What we got wrong, and what we did not

Thursday's article closed with a section headed "What we are not saying". It refused to forecast, refused to set a level, and refused to call the 200-day average reclaimed as a state of affairs. All three of those refusals were right and they cost the reader nothing.

But one sentence went further than the data supported. Describing the roughly $2.7 billion of shorts that had been liquidated, we wrote that "that particular fuel has been spent rather than stored". The implication a reader would reasonably draw is that the move was near its end.

Bitcoin closed Thursday at $73,025 and reached $79,500 on Friday morning. Over three sessions it is up 20.83% from Tuesday's close of $64,725.

The error was not in the mechanism, which the open interest data confirmed then and confirms now. It was in the aside about what the mechanism implied for what came next. Describing a mechanism is measurement. Saying what it means for tomorrow is forecasting wearing measurement's clothes, and we have written four separate times this month that we do not do that. We did it in a subordinate clause and it was wrong within eighteen hours.

Two legs, two different mechanisms

Open interest is the cheapest honest test available for telling forced buying from chosen buying, and it separates Thursday from Friday cleanly.

LegPriceOpen interestChangeReading
Thu 08:00 to 09:00 UTC$69,816 to $71,592110,370 to 108,284 BTC1.89% lowerContracts closing: covering
Thu 23:00 to Fri 09:00$73,025 to $79,500 high107,070 to 110,629 BTC3.32% higherContracts opening: new positions
Fri 09:00 to 21:00$77,911 to $78,172110,629 to 106,060 BTC4.13% lowerLeverage coming off

The middle row is the answer to the obvious question about Friday. This was not the same trade as Thursday. Nobody was compelled to buy at $76,000. Open interest climbed through the whole overnight session and into the European morning, which is what it looks like when traders open new positions because they want the exposure.

The third row is the one worth sitting with. From the 09:00 peak through the close of the US session, open interest fell 4.13% while the price ended that stretch 0.34% higher. Roughly 4,570 Bitcoin of open exposure came off the board and the price did not give the gains back.

That combination cuts against the simplest bearish reading of an overbought market. A market that has just shed leverage while holding its level is in a different condition from one still carrying it. We are describing the condition, not what happens next.

The ten minutes that turned it

Both candles below are closed five minute candles from Binance, one spot and one perpetual, computed the same way our dashboard computes the live reading.

Candle, UTCPricePerp valuePerp buy sharePerp net flow
08:55$78,658 to $79,258, high $79,500$1,138.2 million56.0%$135.7 million bought
09:00$79,258 to $77,853$1,428.9 million40.4%$275.3 million sold
09:05$77,853 to $77,834$797.7 million50.5%$7.5 million bought

The candle that made the high moved $135.7 million of net buying. The candle that took it away moved $275.3 million of net selling, roughly twice as much, and it was the largest net flow in either direction on the whole session. Whoever wanted out at $79,000 wanted out more urgently than the last buyers wanted in.

Now put spot beside it, because this is where the two books disagree. Over the same two candles the spot tape read $12.5 million bought, then $11.7 million sold. Nearly symmetric. A reader watching only spot would have seen an ordinary turn. The asymmetry that actually characterised the reversal existed almost entirely on the futures book.

This is the same lesson as Thursday arriving from the other direction. Thursday the spot buy share understated the size of a move up. Friday it understated the violence of a move down. The metric is a ratio on one venue, and the venue it is on carries about a tenth of the value.

But the composition is shifting, and that is new

Having spent two articles explaining what our spot number cannot see, honesty requires reporting that it is currently seeing considerably more.

SessionSpot net buyingPerp net buyingPerp to spot
Thursday, full 24 hours$27.5 million$1,200.4 million43.7x
Friday, 22 hours so far$132.5 million$886.2 million6.7x

Friday's spot net buying is close to five times Thursday's across a shorter window, while futures net buying is lower. The ratio between the two books has compressed from about forty four to one down to under seven to one.

That is the signature of a move being carried by people buying the asset rather than by contracts changing hands, and it is the first session in this run where our own metric has been measuring a representative share of what is happening. We are noting it as a description of two sessions. Two sessions is not a trend and we are not going to call it one.

Two warnings in the coverage, and we could only reproduce one

A rally of this size brings caution pieces, and two specific claims circulated on Friday. We checked both against our own data rather than repeating them.

The overbought reading checks out. Computing RSI(14) with Wilder smoothing on completed daily Binance closes gives 80.32 at Thursday's close, the highest in the 384 days our window covers. Including Friday's still open candle it reads 85.86, which is consistent with the 84.45 figure reported elsewhere. We flag the difference because an RSI quoted during a live session is not a closed reading, and readers should know which one they are being shown. Bespoke Investment Group separately described the market as more than four standard deviations overbought during Thursday's move.

The funding rate warning we could not reproduce. Several outlets reported bitcoin funding at a 20 month high, described as traders paying steep fees to stay long. On Binance BTCUSDT the 16:00 UTC settlement was 0.0100%. That is technically the highest value in the last 500 settlements, going back to March 8, 2026. But the reason nothing is higher is that 0.0100% is a ceiling, not a peak. The rate has printed exactly that value in 25 of those 500 settlements, including four times in the last thirty, and 0.01% per eight hours is this venue's baseline rate.

Against a median of 0.0027% it is elevated, and we would not describe a rate sitting on a level it touches every few weeks as a warning. Other venues and aggregate measures may show something different, and we are not saying those reports are wrong about whatever they measured. We are saying we cannot see it here, and this is the venue our readers watch.

Four days of flows, and a concentration worth watching

The unforced demand extended for a fourth session, and the size is now material.

DateUS spot Bitcoin ETFsNote
August 17$297.56 millionFirst reported as $137.3 million
August 18$189.30 million
August 19$517.19 millionEight of twelve funds positive
August 20$606 millionLargest since May 1, IBIT $503 million
Four day totalAbout $1.61 billionEther funds added $221 million on Thursday

One detail cuts against the comfortable reading. On August 19 the money was spread across eight of the twelve funds, with BlackRock taking about 55%. On August 20 IBIT alone took $503 million of the $606 million, roughly 83%, and VanEck's HODL actually saw a small outflow. A day carried overwhelmingly by one issuer is a different thing from a day where the whole complex takes money, whatever the headline total says.

Friday's flow figure is not published at the time of writing. It normally lands after the US close and we will not estimate it.

Where things stand

AssetPrice24hSpot buy share, 22h partial
Bitcoin$78,172+7.36%52.0%
Ether$2,480.83+6.89%53.4%
XRP$1.4042+10.60%48.5%
Solana$92.89+6.08%54.1%
Dogecoin$0.09027+12.36%52.9%
BNB$679.40+3.97%
Tokenised gold$4,595+1.95%

Look at the XRP row. It is up 10.60% on the day and its spot taker buy share is 48.5%, which is to say more value crossed the spread selling than buying while the price rose more than any major except Dogecoin. That is the clearest single illustration on the board of why a buy share is not a direction indicator.

Bitcoin's high of $79,500 is its highest print since May 15, and the price sits 13.32% above the 200-day moving average at $69,015, a level it had not closed above since November 2, 2025 until Wednesday. Total market capitalisation is $2.646 trillion, up 3.83%, with Bitcoin dominance at 59.21%, up from 57.13% on Wednesday.

Fear and Greed reads 72, Greed. It read 62 yesterday, 46 on Wednesday and 31 on Monday. A 41 point move in five sessions.

And the ETH to BTC ratio, which we have now reported three days running: 0.03173, down 0.47%. We reported it rising twice and said explicitly that two observations are not a trend. The third observation went the other way. That is the whole reason for the rule.

What we are not saying

We are not saying the rally continues, and after yesterday we are going to be stricter about the subordinate clauses too. The measurements here describe three legs of price action that have already happened.

We are not saying the deleveraging is bullish. Open interest falling 4.13% while price held is a fact about condition, not a signal. It cuts against one bearish argument. It does not constitute a bullish one.

We are not setting a level, including $79,500. A high that failed once is a high that failed once. This site retired directional thresholds on August 3 for good reasons that have not changed.

We are not calling the composition shift a trend. Two sessions of spot carrying a larger share of net buying is two sessions.

What we are saying is that the mechanism behind Friday was different from the mechanism behind Thursday, that open interest is how you tell them apart, that the reversal at the high was carried by futures rather than spot, and that of the two warnings circulating about this market, one reproduces on our data and one does not. Every figure above came from public endpoints anyone can query.