The Backroom Brief

Ten Stories

1. The breakeven fell on the day the Fed hiked, and that is the whole story

Two numbers from the Treasury's own daily curves settle what Wednesday meant. The ten-year real yield rose six basis points to 2.68 percent. The ten-year breakeven, which is the nominal yield minus that real yield, fell five to 2.33. Inflation compensation went down while the real cost of money went up.

That pairing matters because the alternative signature looks nothing like it. When a market is pricing an inflation shock, the breakeven rises and the real yield stays flat or falls, which is exactly the pattern this desk misread on 2 September and had corrected by four reviewers. Wednesday is the mirror image. The front end carried the move: the two-year rose seven basis points to 4.74 percent against the ten-year's one, to 5.01, and the par 2s10s flattened from 33 basis points to 27. The twenty-year sat at 5.39 and the 20s30s inversion held at minus four.

What it means is that the bond market took the projections at face value. The Fed said one more increase is the median view, and 16 of the 18 participants marked at least one, per Figure 2 of the projection table. The curve responded by lifting the part the Fed controls and leaving the part it does not roughly where it was. On the most favourable reading that is a market accepting a tightening cycle rather than bracing for prices to run away. Two other readings fit the identical data. A market pricing weaker demand produces the same falling compensation. So does a supply story: this desk reports on the same page that Norway's manager wants to hold less government paper, that Tuesday's twenty-year left dealers with 16.74 percent of the issue, and that nineteen billion dollars of inflation-protected paper auctions today. Scarcer buyers of duration lift a real yield mechanically, and dealers making room before an inflation-linked auction compress a breakeven for reasons that have nothing to do with belief. One session cannot separate the three.

Watch the breakeven, not the headline yield. If the ten-year breakeven turns back up through 2.40 while the real yield keeps climbing, the reading flips from credible tightening to a market pricing a Fed that is behind. Thursday morning has both easing off: the CNBC benchmark ten-year is 4.988 percent and the two-year 4.707, each down a couple of basis points, though that is a different vendor series from the Treasury par curve above and the two never belong in one table.

2. Hong Kong had to hike and Brazil got to cut, within hours of each other

The Hong Kong Monetary Authority raised its base rate 25 basis points to 4.25 percent on Thursday morning, its first increase in more than three years. It did not decide to. The Linked Exchange Rate System pegs the Hong Kong dollar between 7.75 and 7.85 to the US dollar, and the base rate follows the Fed through a formula. The currency barely moved, at 7.8451 against 7.8447, which is the mechanism working exactly as designed.

Brazil's Copom cut the Selic to 13.75 percent on Wednesday, a fifth consecutive quarter-point reduction, unanimously, in line with 48 of 51 economists polled by Reuters. Valor Economico put the two decisions in one headline and framed them as opposite paths. It is the cleanest illustration available of what a reserve currency does to everyone else: an economy with a peg imports the tightening whether it wants it or not, and an economy with a floating currency and disinflation gets to go the other way on the same afternoon.

The stake is distributional. Hong Kong now runs a 4.25 percent base rate into a property market and a banking system that did not ask for one, while the Hang Seng fell 0.52 percent to 24,586.32 at the close. Korea is caught in between: the Bank of Korea sits at 3.00 percent, the gap to the US has widened to a full percentage point, the won fell 0.71 percent to 1,386.71, and the Korea Times led its paper on the pressure that now sits on the BOK. Seoul opened up 0.91 percent and closed down 0.04, giving the whole gain back inside one session.

Watch the Bank of Korea and the RBA. Australia's cash rate is 4.35 percent, already above the new US range, with the next decision on 28 and 29 September. If Seoul moves before its scheduled meeting, the peg-and-gap mechanism has become a genuine constraint rather than a talking point.

3. Crypto rose into the hike, and ether's funding flipped to the other side of the book

Every one of the fourteen major coins this desk tracks closed the 24-hour window higher. Bitcoin rose 1.24 percent to $76,412, ether 2.30 percent to $2,441.15, solana 3.51 percent to $99.96, and the smallest names led: BONK up 6.99 percent, SUI 5.54, AVAX 4.72, LINK 4.29, all per CoinGecko. A tightening central bank and a broad bid in the highest-beta corner of risk is not the pairing the textbook offers.

The mechanism underneath changed more than the price did. Binance ether funding is now positive at 3.694 percent annualised, which means longs pay shorts. Yesterday it was negative at 4.62 percent, which meant shorts paid longs. That is a full reversal of the setup this desk built a card on, and it happened while the all-account long-short ratio fell from 3.30 to 2.9952, or 75:25 (3.00) in the printed convention. The crowd got slightly less crowded and started paying for the privilege instead of being paid.

The counterweight sits in the volume column. Bitcoin's 24-hour traded volume fell 23.58 percent against the same window a day earlier, ether's 17.81 percent, and XRP's 38.57 percent even as every price rose. A rally on shrinking participation is a thinner thing than a rally on rising participation. On the seven-day view eleven of the fourteen are still negative, bitcoin at minus 2.6 percent and ADA at minus 7.7, so this is a bounce inside a down week rather than a trend turning.

Watch $2,500 on ether. Friday's Deribit expiry has max pain at exactly that strike, computed at this desk from the full book rather than taken from a vendor, and it sits about 2.4 percent above spot. If ether reclaims it while the account ratio stays above 3.0, the crowding card that has run since Tuesday is dead on its own stated terms.

4. Norway's fund asked permission to cut its government bond book by a fifth of the benchmark

Norges Bank Investment Management sent its owner a recommendation on 1 September proposing that the government bond share of its fixed-income benchmark fall from 70 percent to 50 percent, with non-government credit rising from 30 to 50, mortgage-backed securities entering at about 13 percent from nothing, and the yen weight going from roughly 5 percent to 8. That is the primary document, published on NBIM's own submissions page.

What it is not is a trade. This is a manager writing to Norway's Ministry of Finance, and the reallocation needs the ministry and ultimately the Storting to sign off. CNBC reported finer sub-figures in early September, putting US Treasuries at 34.1 percent falling to 21.9 of the bond portfolio, but those four numbers could not be found inside the primary extract and remain secondary. This desk printed a Norges Bank holding as a primary-source fact in August, had to correct it, and is not repeating the error: the sentence is that Norway's manager recommended cutting the Treasury weight, not that Norway sold Treasuries.

The stake connects straight back to Tuesday's auction. The world's largest sovereign fund is asking to hold structurally less government paper at the same moment the twenty-year reopening left primary dealers with 16.74 percent of the issue against 10.93 percent at the prior sale in August, and the indirect share, which carries foreign and real-money demand, fell to 52.12 percent from 55.09. One is a proposal and the other is a settled auction. Both point at the same question of who funds the long end.

Watch the ministry's response and today's auction. The Treasury sells 19 billion dollars of ten-year inflation-protected paper, CUSIP 91282CRE3, with results in the early US afternoon, after this brief goes out. TreasuryDirect confirms every result field is still blank.

5. China bought gold again in August while the real yield was rising

China's State Administration of Foreign Exchange reported gold reserves of 76.73 million troy ounces for August, up from 76.08 million in July. That delta is 650,000 ounces, or 20.22 tonnes. The monthly series for 2026 runs 74.19, 74.22, 74.38, 74.64, 74.96, 75.44, 76.08 and 76.73 million ounces, which is a rise at every step from January to August, seven consecutive increases. These figures come from the filing itself, downloaded and parsed, not from coverage of it.

Gold went up on hike day too. December futures settled at 4,387.50, a rise of 1.26 percent, on a session when the ten-year real yield rose six basis points. The textbook says gold falls when real yields rise, because the metal pays no coupon and a higher real rate raises the cost of holding it.

The honest version is that the window decides the answer. Measured across the last thirty days the relationship is behaving exactly as the textbook requires, with gold down roughly 2 percent while the real yield climbed about 21 basis points. Measured on Wednesday alone it inverted. Both are true, and the desk that only prints the one supporting its story is the one that gets caught. A steady official buyer adding 20 tonnes a month is a plausible reason a one-day inversion can happen without the thirty-day relationship breaking.

Watch whether September's SAFE print continues the run, and watch the real yield. If gold keeps rising through a real yield above 2.70 percent for a week rather than a day, the central-bank bid is overwhelming the discount-rate mechanism and the thirty-day relationship is the one that breaks.

6. Warsh hiked, Trump objected, and no editorial board called it a mistake

Kevin Warsh's first decision as chair was a unanimous increase, and his press conference gave the week its quotable line: inflation is "too high, for too long." The Wall Street Journal's headline carried the same phrase and Valor Economico's Portuguese rendering of the presser has him saying inflation is not converging to target fast enough. Two independent routes, one message.

The reaction split along a line worth naming. Every tier-one outlet sampled treated the hike as defensible. The Economist argued for it outright. The Financial Times and the New York Times both framed it as the Fed asserting independence from a president who wanted cuts. The Wall Street Journal stayed mechanical. Bloomberg was the only one to push back, and even then its scepticism was about durability rather than correctness, asking whether Warsh will be made to walk the talk.

So the dissent in this news cycle is political, not analytical. Trump is the one calling for cuts, at one point saying rates should be one percent or less. When the entire financial press agrees a decision was right and the objection comes from the executive branch, the risk is not that the hike was wrong. The risk is that agreement this broad has stopped anyone checking the arithmetic, which is the condition under which consensus trades tend to get repriced.

Watch the press conference transcript against the projections. If Warsh's hawkishness is not matched by the dots at the December meeting, Bloomberg's doubt gets its evidence and the credibility that Reuters says was bought this week gets tested.

7. The crypto bill's own contract stopped existing

Every Kalshi market on the CLARITY Act cloture vote is finalised, settled on 15 September. The threshold ladder is worth reading because it does something the public record cannot: the "above 48 Yea votes" contract resolved yes and the "above 50 Yea votes" contract resolved no, which brackets the vote at 49 or 50 without touching a Senate source. The 0.98 and 0.02 on the board are last traded prices rather than settlement values. Both senate.gov and congress.gov return 403 to this network on every path tried, so the 49 to 50 tally is otherwise press-sourced only. A prediction market just served as an independent witness to a roll call.

The ticker for the bill becoming law now returns zero markets. It does not exist. The live successor asks whether any crypto market-structure bill becomes law, and it prices 8 percent before January 2027 on 2.25 million dollars of volume, 3 percent before November and 1 percent before October. This desk is naming the substitution rather than quietly comparing across it: yesterday's issue quoted a Polymarket contract at 5.15 percent, and Polymarket has no working route from this vantage today, so the 8 percent above is a different venue asking a broader question and the two numbers are not a move.

The stake is that the reconsideration motion Senator Tillis preserved by voting no still has no date. Two windows have been floated, before the 30 September funding deadline or a lame-duck session, both from a single outlet and neither confirmed. The market's answer is that a floor window is not the same as a law.

Watch the calendar collision. Floor time before 1 October competes directly with government funding, and the successor contract prices that at 1 percent.

8. The funds sold half a billion into a rising tape

US spot bitcoin ETFs lost $295.98 million on Wednesday and ether funds lost $224.11 million, for a combined $520.09 million, from SoSoValue's per-fund data. Tuesday was worse at $591.80 million combined, which is bitcoin's $450.33 million plus ether's $141.47 million. That Tuesday figure has circulated all week attributed to third-party coverage; it is now confirmed to the cent from the flow data itself.

The per-fund detail is broad by headcount and concentrated by weight. BlackRock's IBIT lost $144.11 million, Ark's ARKB $84.40 million, Fidelity's FBTC $52.72 million and Grayscale's GBTC $18.22 million, with only Morgan Stanley's MSBT taking money in at $3.47 million. Four of five funds sold, but IBIT alone is 48 percent of the dollars. Measured against each fund's own assets the ranking inverts: ARKB shed about 3.6 percent of itself, against roughly 0.24 percent for IBIT, so the outlier is a small fund losing a chunk of its holder base rather than a market-wide exit. On the ether side BlackRock's ETHA lost $110.03 million and Fidelity's FETH $55.58 million.

Set that against a spot market where all fourteen majors rose and the tension is the story. The listed vehicle sold while the underlying bid. Before reaching for a buyer somewhere else, note that this needs no buyer story at all: creation and redemption baskets settle with a lag, so Wednesday's print can reflect orders placed before the decision, and a basis-trade unwind produces exactly this shape without anyone changing their mind about owning bitcoin. Note also the scale. Half a billion dollars is under two percent of bitcoin's own $29.76 billion of daily turnover, which is well inside what plumbing can generate. The desk cannot identify a marginal buyer from this data, because Glassnode, CryptoQuant, Arkham and Blockchair all refuse this network and no whale movement with a confirmable date inside 48 hours was found. It also has not established there is a separate one to find.

Watch whether Thursday's flows turn. Two consecutive sessions of outflows above half a billion dollars against a rising spot price is a divergence that resolves one way or the other inside a week.

9. Fear and Greed hit 50 on a day everything went up

The index printed 50, Neutral, on Thursday. It was 51 on Wednesday and 69 on Tuesday, so the drop that matters was Tuesday to Wednesday, at 18 points. Against the full thirty-day series the current reading is the lowest since 19 August, when it was 46 and classified Fear. The month in between was almost entirely Greed, touching 74 twice.

What makes it odd is the price action beside it. Sentiment sits at its lowest in a month on a session where bitcoin, ether, solana and eleven others all rose, the VIX fell 8.41 percent to 16.22, and five European indices opened higher. Sentiment indices lag the tape by construction, because they blend volatility, momentum and survey inputs that are still digesting Tuesday.

The positioning data agrees with the caution more than the price does. Bitcoin's all-account ratio is 60:40 (1.49), well off Tuesday's crowding, and BONK is the only one of the fourteen with more shorts than longs at 44:56 (0.80). Volume fell across most of the table. This looks like a market that went up without conviction.

Watch for the index to catch up rather than the price to roll over. If Fear and Greed climbs back through 60 while volumes stay down, the rally is being chased rather than funded.

10. Japan's curve flattened into a decision it has to make tomorrow

The Bank of Japan concludes a two-day meeting on Friday and has not decided anything yet. Its guideline is unchanged at around 1.00 percent per the bank's own page. Press expects a quarter-point increase to 1.25 percent, which would be the highest since 1995, and that expectation is a preview rather than an outcome.

The curve moved ahead of it in the shape you would expect. The two-year rose to 1.868 percent while the thirty-year fell about four basis points to 4.074 and the ten-year eased to 2.991, narrowing the 2s10s to roughly 112 basis points from 117 in yesterday's issue. Japan's slope is still by far the steepest of the majors: Germany is at 28.2 basis points after flattening from 36, the UK at 51.8, and the US par curve at 27.

The yen is the pressure point. Dollar-yen sits at 155.80, Nikkei Asia led its paper on the yen weakening to the 156 level as the Fed did what was expected, and a Reuters headline on Thursday has Japanese officials promising to keep moves orderly, which is verbal intervention and not intervention. The Nikkei still closed up 0.33 percent at 64,136.25.

Watch the gap between the decision and the currency. If the BOJ hikes and the yen weakens anyway, the market is telling Tokyo that a quarter point does not close a spread this wide, and the intervention conversation stops being verbal.

Crypto

Price action

Prices, volumes and both change columns from CoinGecko, pulled in a single pass. Bull:bear is the share of Binance futures accounts long versus short over the latest hour. It counts accounts, not position size. Bonk is the only name on the table with more short accounts than long.

All fourteen rose on the day and eleven are still lower on the week, so this is a bounce inside a losing week rather than a turn. The dispersion is informative. The largest coin gained least, at 1.24 percent, and the smallest gained most, at 7.00 percent, which is the ordering you get when risk appetite returns to the tail rather than when capital rotates into quality. Volume went the other way on ten of the fourteen, with XRP's traded value down 38.57 percent against the same window a day earlier and bitcoin's down 23.58.

Market structure

Total crypto market value is $2.626 trillion on $92.01 billion of 24-hour volume, with bitcoin dominance at 58.32 percent and ether at 11.33, per CoinGecko's global endpoint. The aggregate is down 1.356 percent over 24 hours even though every coin in the table above is up, which is the arithmetic of a broad market where the long tail outside the majors fell.

The trending list is where the day's rotation shows. CoinGecko's trending categories are led by cat-themed tokens at plus 16.68 percent, quantum-resistant names at plus 15.33, zero-knowledge at plus 14.69 and privacy coins at plus 10.19. Zcash sits at $1,369.82, up 16.98 percent, and Monero is one of only two names on the trending list that fell. A privacy and post-quantum bid on the day a central bank tightens is not an obvious pairing, and this desk does not have the flow data to explain it.

Fund flows

US spot bitcoin funds lost $295.98 million on Wednesday against $450.33 million on Tuesday and a $160.04 million inflow on Monday. Ether funds lost $224.11 million on Wednesday and $141.47 million on Tuesday. The combined Tuesday figure is $591.80 million and the combined Wednesday figure is $520.09 million, both from SoSoValue's per-fund series rather than from coverage of it.

Only funds with a non-zero Wednesday flow are listed, and both columns now sum to their stated totals: the five bitcoin funds to minus $295.98 million and the six ether funds to minus $224.11 million. SoSoValue's two endpoints disagree on cumulative bitcoin inflow by $21.42 million, $54.568 billion against $54.590 billion, which is a mark-to-market against flow-only difference. Both are reported rather than reconciled. The ether endpoints agree to the cent.

Derivatives

Funding is positive across the three majors on Binance, which means longs pay shorts in every case. Bitcoin sits at 3.514 percent annualised, ether at 3.694 and solana at 0.967. Ether is the one that changed: it was negative at 4.62 percent annualised in yesterday's issue, so the side that pays has switched. Across the ten days of eight-hour prints available from the endpoint, bitcoin was positive on all thirty and ether was negative on seven of thirty, with a low of minus 11.837 percent annualised.

Open interest needs one label to be readable. The live Binance figure is 108,337.858 bitcoin and 2,312,640.549 ether at the time of the pull. The daily series marked at the UTC turn reads 107,918.184 bitcoin, about eight hours older. This brief uses the live figure as current open interest and the daily series only for trend, because mixing the two read as a contradiction to five reviewers in a prior issue.

On Hyperliquid, ether funding annualises to 10.950 percent, which lands almost exactly on the structural floor this desk has noted in past runs, while bitcoin is at 5.168 and solana 9.264. The order books disagree with each other: bitcoin and ether are both ask-heavy, at depth ratios of 0.786 and 0.853, while solana is bid-heavy at 2.109. Bitcoin's top ask alone carries 21.03 coins across 86 orders against roughly a quarter of a coin across the top five bids, which is a thin bid against a thick offer right at the touch.

There is no cross-exchange liquidation total in this issue. Binance's public force-order endpoints now return a 404 on one path and require a key on the other, and Coinglass and Coinalyze both need keys. Rather than present one venue's partial as a market total, this desk reports the absence.

Options

Friday's Deribit expiry carries 22,499.9 bitcoin contracts of open interest against 140,426 in ether. Max pain computes to 77,500 for bitcoin and 2,500 for ether, each derived at this desk from the full book by minimising open-interest-weighted intrinsic value across every listed strike rather than taken from a vendor. Both sit above spot, bitcoin by about 1.4 percent and ether by about 2.4. The put-call ratio is 0.802 on bitcoin and 0.505 on ether, and at-the-money implied volatility is 30.66 percent and 42.66 percent respectively.

The bigger board is next Friday. The 25 September expiry carries roughly $14.32 billion of bitcoin notional with max pain at 72,000, about 5.8 percent below spot, and $1.86 billion of ether notional with max pain at 2,200. Max pain is mechanical open-interest arithmetic and not a forecast, but a gap that size between the near expiry and the following one is worth holding in view.

Traditional Markets

The American session

Wednesday closed lower across the board and the dispersion says more than the direction. The Dow fell 1.21 percent to 51,461.90, a drop of 631.21 points, while the Nasdaq Composite was effectively unchanged at 25,978.43, down 0.01 percent. The S&P 500 fell 0.45 percent to 7,551.81 and the Russell 2000 fell 0.40 to 2,858.81. A session where the industrial average loses more than a percent and the technology index loses nothing describes rotation out of the rate-sensitive and the cyclical rather than broad de-risking.

The volatility unwind came afterwards. The VIX closed Wednesday at 17.71 and traded at 16.22 on Thursday morning, down 8.41 percent, per the CNBC benchmark series. The CBOE feed's own VIX record passes every staleness test with real session ranges and a genuine price change, which is worth stating because the gold and crude volatility indices from the same source did not, and are excluded below.

What the curve did with it

Treasury official par yield curve, percent. Slopes computed here from the printed levels. This is not the same series as the CNBC benchmark quotes used in the ticker and in Thursday-morning references, and the two are never mixed in one row.

Treasury real yield curve, percent. The breakeven column is the nominal ten-year minus the real ten-year, computed here from the two official series rather than taken from a vendor.

Commodities and currencies

Settlements and live prints are separate columns on purpose. Gold is the December futures contract throughout this brief; a spot quote for the same metal reads several dollars lower and is a different instrument. Brent-WTI is 3.40 on the settles and 3.28 on the live prints.

Crude fell hard on the decision day for a reason that has nothing to do with the Fed. The United States said a damaged Saudi pipeline would restart within days, and Brent gave back 2.69 percent after a 2.90 percent gain on Tuesday. Wednesday's low printed 104.00. That number matters to this desk because it is the exact upper bound of an open bias zone, discussed in the Bottom Line.

The dollar did almost nothing. The dollar index sits at 99.99, up 0.01 percent, with the euro at 1.1469, sterling at 1.3382 and dollar-yen at 155.84, down 0.27 percent. A quarter-point increase with a hawkish projection set that moves the currency by a hundredth of a percent is the clearest evidence available that the decision was fully in the price.

What comes next on rates

Futures-implied probabilities. Buckets are counted from the new 3.75 to 4.00 range, so 4.00 to 4.25 is one further increase and 4.25 to 4.50 is two. The source page's own header still labels the current range as 4.00 to 4.25, which contradicts the Fed statement; the bucket boundaries are consistent with 3.75 to 4.00 and the header is treated as a stale cache.

The week-over-week column is the story. Seven days ago the December two-increase bucket was 8.9 percent and now it is 40.3. The hold bucket went from 40.8 to 12.1 over the same stretch. Whatever the Fed said on Wednesday, the repricing toward a genuine cycle happened across the week, not in the hour.

The projections themselves moved further than the statement did. The 2026 median dot is 4.1 percent against 3.8 in June, the 2027 median is 4.1 against 3.6, and the longer-run value is 3.2. Unemployment was revised down to 4.1 percent from 4.3, headline PCE up to 3.7 from 3.6 and core PCE to 3.4 from 3.3. A committee that marks down unemployment and marks up inflation in the same round is describing an economy with less slack than it thought, which is the argument for the increase it just made.

Data and the calendar

August retail sales landed on the morning of the decision and beat: $773.9 billion, up 1.2 percent on the month, with July revised to minus 0.5 percent from minus 0.6, per the Census advance report. Sales are up 6.0 percent on the year. A committee reading a 1.2 percent retail print hours before voting had its evidence of resilient demand handed to it, which also cuts against the weaker-demand reading of the falling breakeven discussed above. Two effects sit underneath that print and pull in opposite directions. Somewhere between $100 billion and $166 billion of tariff refunds is flowing back to payers after the court rulings, which is a demand stimulus arriving in the same week the Fed is trying to cool demand. Removing tariffs is also disinflationary at the price level, which is a live non-Fed explanation for why inflation compensation fell that owes nothing to credibility.

Today brings the Philadelphia Fed survey, jobless claims, housing starts and building permits, alongside the Bank of England decision and the ten-year inflation-protected reopening. Next week's dated calendar could not be retrieved from either of the usual routes and is left out rather than filled with a vendor's projected values, since that vendor has been observed printing figures in its actual column for events that have not happened.

International Markets

Asia Pacific

Index levels are closes only where the print is timestamped after the local close. Tokyo and Seoul close at half past three in the afternoon local time, Singapore at five, and the Indian and Singaporean figures here were taken while both markets were still open and are labelled accordingly. The Monetary Authority of Singapore's site was unreachable on both attempts, so no policy stance is printed for it.

Europe and the Americas

European levels are intraday prints taken during the morning session, not closes. The Brazilian, Canadian and Mexican equity feeds were serving prior-session values at the time of the pull; the Mexican print is two sessions old and its change column is left empty rather than presented as current.

Europe opened green with yields higher, which is an unusual pairing and the most interesting thing on this page. All five indices pulled were up and Bunds, Gilts and OATs were all fractionally higher in yield at the same time. Rising yields with rising equities reads as repositioning after an event passed without damage rather than as risk appetite.

The German curve did what the American one did, and the size of the move depends on which series you read. The Bundesbank's official par curve put the 2s10s at 36 basis points at Wednesday's close. A traded-benchmark series has it at 28.2 on Thursday morning, with the two-year up about 2.9 basis points against the ten-year's 0.93 and the thirty-year flat. Those two figures are different vendors on different days and the Thursday legs alone account for roughly two basis points of flattening, not the eight the pair implies, so most of any move belongs to Wednesday itself. The direction is not in doubt in either series. Two curves flattening from the front, in two currencies, is the market applying one decision to two central banks. The ECB has already moved, having raised its deposit rate to 2.50 percent on 10 September.

France is the exception that is not about the Fed at all. The OAT-Bund spread computes to about 97 basis points from today's prints, against roughly 85 in late August and about 55 at the start of the year. A budget that collapsed a government and a 2027 budget that must pass by year-end are doing that, and a quarter point in Washington is not.

Geopolitics

Hormuz, day two hundred

The Strait of Hormuz blockade reached its two hundredth day on Thursday. Traffic through it is running at roughly two to eight vessels against a pre-crisis baseline near 85 a day, war-risk insurance is quoted around forty times its pre-crisis level, and 431 vessels are holding position off-berth in the region. These figures come from a shipping tracker rather than a press outlet and carry that caveat: they are an aggregator's numbers, not a wire's.

The most recent attack was on 13 September, when a vessel transiting near Limah in Oman was struck by a projectile and caught fire without casualties, per a UK Maritime Trade Operations warning. That was the first reported strike since a Saudi blockade declaration in July. Separately, Saudi infrastructure was struck around the same date, and the American statement on 16 September that the damaged pipeline would restart within days is what turned crude down.

The gap between the official line and the traffic count is the thing to hold. The United States has repeatedly said the waterway is open or being cleared. Two vessels a day against a baseline of 85 is what the trackers show. Both statements can survive together only if "open" means legally transitable rather than commercially used, and the insurance quote settles which of those the market believes.

Two things this desk cannot tell you, and they matter because a trade card rests on them. Oil markets price barrels, not hulls, and nothing in the tracker data gives the size mix of the vessels still moving against the size mix of the baseline, so a count falling from about 85 to between two and eight does not convert into a known percentage of lost supply. And the overland pipeline struck this month is the standard route Gulf producers use precisely to bypass the strait, which cuts against the blockade story rather than adding to it: its restart restores a route around the chokepoint rather than reopening the chokepoint. A throughput collapse of the size the vessel count implies, sustained for two hundred days, sits awkwardly beside Brent at $104.93. Either the count overstates the barrel loss or the market is underpricing the disruption, and this page cannot yet say which.

No diplomatic movement has been reported within the last week. The most recent items on reopening talks are dated late August, which makes them roughly three weeks stale, and no release from strategic reserves was found.

Russia, Ukraine and the truce that is not one

Ukraine struck a refinery at Yaroslavl on 15 September and has also hit the Syzran refinery in the Volga region, per Bloomberg. Russia struck Ukrainian power sites on 13 September, wounding 18, and struck again on 17 September, per a same-day Cyprus Mail report. An energy truce was announced by Trump and has been confirmed by neither government, and strikes on energy infrastructure have continued through the window in both directions. The claim and the tape disagree.

Sanctions mechanics are where this desk has to report a gap rather than a number. The last confirmed change to the Russian crude price cap is the European mechanism that set it at $44.10 a barrel effective in January, and an American general licence issued in March effectively suspended the cap for covered cargoes for US persons. No September update on waiver status could be found through any route tried.

Tariffs after the court

The Supreme Court has invalidated most of the administration's tariffs, per ABC News on 16 September, the culmination of a sequence of rulings running back through the spring on the emergency-powers and Section 122 authorities. That is a single outlet for a major ruling, and unlike the Senate tally elsewhere in this brief it has no second witness, so it is reported at the confidence one source supports. Refunds have reached somewhere between $100 billion and $166 billion depending on the counter. Twenty-five states have sued over the tariffs imposed since.

The response has been to find other authorities. A 1930-era statute was invoked for 50 percent tariffs on Canada, per a single outlet in late August. On semiconductors, the Section 301 rate on the relevant list stands at 25 percent stacking on a pre-existing 50 percent, and the chip-specific announcement has slipped to 2027. Handelsblatt's lead this morning paired the German market's rise with a tariff threat, so Europe is pricing this as live rather than settled.

On-Chain and Whale Intelligence

The stablecoin float is flat to slightly lower on the week and the composition is shifting underneath. Tether stands at $183.23 billion, down $41.58 million over seven days, with its share of the tracked market at 58.63 percent against 59.34 a month ago. USD Coin is at $73.75 billion, down $765.69 million on the week but up $1.76 billion on the month, with share at 23.60 percent. The total tracked float is $312.51 billion, up 1.37 percent over thirty days. Every delta here was computed from the raw series rather than taken from a precomputed field.

There is no confirmed whale movement inside the last 48 hours, and that is a finding rather than an omission. Several candidate items surfaced with headline dollar figures attached, and every one of them either dated to early September or carried no extractable event date at all. This desk has twice been burned by republished articles wearing fresh timestamps, most recently having to correct a correction, so an undated flow figure does not get printed.

Mining is the one place where behaviour changed in-window. MARA bought 1,292 bitcoin on 16 September for about $98.64 million, having been a net seller of 23,093 coins earlier in the year. Difficulty rose 1.31 percent to 127.45 trillion at the early-September retarget and hashprice rose from $32.42 to $39.63 per petahash per day around the same adjustment. Sustained hashrate was running near 914 to 915 exahash in late August, and no fresher sustained figure was available; a single-day print above 1,000 exahash is a spike rather than a level.

Strategy is unchanged and that is confirmed from the filing rather than from coverage. Its 8-K covering 8 to 13 September states plainly that it neither bought nor sold bitcoin nor sold shares under its at-the-market programme that week, leaving it at 845,050 coins at an average cost of $75,412. Spot sits 1.33 percent above that average. The company did repurchase $139.3 million of one preferred line in the same week.

Prediction Markets

One venue is missing from this issue and the reason is not that it went quiet. Every route to it from this desk's connection fails: the public interface returns nothing on either resolved address, and the tooling that wraps it surfaced a certificate mismatch, which is the signature of a network interception rather than an outage. That is a block at this vantage, not a provider failure, and the distinction matters because yesterday's issue quoted prices from it. Everything below is from the other venue.

Calibration is p raised to 1.31, over itself plus one minus p raised to 1.31, which sharpens probabilities away from the middle. Because it is a pure function of the raw price, two different questions trading at the same price produce identical calibrated values and identical edges. That is arithmetic, not a duplicated pull. Edge is in percentage points.

The October contracts are the ones to read carefully, because the two available venues disagree about direction and not merely about degree. Futures-implied pricing puts one more increase in October at 55.1 percent, so the increase is favoured. This venue prices the hold at 57.0 percent and the increase at 46.0, so the hold is favoured. Those two markets are structured differently and do not sum the same way, which is part of the gap, but a nine-point spread with opposite favourites is the kind of disagreement that used to be the basis of a card on this page. It is not today, for reasons given in the Bottom Line.

Two absences are worth naming. There is no live contract on the bill this desk has tracked for a week; the ticker that asked whether it becomes law returns no markets at all, and what remains is a broader successor question. And there is no contract anywhere on this venue for a Hormuz blockade or closure, despite the blockade being in its two hundredth day and setting the insurance premium inside the crude price. The only live strait-related contracts ask whether the waterway will be renamed. A market with two hundred days of a supply shock and no instrument on the mechanism is telling you where the liquidity is, and it is not where the risk is.

Sentiment and Positioning

The table splits cleanly in two and the split is the point. Everything measuring sentiment is soft: the index at its lowest in a month, dominance flat, volumes down across most of the price table. Everything measuring positioning has become less extreme rather than more: bitcoin's account ratio fell from 1.78 to 1.49, ether's from 3.30 to 2.9952, and the funding that was charging the short side is now paying it.

That combination rules out confirmed new money without establishing what replaced it. Confirmed new money shows up as rising volume and rising open interest together, and volume fell on ten of fourteen names. But open interest rose, and rising open interest into a rising price is the signature of new longs opening rather than shorts closing, because a short covering its position removes open interest rather than adding it. So the honest description is a thin, concentrated bid: fewer participants taking larger positions, with a sentiment gauge that has not moved off neutral. This desk cannot separate that from a forced unwind earlier in the week, because the cross-exchange liquidation total that would settle it does not exist on any free route, as the derivatives section says.

Cross-asset linkages

Credit spreads carry a reporting lag and the latest available print is 15 September, so none of the credit column yet reflects the decision. Sovereign default swap levels are as dated in the source. The gold and crude volatility indices are excluded from this issue: both failed a staleness test, showing zero volume, zero session range and a midnight timestamp, and computing a same-day ratio from a carried print against a live one would be a fabricated comparison. India's row was dropped for carrying an impossible future date.

France paying more than Italy to insure its debt is the quiet anomaly on this table, and it is consistent with the OAT-Bund spread at 97 basis points. The credit stack otherwise says the weakest tier is deteriorating alone: the bottom rung made another 90-day high while investment grade did not move at all and the broad high-yield index sits mid-range. Either that is the leading edge of a refinancing problem or it is a composition quirk in a small index, and two more sessions will not settle it.

The Bottom Line

1. The Fed raised to 3.75 to 4.00 percent, twelve to none, and the ten-year breakeven fell five basis points to 2.33 while the ten-year real yield rose six to 2.68. The front end carried it: two-year up seven, ten-year up one, par 2s10s flattened from 33 to 27. The flattering reading is that the market accepted the tightening. Two rival readings fit the same numbers, weaker demand and simple scarcity of duration buyers into a heavy supply week, and nothing on this page separates them yet. Today's auction is the first test.

2. The projections moved further than the statement. Sixteen of eighteen participants mark at least one more increase and four mark two, counted from the projection figure itself. The December two-increase bucket went from 8.9 percent a week ago to 40.3 today.

3. One decision forced opposite answers elsewhere. Hong Kong raised to 4.25 percent mechanically through the peg, Brazil cut to 13.75 on the same afternoon, Britain decides today with three of nine already dissenting for a hike, and Japan decides tomorrow with its 2s10s at 114 basis points.

4. Risk went up anyway. All fourteen majors rose, the VIX fell 8.41 percent, and five European indices opened higher with yields higher too. But volume fell on ten of fourteen coins and the listed funds sold $520.09 million on the day, so the bid was not coming through the American wrapper.

5. The risk this page is least hedged against is Japanese. Today's broad, low-volume rally across crypto and European equities is the shape a yen-funded carry trade makes, dollar-yen sits at 155.80 with officials already talking about orderly moves, and the Bank of Japan decides tomorrow into the steepest 2s10s of the majors. A hawkish surprise there unwinds the funding leg of exactly this move, and no card on this page is positioned for it.

6. The things this page cannot see are named rather than smoothed over. One prediction venue is blocked at this connection, there is no free cross-exchange liquidation total, official foreign-holdings data is stale by nine months, and two volatility indices failed their own staleness test.

What this desk got wrong

Yesterday's issue compared Tuesday's twenty-year reopening against what it called an $18.1 billion original issue in August. The official auction record for that security shows the prior sale was on 19 August and it was a $16 billion reopening, stopping at 5.204 percent with a cover of 2.53. The comparison held up, the stop yield did rise 21.6 basis points month over month, and the label on it was wrong.

The gold watch carried from 10 September needs the same treatment. It was built on gold rising while real yields rose, and on Wednesday it did exactly that, up 1.26 percent against a six-basis-point rise in the ten-year real. Measured across thirty days, though, the relationship is behaving conventionally: gold is down about 2 percent while the real yield climbed roughly 21 basis points. The anomaly is real on the day and absent on the month, and this page has been printing the day without the month.

Last issue's views, marked

Trade views

On concentration. Five cards carry nine conviction points between them. Six of those nine sit on the rate path, through the breakeven card, the supply card and the credit card, which all resolve against the same set of yields. Only the crude card is genuinely independent, and the crypto card is partly rate-linked through risk appetite. No card on this page carries a filled entry, so capital at risk today is zero, though zero risk is not zero cost. Over windows of two to three weeks the 4.707 percent the two-year pays is worth roughly 18 to 20 basis points of carry, not the headline rate, and the larger cost is missing the one entry actually in reach, Brent sitting on its zone boundary, while waiting for a fill a single headline could erase. The outcome this page spends least time on is the one that would hurt most: an inflation print that lifts the breakeven back through 2.40 would break the first card and the credit card together, and would make the crude card right for a reason this desk did not write down.

Catalyst calendar