Eleven days that went nowhere

Start with the shape of the period, because the flatness is itself the story.

Bitcoin closed August 3 at $63,520. It now trades at $63,079, down 0.69% over eleven sessions. In between it rallied steadily to a period high of $65,474 on August 9, then gave all of it back across the following week. The period low was $62,300, set on August 3 itself. Total volume across the stretch was $9.4 billion at an aggregate taker buy share of 50.8%, which is as close to perfectly balanced as this measure gets.

Sentiment matched the price. The Fear and Greed index has printed 29, 30, 31, 30, 29, 27, 29, 29 across the last eight readings, pinned in a four point band the entire time. Total crypto market capitalisation is $2.253 trillion, effectively unchanged, with Bitcoin dominance at 56.2%.

A fortnight of that would normally make for a thin article. What makes it worth writing is that our own readings and the price parted company inside it, and working out why produced something more useful than another flat market summary.

The divergence, and then the explanation

Here is the test we have been running: does the daily direction agree with which side of 50% the buy share sits on? In late July it did, almost perfectly. In August it stopped.

PeriodAgreeDisagreeFlat
Jul 25 to Jul 31601
Aug 3 to Aug 14444

Four and four is a coin flip. And the individual disagreements are not marginal. On August 7 price rose 0.93% on a 47.8% buy share. On August 10 price fell 1.44% on a 50.9% buy share. On August 14, today, price is down 0.65% on a 53.9% buy share. Those are large moves against the reading, not noise around the line.

Then the ETF data arrived and the whole thing resolved. In the week ended August 7, US spot Bitcoin ETFs recorded net inflows of about $853.54 million, described as the largest weekly total since mid-April, and roughly $693 million of that went into BlackRock's IBIT alone. That is 81% of a week's institutional demand concentrated in one fund.

ETF creations are not taker buys. When money enters a spot Bitcoin ETF, authorised participants source the coin and the fund issues shares. That process does not send market orders across the spread on Binance spot, which is precisely and only what our buy share measures. So during the week price was being set by a bid our instrument is structurally blind to.

The reverse then confirmed it. ETF flows turned negative in the following week, with about $144.67 million out on August 11 ending a five session inflow streak, and about $131 million out on August 13. Price fell through that stretch while our Binance buy shares moved back above 50%: 50.9%, 49.3%, 51.3%, 51.8% and 53.9%. Aggressive buyers were present on Binance the whole way down, and they were the smaller force.

What that means for how we use this metric

This is the part worth keeping, because it is a permanent limitation rather than a bad fortnight.

The taker buy share answers one question precisely: on this venue, right now, which side is paying the spread to get filled? That is genuinely valuable, and the July results were not an accident. It is also, by construction, a measurement of one venue. When the marginal price-setting buyer is an ETF authorised participant, an OTC desk, or a corporate treasury working an order away from the public book, the metric will show you the residual retail and speculative flow on Binance and nothing else, and it can point the wrong way for days.

So the honest framing from here is that our readings tell you who is aggressive on Binance spot, and that this is a good proxy for direction only while Binance spot is where the marginal buyer lives. In July it was. In the first week of August it was not, and the tell was available in public data the whole time: a nine figure ETF week with 81% of it in one fund.

Practically, that means pairing the two series rather than trusting either alone. When the buy share is below 50% and price rises, the first thing to check is whether ETF creations explain the gap. When the buy share is above 50% and price falls, check whether redemptions do. Over these eleven days that single check would have resolved every one of the four disagreements.

Two errors we have to own, and they are the same error

Separate from the venue problem, we published two things in the last fortnight that were wrong, and both failed the same way.

The gold unwind. On August 3 we presented five consecutive falling readings in Tether Gold's taker buy share, 67.9% then 47.2%, 40.3%, 38.2% and 34.8%, described the hedge unwind as "still running," and called it "the clearest available read on where the crypto bid is coming from." It reversed the next day. Tether Gold printed a 61.9% buy share on August 4, gold rose 4.90% on August 5 and a further 2.24% on August 7, and the series has oscillated between roughly 46% and 60% ever since. Today it reads 45.3% with Paxos Gold at 38.9%, and we are explicitly not calling that a trend.

What makes this worse is that we had already identified the error. On July 29, writing about the same gold rotation, we said plainly that "nobody should have extrapolated it into a trend." Five days later we extrapolated it into a trend. Knowing the failure mode and naming it in print turns out not to be the same as avoiding it.

The intraday reading published as a fact. That 34.8% figure was measured mid-session. Tether Gold's full day for August 3 closed at 43.4%. This is the third time in a fortnight we have characterised a session from a partial reading: the same thing happened with Sunday's volume on August 2, which we corrected on August 3, and it has now happened again inside the very article containing that correction. The fix is mechanical and we are adopting it: any figure describing a whole session gets recomputed after that session closes, or it is labelled as partial in the text.

And the July record needs trimming. We published that week as seven agreements out of seven. On a stricter test that treats moves under 0.15% as flat, it was six clear agreements and one flat day, because one of the seven moved just 0.11%. The pattern was real. It was one session less impressive than we said.

The call that did hold: CLARITY

On August 3 we argued that the whole bill came down to one mechanical fact, that no cloture motion had been filed, and that Majority Leader John Thune saying he wanted to "at least begin consideration" was the tell that passage was off the table. That is exactly what happened.

The Senate adjourned on August 8 without a final vote. But early that Saturday morning, after an overnight voting session, Thune filed a motion to proceed on the Digital Asset Market Clarity Act. That is the first procedural step in the cloture process, and it keeps the bill on the docket. Reporting puts the resulting procedural vote at September 15, with the Senate returning on September 14 to a roughly three week floor window. Beginning consideration without passing anything is precisely what he said he wanted.

The blocking issues also came in as described. We said on August 3 that the obstacle was conflict-of-interest provisions rather than market structure, and that the five Wall Street endorsements had not addressed the actual question. CoinDesk now lists three sticking points: government ethics provisions, particularly restrictions on senior officials and President Trump backing crypto projects, stablecoin rewards and yield, and illicit-finance protection details. And the compromise text we flagged as the thing to watch has, per the same reporting, sat unanswered for at least a week at the White House.

One number has moved against us. On August 3 the reported arithmetic was 60 votes against 53 Republican seats, implying about seven Democrats. CoinDesk now reports the bill needs at least ten Senate Democrats, which implies some Republican defections are being assumed. The gap got wider, not narrower.

What actually drove the money: the jobs report

The ETF week did not appear from nowhere, and its cause connects directly to the Fed coverage we have been running since late July.

An unexpectedly weak US jobs report for July, released on Friday August 7, cooled bets on further Federal Reserve rate hikes. That is the same Friday the record $853.54 million ETF week closed on, and it is cited as a driver of the buying.

Put that against the July 29 decision. The Fed held on a 9 to 3 vote, with Beth Hammack, Neel Kashkari and Lorie Logan all preferring a quarter point increase, and we wrote at the time that September would become the live meeting and that the dissenters' case depended entirely on whether the incoming data backed them. A weak labour print does not back them. Combined with core PCE easing to 3.3% and second quarter GDP coming in at 1.5% against 2.1% expected, the hawkish argument inside that committee has had three consecutive data points go against it.

Which produces a genuinely unusual calendar collision. The Fed meets September 15 and 16, with the Summary of Economic Projections and the dot plot returning after two meetings without one. The CLARITY procedural vote is reported for September 15. The two catalysts this site has tracked separately for a month now land on the same day.

Where the board sits today

Today's readings, offered as a description of today and nothing more.

PairTaker buy share, 24h24h change
Bitcoin54.0%0.18% lower
BNB50.6%0.32% lower
Ethereum49.0%0.18% higher
Tether Gold45.3%0.32% higher
XRP45.3%0.57% lower
Solana42.8%0.67% lower
Paxos Gold38.9%0.28% higher

Bitcoin has the strongest aggressive bid on the board at 54.0% and is nonetheless slightly lower on the day, which is the same pattern as the rest of this week and, on the analysis above, is what you would expect while ETF flows are negative. Both tokenised golds are higher on price with weak buy shares. We are reporting that and declining to build a story on it, for reasons the correction section covers.

The longer-term ETF picture remains the constraint worth remembering. Even after the best week since mid-April, US spot Bitcoin ETFs are approximately $4.5 billion in net outflows for the year. One strong week against that backdrop is a change in direction, not a change in position.

What we will be measuring into September

As set out on August 3, this list is scheduled events and things we will measure, with no if-then thresholds attached to price levels.

  • September 15, twice over. The CLARITY procedural vote and day one of the Fed meeting that brings back the dot plot. We will cover them as one session because that is how the market will experience them.
  • Whether the White House answers on the ethics text. It has reportedly sat unanswered for at least a week and requires the President's sign-off to advance as a bipartisan bill. This remains the item on which the Democratic votes are conditioned.
  • Daily ETF flows, paired with the buy share from now on. This is a change in what we publish. Given the venue limitation described above, a buy share reading without the ETF flow beside it is an incomplete picture, and we will stop presenting it as a complete one.
  • The vote count, seven or ten. Reported requirements have drifted from about seven Democrats to at least ten. Which figure the reporting settles on tells you how many Republican defections are being priced.
  • Whether the September dot plot vindicates the three dissenters. Core PCE at 3.3%, GDP at 1.5% and a weak July jobs report have all gone against them. The dot plot is where that gets written down.
  • The $4.5 billion year-to-date ETF deficit. The single number that puts every good week in proportion.

How to read this as a trader

The useful lesson from eleven flat days is about instruments rather than about Bitcoin.

Every measurement has a boundary, and the boundary is usually a venue. Our buy share is computed from Binance spot candles, so it sees Binance spot and nothing else. That was sufficient for most of July, when the marginal buyer was on the book. It was insufficient in the first week of August, when $693 million went into a single ETF and never touched the spread. The metric did not break. It was asked a question about the whole market when it only has evidence about part of it, and the part it cannot see happened to be the part that mattered.

The practical form of that is a habit rather than an indicator: when your best instrument disagrees with price for more than a session, stop asking whether the instrument is broken and start asking who is transacting somewhere you are not looking. In this case the answer was published daily and freely, in ETF flow data, the entire time.

And the second lesson is duller and probably more expensive. Twice in a fortnight we described a session using a number measured before that session closed, and once we extrapolated a five point series into a trend that reversed within a day, having warned against exactly that five days earlier. Neither error came from bad data. Both came from wanting the story to be tidier than the evidence. The fixes are mechanical: recompute after the close, and refuse to name a trend from five observations. Our live dashboard separates aggressive buying from aggressive selling minute by minute across the most liquid pairs, and it is precise about what it measures, which is one venue. For the run-up to this, see the arithmetic and the missing cloture petition, the endorsements and the falling odds, July's 7.24% on a 49.5% buy share, and the Fed's 9 to 3 hold. If the buy and sell split is new to you, the volume guide covers the mechanics.