A line it had not closed above since November

The 50-week moving average is the average of the last fifty completed weekly closes. It is slow on purpose. One number summarises about a year of price, which is why crossing it registers as an event rather than noise.

Thursday's close of $81,270.37 was above it. The line for this week, on completed weekly closes only, sits at $81,070.91. That is a margin of $199.46, or 0.25%.

Working back through every daily close on Binance, the previous close above that line was November 11, 2025. The gap is 296 days, and there were no other closes above it anywhere in 2026.

It lasted one session. Friday closed at $79,660.77, which is $1,410 below the line. Saturday closed at $79,831.75. Sunday is partial as we write and trades near $79,840, essentially flat on the session.

The line is also moving

Week beginning50-week averageChange from prior week
Aug 10$83,398
Aug 17$82,491$907 lower
Aug 24$81,823$668 lower
Aug 31$81,071$752 lower

That is $2,327 lower in three weeks, an average of $776 a week, because the strong weeks of late 2025 keep dropping out of the back of the fifty week window. On August 26 we measured this line at $81,823 and on Thursday at $81,071. Both were right for their week. When this week's candle closes tonight, the line moves again, to roughly $80,363.

So the ceiling is descending toward the price at about $700 a week. Reaching it does not require a rally. It requires patience, or nothing at all.

The week is still up 2.78% from its $77,682 open. A green week that fails at the line is not the same thing as a red week, and we would rather state both facts than the convenient one.

The busiest hour of the week went down

The August employment report was released on Friday at 12:30 UTC. It showed 162,000 jobs added, close to triple the 53,000 consensus, with unemployment steady at 4.1% and the June and July figures revised higher. Fed funds futures moved September hike odds from 49.4% to 58% ahead of the September 15 and 16 meeting.

Here is the hour by hour tape on Binance BTCUSDT spot around the release.

Hour, UTCSessionSpot volumeBuy share
Fri 11:00+0.06%$41.6M52.8%
Fri 12:002.18% lower$295.3M40.6%
Fri 13:000.09% lower$153.0M46.2%
Fri 14:000.58% lower$143.8M45.3%
Fri 15:00+0.65%$74.7M57.3%
Fri 16:00+0.40%$84.4M51.2%

The release hour fell 2.18% on $295.3 million at a 40.6% buy share, which means aggressive sellers took 59.4% of it. That is not a drift. That is people hitting bids.

Now set it beside the rally it undid. Thursday's two working hours were 14:00 at $200.4 million and 15:00 at $246.9 million, the largest hour of that session. Friday's single release hour traded 19.6% more than the biggest hour of Thursday's advance.

Across all four sessions from Thursday to Sunday, the busiest single hour of the week is a down hour. That is the cleanest one line summary of the week we can give you.

Buyers did arrive afterwards. The 15:00 hour gained 0.65% at a 57.3% buy share and 16:00 added 0.40% at 51.2%. They stopped the fall. They did not take it back, and the close finished the day 1.98% lower at $79,660.77.

Two days ago our own tools disagreed. The next two sessions picked one.

On Thursday we published that the buy share and open interest read the same session differently, and that we would show both rather than choose the flattering one. Here is what each said, and here is what followed.

The buy share said conviction: 54.7%, above our 90th percentile of 52.4%, clearly separated from the three failed attempts at 49.8%, 51.0% and 47.1%. Open interest said covering: down 0.60% on the same session, meaning contracts were being closed rather than opened.

SessionCloseChangeBuy shareOpen interestLong/short
Thu Sep 3$81,270+5.08%54.7%0.60% lower1.2148
Fri Sep 4$79,6611.98% lower48.4%+4.71%0.7816
Sat Sep 5$79,832+0.21%46.7%4.26% lower1.0165
Sun Sep 6, partial$79,842flat45.3%1.54% lower1.0475

Price gave back $1,438.62 of the $3,930.36 gain, which is 36.6%. The buy share fell on three consecutive sessions, from 54.7% to a partial 45.3%. And on the down session, open interest rose.

The covering read is the one that survived contact with the following days.

We want to be exact about what that does and does not mean. It is one resolution of one disagreement. It makes open interest the tool that was right this time, not the better tool in general. We have spent two weeks writing that four observations do not make a rule, and one observation makes even less of one. What it does establish is narrower and more useful: when we said we would not pick, not picking was the correct answer, because we had no basis for picking and the data settled it 48 hours later at no cost to anyone.

Friday's fall was positioning, not panic

That open interest number deserves its own paragraph. On a session where price fell 1.98%, open interest rose 4.71%, from 107,652 BTC to 112,718 BTC. It is the largest single day increase in the thirty days Binance publishes. Falling price with rising open interest means new contracts opened into the weakness. That is people taking a side on the September 15 and 16 meeting, not people being forced out of one.

Then Saturday reversed it: open interest fell 4.26%, the largest single day decrease in the same thirty days. The biggest build and the biggest unwind of the month landed on consecutive sessions.

The account ratio agrees. It settled at 0.7816 on Friday, meaning 56.1% of accounts were positioned short. On Thursday we reported a live intraday reading of 0.7979, and the session finished more skewed rather than less. By Saturday it was 1.0165 and by Sunday 1.0475, so the skew lasted about as long as the line did.

We have now flagged a sub parity reading three times. Once a squeeze followed within a day. Once nothing happened. This time the skew simply unwound with the price roughly flat. Three instances, three different outcomes, which is our own evidence that this configuration describes the present rather than forecasting the future.

The biggest up day of the run was also the thinnest

Here is a measurement we had not made before this week. Take the twelve largest single day gains in the last 400 sessions, then sort them by how much actually traded.

SessionGainCloseSpot volume
Feb 6+12.19%$70,580$6.13B
Aug 21+7.27%$78,338$3.40B
Mar 4+6.33%$72,667$3.20B
Aug 20+5.32%$73,025$2.56B
Dec 2, 2025+5.79%$91,278$2.52B
Mar 2+4.64%$68,830$2.16B
Jan 13+4.51%$95,414$2.15B
Feb 25+6.13%$67,988$2.06B
Mar 23+4.49%$70,906$1.98B
Aug 19+7.12%$69,335$1.95B
Apr 13+5.20%$74,418$1.75B
Sep 3+5.08%$81,270$1.58B

Thursday is last. Its $1.58 billion is the smallest of the twelve: 46.4% of the August 21 session thirteen days earlier, and about a quarter of the February 6 session.

Put another way, across those same 400 sessions Thursday's volume ranks 255th, the 64th percentile. A session in the top 2.3% by size traded in the middle of the pack by activity.

This is not a criticism of the move. Price is price and the close was real. It is an observation about who was in the room. The largest advance of this run happened with fewer participants than any comparable advance in fourteen months.

And the same session set an ETF record

Reporting puts US spot Bitcoin ETF inflows at $730.9 million on Thursday, the largest single day since January 14. We did not measure that and we are relaying it.

Side by side those two facts look contradictory: the biggest ETF day in almost eight months landed on the thinnest big up day in 400 sessions. They are not contradictory. This is the first of the three blind spots we have documented on this site. ETF creations settle between the issuer and its authorised participants, away from the public order book. Whatever they bought, and wherever they bought it, our taker buy share never saw it.

So the thin tape is not evidence that nobody was buying. It is evidence that whoever was buying was not doing it where we can watch.

What people are actually searching for

Attention is the other half of a market, so it is worth measuring too. These are third party search estimates rather than our own data, and we label them that way for the same reason we label a partial session.

Search termEstimated monthly volumeReported change
Bitcoin3.35M33% lower year over year
Day trading246Kabout 5x
Clarity Act201Kabout 25x
Crypto ATMs110Kmore than 12x year over year
Tokenization18.1K+22%
AI trade14.8K+83%
Prediction markets14.8Kvery large, off a small base

On the normalised search index, global crypto interest had fallen to a band of 26 to 30 by June 2026 against the August 2025 peak.

The pattern is not that attention left. It is that attention moved: away from the price of the asset and toward the rules around it, the machines under it and the places to trade it. The single fastest growing term on that list is a piece of legislation.

And this is where the previous section earns its place. We can check the exchange analogue of that claim ourselves, and it agrees. The largest up day of this entire run traded less than half the volume of a smaller up day thirteen days earlier. Two independent measures of attention, one from a search engine and one from an order book, describing the same period the same way. We would not lean on either alone. Together they are worth a paragraph.

The most searched question, answered with numbers

The most searched crypto question of 2026 is some variant of why is Bitcoin down today. For Friday, September 4, the answer has four parts and only the last one is guesswork.

The trigger is reported. Payrolls at roughly triple the forecast, and September hike odds moving from 49.4% to 58%.

The mechanism is measured. A 2.18% fall in a single hour, on $295.3 million, at a 40.6% buy share. Sellers were the urgent side and they were urgent for sixty minutes.

The character is measured. Open interest rose 4.71% that day, the largest daily build in thirty days. Contracts opened rather than closed, so this was positioning ahead of a meeting, not forced liquidation.

The fourth part is what happens next, and we do not do that. The numbers above are the reason. The same buy share that separated Thursday's success from three failures then fell for three straight sessions without telling anyone where the price would be on Monday.

What we cannot see this week

Three things move price that no taker metric on any venue can observe, and all three were live across these four sessions.

ETF creations. $730.9 million on Thursday, settled off the public book. Our buy share cannot see it, and neither can anyone else's.

Futures liquidations. The perpetual book is roughly fifteen times the size of the spot book we measure. Friday was less liquidation driven than usual, since open interest rose rather than fell, but the structural point holds. Our spot tape is the smaller room.

Maker repricing. When a resting order is cancelled and replaced lower, no trade happens and nothing is recorded. Price can travel a long way on very little volume this way, and the 12:00 hour on Friday almost certainly contained some of it.

There is a fourth limit, and the search table quietly points at it. We measure Binance. If attention is genuinely moving toward regulated venues, crypto ATMs and prediction markets, then a growing share of what matters is happening somewhere our data does not reach. That is not a flaw we can fix with a better query. It is a property of the instrument, and the honest response is to name it rather than to widen the claim.