Markers Monday: Bitcoin’s $1.44bn of New Open Interest in August Was Entirely Revaluation — the Coin Count Fell 2,196 — and We Were Wrong About the Greed Record

For three days this desk has owed its readers a number. On Thursday we retired the funding-rate evidence for the claim that August’s rally was “unlevered.” On Friday the replacement — a one-day fall in open interest — round-tripped. On Saturday it round-tripped again. On Sunday we called it churn and wrote the homework down as unpaid for a third time: re-derive the unlevered-rally frame from the cohort spread and the basis, or stop implying it.

It is paid this morning, and the answer is better than the frame was.

The find: August’s $1.44 billion of new open interest is entirely revaluation

Pull Binance BTCUSDT perpetual open interest as daily snapshots across the whole month, in both units the exchange publishes — coins and dollars — and the decomposition is unambiguous.

The two changes multiply out exactly: 1.23671 × 0.97976 = 1.21168, against a published dollar change of +21.168%. Not approximately. Exactly, to five decimal places.

So: the dollar value of open bitcoin perpetual positions on Binance rose by $1.44 billion in August, and every cent of that increase — and $170 million more — came from the price of coins that were already on the book. The number of coins fell by 2,196.15. The rally did not attract new perpetual leverage. It was financed somewhere else.

That is the datum the frame needed, and it is worth noting precisely what it does and does not license. It does not license the word “unlevered,” which was always too strong and is now demonstrably wrong — see the carry section below, where existing longs are paying real money. What it licenses is narrower and more useful: leverage did not grow into this move. Notional exposure is flat to down while spot is up 24%. Anyone reaching for a “this is all leverage” explanation of August has to explain 2,196 fewer coins on the largest perpetual book in the market.

The frame is therefore retired in its loose form and replaced with the specific one. Three days late, but with a number that survives arithmetic.

And where the leverage that does exist is being paid for

Two prices, both taken at 06:13 UTC on 31 August from Binance COIN-M, each contract measured against its own index rather than against a spot venue:

Two tenors, four and a half months apart, annualising within 27 basis points of each other. That is a flat carry curve, and a flat carry curve is what a market looks like when nobody is bidding aggressively for term leverage.

The comparison that gives it meaning is the risk-free leg. On 28 August the US Treasury constant-maturity par yield was 3.90% at three months and 3.94% at four months (own pull of the Treasury daily par yield curve). So the cash-and-carry premium over bills is +145bp at the September tenor and +114bp at December. Positive — there is a real carry trade here — but thin. This is not the 10–20% annualised basis that characterises a leveraged bitcoin bull market.

The perpetual tells a slightly different story, and the difference is itself informative. Across the last 90 funding settlements the mean rate is 0.006602% per eight hours, which annualises to 7.23% — higher than either dated contract. The leverage premium in this market is concentrated in the perpetual, not in the term structure. Longs holding perps are paying about 7.2% a year for the privilege; longs willing to hold a dated contract to September are paying about 5.3%. Both are being paid. Neither is being paid enthusiastically.

Cohorts: the size money is long, the crowd is not

Binance’s three long/short ratios at the 31 August daily snapshot: top traders by position 2.0586, top traders by account 1.1580, all accounts 1.0812.

The spread that matters — top-trader positions minus all-accounts — is 0.9774. Its ten-day range is 0.8949 to 1.2552. Today sits in the lower half of that range, and the direction of travel over the last three sessions is down: 1.1511, 0.9562, 0.8949, 0.9774. The size money is still roughly two-to-one long while the crowd is barely off flat, but the gap has been closing, not widening, since the 28 August drawdown.

One caveat this desk has printed before and repeats because it keeps mattering: these are daily cohort buckets, not intraday readings. A ratio dated 31 August is a snapshot, not a session.

Correction: our “longest greed streak” figure was a pull-depth artifact

Yesterday we published that bitcoin’s current run of Fear & Greed readings at or above 55 was “the second-longest of thirteen months,” against a window record of 13 days set 7–19 August 2025. The 13-day figure is real, but it is not the record, and calling it one was our own error.

The cause is the defect this desk named as a standing learning three weeks ago and then walked into again: a streak measured inside a fixed-depth pull is a statement about the depth of the pull. Sunday’s figure came from 400 readings, a window that opens on 27 July 2025. There is a greed run of 40 consecutive days from 24 June to 2 August 2025 whose tail pokes into that window; the pull saw six days of it and the 13-day August run looked like the maximum.

Pulled properly — the full alternative.me history, 3,130 daily readings from 1 February 2018 to 31 August 2026 — there have been 98 separate runs of at least one day at or above 55. The current run is 12 days. Here is where that actually ranks:

The current streak is the 23rd-longest of 98, not the second-longest of anything. The record is eight times its length. Two secondary figures from Sunday move with it: the window maximum was reported as 75 on 14 August 2025, and over a 420-day window it is 79, set 12 July 2025; the window mean was reported as 30.8 and over 420 days it is 32.7.

What survives the correction intact is the claim that actually carried the article: 2026 has produced 243 daily readings and only 13 of them have printed at or above 55 — twelve of those are the current run, and the thirteenth was 15 January. That is a fact about this year and it does not depend on window depth. The streak is unusual for 2026. It is thoroughly ordinary for the index.

Monday’s reading, incidentally, is 62 — down seven points from Sunday’s 69, the largest single-day fall of the run, and now seven points from settling marker F2.

Three markers settle on tonight’s prints

This is the heaviest settlement day of the month, and none of the three is decided at press time.

A2 is effectively decided. An 8.38% single-day fall has happened in bitcoin’s history, but assigning it meaningful probability on a month-end Monday would be theatre. We will grade it on tonight’s settled close rather than call it now.

B2 is the interesting one, and the bar is not arbitrary. $3,424.9m is exactly the net inflow US spot bitcoin ETFs took in October 2025 — the most recent month larger than August 2026 has managed so far. August currently stands twelfth of the 32 months since these funds launched. Clearing the bar makes it eleventh, and makes it the best month since September 2025 ($3,511.0m). Missing it leaves August as the best month since October 2025, which is still comfortably the best month of 2026: April, the year’s runner-up, managed $2,021.7m.

A number worth holding next to that: 2026 is still net negative on ETF flows, at −$1,894.6m year to date. August has recovered 63.7% of a deficit built up over seven months. The settlement source for B2 was declared in print on Sunday and does not move: B2 was set on Farside and settles on Farside, notwithstanding that SoSoValue has August at $3,306.1m, a $16.3m disagreement that would move the required amount to $118.8m on the other tracker.

B3 got materially more interesting on Sunday, and it is covered separately in today’s fourth article. In brief: Michael Saylor posted Strategy’s accumulation chart to X with the caption “We’re ₿ack” — a pattern that has historically preceded Monday purchase disclosures. The filing typically lands around 08:00 ET, which is after this article publishes. B3 settles on the document, not on the meme.

On chain: the projection retraced, and the retarget slid to Sunday

Tip height 964,835; retarget height 965,664; estimated 6 September 00:00 UTC, which is about two hours later than yesterday’s estimate. Previous retarget −1.3122%. Fees remain at 1 sat/vB across every tier — the eleventh consecutive day this desk has recorded an empty mempool at every priority level.

Yesterday the projection moved 1.79 points in 24 hours and we used that to kill two markers. Today it moved −0.53 points in the opposite direction, on seven more points of elapsed period. The honest reading is that a projection at 59% elapsed still carries roughly half a point of noise, and the correct lesson from yesterday was not “the projection reversed” but “a projection is not a reading.” F1 — that the realised retarget prints at or above 0.00%, settling on mempool.space — is currently passing with 829 blocks to run. D2, which needs a print between −1.50% and 0.00%, is failing on the wrong side by a third of a point and is now within noise of its band.

The hashrate deserves a sentence of its own. 904.5 EH/s is a new high for the series this desk tracks, on the third day of a month in which bitcoin gained 24%. The miner-capitulation premise that C3 and D2 were built on is not merely retired; it is being contradicted daily.

Funding: one new cap print, and the asymmetry holds

500 settlements back to 16 March 2026, refreshed this morning. Maximum exactly 0.010000%, count at the cap now 42 (41 yesterday; the new print landed 30 August 08:00). Zero settlements above the cap. Two below −0.0100%, minimum −0.012276%. 371 of 500 positive, mean 0.002769%.

The cap asymmetry we first printed on Sunday is unchanged and remains the single most useful thing in this series: the rate has printed below −0.0100% twice and above +0.0100% never. A genuinely symmetric clamp would have produced a mirrored floor. It did not. D1 — any settlement above 0.0100% before 30 September — is therefore improbable rather than impossible, which is the state a marker should be in.

Full-cap days are unchanged at four of 166 complete days in the window: 22, 23, 24 and 29 August, a 2.4% base rate with all four instances inside the last ten days. Sunday broke the pattern — 30 August printed 0.008283%, 0.010000%, 0.007058% — and Monday’s 00:00 settlement came in at 0.007271%.

The marker board

Two new markers, each naming its settlement source in the same sentence as its bar — the convention adopted on Sunday after a marker was graded against the wrong Treasury series.

G1: Binance BTCUSDT perpetual open interest, daily snapshot, closes any September session below 100,000 BTC. Settles on the Binance futures openInterestHist endpoint, 1d period. This is the falsifiable version of this morning’s find: if positions genuinely are not being added, the coin count should be capable of falling further, not merely failing to rise.

G2: the Binance COIN-M September quarterly basis, measured mark against its own index, annualises above 8.00% on any day up to 25 September, settling on the Binance dapi/v1/premiumIndex endpoint. That is the level at which the carry story stops being “thin” and starts being a leverage story. At 5.349% this morning it needs a substantial repricing, and if it never gets one, the flat-curve reading published above is the one that stands.

What we still owe

One thing, and it is a design problem rather than a data problem. The homework paid this morning was paid with a monthly decomposition, and monthly decompositions are exactly the kind of statistic that flatters whoever chooses the endpoints. 2 August and 31 August were chosen because they are the ends of the pull, not because they are meaningful. The intra-month path was not flat: open interest peaked at 111,988.29 BTC on 15 August and bottomed at 105,531.28 on 24 August, a 5.77% range that the two-point comparison erases entirely.

The conclusion survives — the high, the low and the endpoints are all below where the month started, so no endpoint choice inside August produces a coin-count increase. But the sentence “every cent came from revaluation” is a statement about two days, and the reader is entitled to know that. It goes on the board as standing practice: any two-point decomposition ships with the range it is hiding.

Disclaimer: This article is for information only and is not investment advice. Bitcoin Mastery is not a financial adviser. Cryptocurrency and derivatives are volatile and you can lose the whole of your capital. Every figure here names its source and the time it was taken; do your own research and consider taking independent professional advice before acting.