informational data, not investment or trading advice.

Yes, and it is common. Between 16 and 24 August 2026 the five address set we track gained 18,143.06 BTC, all of it from three sweeps out of a single hot wallet that is not on the list, with change returned to that same wallet and an identical 5.03 satoshi per virtual byte fee on each. Nothing entered the operator control. The list simply started counting coins it had not counted before.

You do not know, you infer it, usually by clustering addresses that co-spend together and matching them against deposits observed from known accounts. It is good enough to be useful and never certain. We publish the addresses we measure so the inference is yours to accept or reject rather than something buried in a methodology page.

No, and assuming they do is the common mistake. We read the same five addresses on 16, 24 and 29 August 2026. The set gained 18,143.06 BTC over the first eight days, then 5,446 satoshis over the next five, with nothing spent at all. Reserve totals sit still and then step, because they change when one operator crosses an address list boundary.

Because they get sprayed with dust. In the five days after 24 August 2026, every transaction touching the five addresses we track was inbound dust, between 330 and 3,148 satoshis, including one spray that hit two of them at once. A dashboard counting transactions or active addresses reports those wallets as busy. A dashboard counting value reports the truth, which is that nothing happened.

Balances change whenever a transaction confirms, so the underlying data updates with each new block, roughly every ten minutes. Most published reserve charts sample once a day, which is why a large sweep can appear as a single vertical step. Querying at a block height rather than a date removes that artifact.

Because a reserve figure only counts coins sitting in a specific list of addresses, and operators move between their own wallets on their own schedule. We read the same five addresses five times between 16 August and 3 September 2026. The total sat perfectly flat for ten days, including a 594 block stretch with 1,400 satoshis received and nothing spent, then rose 4,834.48 BTC in one block on 3 September. The jump is the moment an internal transfer crosses the list boundary, not the moment money moved.

Whatever is staged in wallets the list does not include, and it can be thousands of BTC. On 2 September 2026 the hot wallet feeding these cold addresses held 14,932.62 BTC, up from 8,524.23 BTC on 24 August, none of it counted. On 3 September it swept 4,834.484 BTC across the boundary and the published total jumped 0.70 percent, yet it still holds 10,863.60 BTC that no reserve figure counts. You cannot size this gap without knowing the operator address set.

Compute the same change two independent ways and check that they agree. Read the address balance at the start block and the end block and subtract, then separately sum every transaction that touched the address in that window. On 10 September 2026 we did this over 287 blocks on the wallet funding this set: the balance delta was 2,245.26856005 BTC and the transaction level net was 2,245.26856005 BTC, matching to the satoshi.

Satoshis, which is eight decimal places. Commercial balance feeds commonly document two decimal places and an hourly cache, and at that granularity you cannot tell an unchanged balance from one that drifted by hundreds of thousands of satoshis. The four consecutive byte-identical readings on this page would be invisible at two decimals, and so would any residual an auditor asks you to explain.

Regularly, and the technique changes. On 28 August 2026 a single transaction sprayed 21 outputs of 330 to 546 satoshis at 0.52 satoshis per virtual byte, hitting two of these addresses. On 2 September 2026 at block 965,137 the same two P2SH addresses each received exactly 700 satoshis, but from two separate one-input two-output transactions sent by the same address in the same second, each paying 448 satoshis at about 1.75 satoshis per virtual byte. Same objective, different construction, and more than three times the fee rate. Any filter written against transaction shape needs to expect both.

Yes, and you should. Every figure on this page came from public block data at a stated block height, so any indexer or explorer will reproduce it. The five addresses are listed in full, the heights are 962,803 and 963,839 and 964,592 and 965,186 and 965,290 and 965,562, and the balances cross-checked to the satoshi against a second independent explorer. What you cannot verify from the chain is the attribution, which is to say whether these addresses really belong to the exchanges people say they do. That part is inference and it always has been.

No, and expecting that is the most common mistake made with this metric. Reserves measured from a cold address list sit perfectly still and then step. Our six readings of the same five addresses run flat, then up 2.70 percent, then flat, then up 0.70 percent in a single block, then flat again. The steps happen when one operator crosses the boundary of somebody address list, not when customer flows change. If your model treats a reserve series as a continuous flow, it will read scheduled treasury housekeeping as market activity.

Because most of the movement never touches a tracked address. Between blocks 965,290 and 965,562 our five-address total changed by 12,313 satoshis while the hot wallet that funds one of them fell by 1,942.39 BTC, paid out in thousands of small customer withdrawals to addresses no reserve list contains. Both facts are true at once. A reserve figure only sees the addresses on the list, so the more of an operator balance sits off-list, the less the metric describes the business.

More often than the metric appears to change, because the value of the series is in catching the step and knowing which transaction caused it. We read every few days and record the block height, the satoshi total and the transaction ids. That is what lets us say a move was one automated sweep at 5.03 satoshis per vbyte rather than an inflow trend. Reading a rounded total once a week gives you the number and loses the explanation.

No, and this is the most common misreading of the metric. Between blocks 965,562 and 965,791, on 4 to 6 September 2026, our five tracked cold addresses moved exactly zero satoshis while the hot wallet that funds them confirmed 1,221 transactions totaling 27,829.24 BTC of gross flow. A reserve chart would have drawn a flat line for two days and been perfectly accurate. Flat means the movement happened at addresses the list does not contain.

Ours are published with dates and falsifiers so you can check them, which is rarer than it should be. Five are graded on this page. Four resolved right, most recently the 16 September call that the next step would be one sweep from a specific funding wallet at 4 to 6 satoshis per virtual byte, which confirmed on 17 September at 5.027, two days earlier than we dated it. The first was right about the mechanism and wrong about the size, and we published that too.

A reserve figure is an outside estimate: someone guesses which addresses an exchange controls and adds up the balances. Proof of reserves is an inside attestation, where the exchange itself demonstrates control of specific addresses and, in the stronger versions, commits to customer liabilities as well. The first can be produced without the exchange cooperating and is only as good as the address list. The second requires cooperation and is only as good as the liability side. Our proof of reserves API page covers the second.

Read every Bitcoin API question

Explore more

See every use case See pricing Back to the Bitcoin API

Stop running a node. Just query Bitcoin.

Run your first query now and get on-chain data back by REST API, SQL or dashboard, with address history back to the genesis block over REST. Informational on-chain data only, not financial advice.

See pricing

History from genesis over REST · SQL · dashboards · addresses, transactions, UTXOs, metrics