Markers Sunday: The Difficulty Projection Swung 1.79 Points in 24 Hours and Killed Two Markers We Called Unresolvable — and We Were Wrong About the 30-Year

Every Sunday this column grades the falsifiable claims it has put in print and sets new ones. This week the largest single move on the board came from the part of bitcoin that is supposed to move slowly. At 06:20 UTC on Sunday the network’s projected difficulty adjustment stood at +0.8494%, with the current period 51.98% elapsed. Twenty-four hours earlier, at 44.05% elapsed, the same projection read −0.9414%. That is a swing of 1.79 percentage points in a single day (own mempool.space pulls, same hour both days). It resolves two markers we recorded on Saturday as unresolvable, and it fails both of them.

Correction: the 30-year rose. We printed that it fell.

Before the scorecard, the correction, because it changes a grade we published. On Saturday we graded marker Y1 — a 30-year Treasury close at or above 5.35% by 28 August — as a failure, and wrote that the yield had fallen, citing CNBC’s 5.168%. We pulled the US Treasury’s daily par yield curve series directly this morning. It does not say that.

US Treasury constant-maturity par yields, own direct pull of the 2026 series, 30 August 2026. Basis-point spreads computed from the printed cells.

The 30-year rose three basis points on Friday, to 5.22%, a fourth consecutive daily increase. The two-year rose fourteen. The curve flattened by eleven basis points, which is a larger and cleaner bear flattener than the one we described. Y1 still fails — 5.22% is thirteen basis points short of its 5.35% bar, not the 18.2 we printed — so the marker’s verdict survives. The sentence around it did not. The mechanical cause is worth naming because it will recur: CNBC’s 5.168% is a live market quote on the on-the-run 30-year bond, and Treasury’s 5.22% is a constant-maturity par yield. They are different series measuring adjacent things, and 5.168% is closest to Treasury’s 25 August cell. We took a real-time quote as a daily close. Rule added: a marker denominated in a daily close settles on the official daily series that publishes that close, named in advance.

The board

C3 and D2: from ‘unresolvable’ to failed in one day

For most of last week the difficulty projection sat between −0.9% and −1.6% with the period less than half elapsed, and we declined to grade C3 (a cut deeper than 2.00%) or D2 (a retarget landing between −1.50% and 0.00%) on the grounds that a projection taken before the halfway mark is mostly noise. That caution was correct and it has now been vindicated in the least comfortable way: the projection crossed zero and kept going.

Own mempool.space pulls at the stated timestamps. Current tip height 964,696; retarget height 965,664; previous retarget −1.3122%; three-day average hashrate 900.1 EH/s; fees 1 sat/vB across every tier.

The mechanism is not mysterious. Blocks came fast enough over the past day to pull the period’s average interval below ten minutes, and once the period is past halfway each additional fast block moves the projection less but moves it more durably. C3 needed a cut deeper than 2.00% and the projection is positive: it is now dead, and it will not come back to life in 968 blocks. D2 needed a landing between −1.50% and 0.00%; +0.8494% is outside that band on the wrong side, and while D2 is not yet mathematically closed, recovering it would require the remaining 968 blocks to run substantially slow. Both graded honestly: called unresolvable yesterday, resolvable today, failing today. The premise behind both — that August’s price weakness would show up as miners capitulating — is now retired, not rolled. Hashrate at 900.1 EH/s is the counter-evidence.

Funding: Saturday spent the entire day pinned at the cap

Our standing funding pull is 500 settlements of Binance BTCUSDT perpetuals, the endpoint maximum, which today reaches back to 16 March 2026. In that window the maximum is exactly 0.010000% and nothing is above it; the minimum is −0.012276%; 370 of 500 settlements (74.0%) are positive; and 41 print at exactly the cap, up from 39 on Saturday. Both new cap prints landed on Saturday.

Yesterday we reported the cap prints as a cluster and left the structure unexamined. Today we counted the runs, and the structure is the finding. The 41 cap prints are not scattered: they form twelve runs of one, eight runs of two, one run of three, and one run of ten. The run of ten began at 16:00 UTC on 21 August and spanned three and a bit days — which means Saturday’s three-in-a-row, which we were about to describe as unusual clustering, is the second-largest run in the window and less than a third of the size of the record set nine days ago. Correcting our own framing before it hardens.

The genuinely rare thing is narrower and we can state it exactly. Grouping the window by calendar day gives 166 complete days of three settlements each. On four of those 166 days — 22, 23, 24 and 29 August — all three settlements printed at exactly the cap. That is a base rate of 2.4%, and every single instance falls inside the last nine days. A full day pinned at the ceiling is not a feature of this market; it is a feature of the last nine days of this market. The run ended at 00:00 UTC on Sunday with a print of 0.008283%.

One structural detail that keeps D1 alive and that we had not previously printed. The 0.0100% level is not a symmetric hard cap. The series has printed below −0.0100% twice in 500 settlements, reaching −0.012276%, while never once printing above +0.0100%. If the ceiling were a hard exchange clamp the floor would mirror it and does not. So D1 — any settlement above 0.0100% before 30 September — is improbable rather than impossible, which is the correct state for a marker to be in. It stays open.

Open interest has now round-tripped twice in three sessions

Binance BTCUSDT futures open-interest history, own pull. Daily rows are 00:00 UTC snapshots, not session closes.

Three consecutive daily moves of more than 1.9% in either direction, and the net across all three is +1.97% — 2,079 BTC on a base of 105,594. On Friday this column recorded “the leverage arrived” as its market learning; on Saturday it recorded that the leverage had left within twenty-four hours and set the rule that a single-session extreme in open interest is a position, not a regime, until it survives the event it was placed for. Sunday is the third data point and it argues the same thing from the other direction: the position has been put back on. What has actually happened over four sessions is that roughly three thousand bitcoin of open interest has been added and removed twice while the underlying moved 2.9% down and 0.6% back up. That is churn, and calling it a build or a flush in either direction would be the same error twice more. E3 remains failing: 108,447.96 against a 115,000 bar is 6.04% away.

A window error we caught in our own cohort data

The daily long/short cohort pull returned, for 30 August, a top-trader position ratio of 2.0826 and an all-accounts ratio of 1.1877. The all-accounts figure is identical to four decimal places to the 29 August row. Two consecutive days agreeing to 0.0001 on a ratio that moved 29% the day before is not a market observation, it is a data artefact, so we pulled the hourly series to check. The daily bucket labelled 30 August is the 00:00 UTC snapshot, not a live reading. The hourly series shows the ratios have moved since: all-accounts from 1.1877 at 00:00 to 1.1358 at 06:00, top traders from 2.0826 to 2.0603, widening the spread from 1.753× to 1.814×.

This is the third week running that a window has produced a fake finding on this desk — a streak counted against a fixed-depth pull, a word count taken across a page boundary, and now a daily bucket read as intraday. The failure mode is identical every time and the fix is identical every time. Standing practice extended: any figure labelled with today’s date now ships with the timestamp of the snapshot it actually is. The direction of travel over the weekend, on the hourly data, is a crowd getting less long, not more.

B2 now depends on which tracker you settle it against

B2 requires August net inflows into US spot bitcoin ETFs of at least $3,424.9 million, settling on Monday’s figures. Through 28 August the month stands at $3,322.4m on Farside Investors, which needs $102.5m on Monday. It stands at $3,306.1m on SoSoValue, which needs $118.8m. The two trackers differ by $16.3 million on the same month — a rounding-scale gap in absolute terms, and a 16% difference in the size of the remaining task.

We are declaring the settlement source rather than discovering it on Tuesday: B2 was set on Farside and settles on Farside. Two operational notes for anyone checking our arithmetic. Farside’s page currently displays a rolling window of fourteen August sessions, 11 to 28 August, summing to $2,601.7m; the six earlier sessions (3 to 10 August, $720.7m) come from our verified pull of 29 August and are carried forward, which is why the month total is not reproducible from a single screenshot today. And across those fourteen displayed sessions the concentration is extreme: IBIT accounts for +$2,243.8m against +$242.3m for FBTC, +$157.0m for the Grayscale Mini and −$92.7m for GBTC. Friday’s −$201.9m, of which ARKB was −$114.9m, remains the only negative session in the visible window.

Three things settle tomorrow

A2 asks for an August close at or above $71,440.63. At $78,119.23 it is passing by 9.35% and would need an 8.55% collapse inside one session to fail; we are calling it now rather than pretending to suspense. B2 needs $102.5m of net inflow. B3 asks whether Strategy’s Monday 8-K discloses a bitcoin purchase greater than zero. The company last bought on 22 June — 520 BTC at about $67,068 — which is 69 days ago, and has sold roughly 6,916 coins since, taking holdings from 847,363 to about 840,447, or close to 4% of the 21 million cap. Its 8-K of 24 August disclosed $5.10bn in a “USD Reserve” plus $1.59bn of newly created “USD Cash”: $6.69 billion of liquidity that has bought no bitcoin for ten weeks. We have no view on what the filing will say and are not pretending to one; that is what makes it worth writing down in advance.

Three new markers for September

F1 — the retarget prints positive. The difficulty adjustment executed at block 965,664, estimated for 5 September, comes in at or above 0.00%. Settles on the realised adjustment reported by mempool.space, not on any projection. This is the direct successor to C3 and D2 and it is deliberately set on the side the evidence now favours, so that a failure is informative.

F2 — sentiment leaves the Greed band. The Crypto Fear & Greed Index prints below 55 on at least one day on or before 6 September. It has now spent eleven consecutive days at 55 or above. Settles on the alternative.me daily series.

F3 — Strategy buys again. Strategy discloses a bitcoin purchase greater than zero in any 8-K filed on or before 28 September. Settles on EDGAR. F3 is deliberately wider than B3: B3 asks about one filing, F3 asks whether the ten-week pause is a pause or a policy.

Scorecard

Two markers closed today, both failed, both graded from a projection that reversed on us. One published sentence corrected against a primary source. One of our own framings corrected before it hardened, on funding run-lengths. One data artefact caught in our own pull and traced to a window definition. The pattern across the week is consistent and not flattering: the arithmetic keeps coming out clean and the question design keeps coming out wrong. Recomputing a number verifies the number. It does not verify that the number was worth computing, and no amount of checking will catch a marker that was set against the wrong series or read off the wrong bucket. That is a design problem and it needs a design fix, which is why every marker set from today names its settlement source in the same sentence as its bar.

Method: monthly returns in this article are computed by Bitcoin Mastery as first-of-month open to last-of-month close, UTC, on Bitstamp BTC/USD daily candles — 4,990 rows from 1 January 2013 to 30 August 2026, plus a separate extension pull covering August 2011 to December 2013. Partial months are excluded and identified. Prices, flows, funding, open interest and on-chain figures are pulled directly at the timestamp stated. Where a third-party figure is cited we name the source and its date; where two sources disagree we print both.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. Cryptocurrencies are volatile and you can lose money. Historical seasonality describes the past; it does not forecast the future. Do your own research and consult a licensed financial advisor before making investment decisions.