Seven weeks ago twelve people sat in a room and split nine to three. The three who lost wanted rates a quarter point higher. The nine who won voted to leave them alone. On Wednesday afternoon the same twelve voted twelve to zero to raise them.

The statement records the vote in its opening line, before anything else. The target range went to 3.75 to 4.00 percent, the first increase since 27 July 2023, which is 1,147 days. The July statement names the three who were outvoted then: Beth Hammack, Neel Kashkari and Lorie Logan, each of whom "preferred to raise the target range for the federal funds rate by 1/4 percentage point at this meeting." Seven weeks later that is precisely what the committee did, and nobody voted against it. The dissenters did not get outlasted. They won, and the other nine came to them.

One clause did not survive the seven weeks either. In July, inflation was elevated "in part reflecting supply shocks that have driven price increases in certain sectors, including energy." In September that clause is gone. Three words are left of it, "Inflation remains elevated," and then a sentence of intent: "Today's policy action will support a timelier return to the Committee's 2 percent goal." Brent spot ran 25.20% in six sessions to $130.80 on 15 September on the EIA's own series, the day before the meeting. The committee took the energy sentence out in the week energy would most have justified leaving it in.

Then the price went the other way.

Bitcoin's low on the UTC day of the decision was $75,447. It was $81,618 at a quarter to nine on Sunday evening, up 8.2% from that low, which printed at ten on Tuesday evening our time, sixteen hours before the announcement. Friday did most of the work on its own: plus $4,753, plus 6.22% in a single session. An asset that pays no yield is supposed to hate a rate rise. It got one from two central banks inside three days and went up.

We cannot tell you why, and we would be careful with anyone who can. Three explanations were circulating by Friday night. The first is the SEC order that legalised tokenised stock trading on Thursday, which is real, dated, and in the RADAR below. The second is oil reversing, which we cannot check: the EIA's Brent series stops at 15 September and does not update until Wednesday, and the secondary reports of Friday's crude price contradict the EIA's own last print by more than twenty dollars. The third is a wave of short liquidations, which exists only in secondary coverage we did not confirm. A verified move with an unverified cause is still a verified move. It is not yet a reason.

Three metrics.

Metric 1, Fear and Greed: 70, Greed. The stamp is taken at midnight UTC, which is eight in the evening here, so this is Sunday night's reading. On the evening of the decision itself it was 50, Neutral. Twenty points in four days, and every one of them came after the hike rather than before it.

Metric 2, ETF flows: plus $433.0 million on Friday, and not one fund had an outflow. That is the part worth repeating. Thursday was plus $159.5 million with IBIT the only fund taking money in at all. On Friday five funds took money in, Fidelity ahead of BlackRock by roughly three to one, and the other seven took nothing. Nobody sold. Wednesday, the day of the decision, was minus $295.9 million. The three days net to plus $296.6 million, and September stands at plus $313.63 million on SoSoValue's published monthly figure. Our 7 October test on the September total is Branch A, net positive, and it was negative as recently as Wednesday evening.

Metric 3, BTC dominance: 58.1%. CoinGecko's global measure late Sunday against a total crypto market of $2.81 trillion, which puts bitcoin itself near $1.64 trillion. The corporate bid is the strange one this week. MSTR closed Friday at $153.92, up 16.39% on the day, on 54.4 million shares, about two and a half times its average volume this month, and it ran about 2.6 times bitcoin's move. The company's most recent disclosure covers 8 to 13 September. It shows no bitcoin bought for a second consecutive week, 845,050 coins held at an average cost of $75,412, and $139.3 million of cash spent buying back its own preferred instead. Friday was a 16% move in a company whose entire argument is an asset it had not added to in a fortnight.

The book is flat for the fourteenth issue running. The STACK grades two dated tests, one contract our own rule told us to act on and we did not, and one reader who called this the other way in August.

📣 THIS SPOT IS OPEN

Baseline Crypto reaches nearly 7,000 Bitcoin-focused readers every Monday and Wednesday morning, opening about half of what we send. Direct sponsor placements are open, premium position, top of the issue, one per send.

Reply to this email for rates.

A unanimous vote is the cheapest thing that committee produces. The argument did not end on Wednesday, it moved to the page where nobody's name is attached to anything.

Start by giving the position everything it deserves, because most of it is correct and the correct part is genuinely surprising. Nine people changed their minds in seven weeks. That does not happen over nothing. Unemployment was revised down two tenths for this year, to 4.1%, while inflation was revised up, and that combination is what a rate rise is made of. The bond market took the committee at its word within hours: ten year breakevens closed at 2.38% the day before the meeting and 2.33% on the day of it, and they have not moved since. Expected inflation fell through a rate increase. That is what credibility looks like when you can see it in one number, and a 12 to 0 vote seven weeks after a 9 to 3 is a committee that really did converge.

Now ask what a vote costs and what a dot costs.

A dissent is attributable. It carries a name in the statement, in every wire story written that afternoon, and in the permanent record of the meeting. A dissent against the chair's first rate increase says out loud that the chair does not control his own committee. The projections published at the same moment carry no names at all. They are a distribution of anonymous forecasts. If you are a participant who is not convinced, the vote line costs you something real and the dot plot costs you nothing.

So look at where the disagreement went. The full range of participants' forecasts for where the federal funds rate sits at the end of this year runs from 3.9% to 4.4%. The midpoint after Wednesday's decision is 3.875%. In plain terms, the most dovish person in that room expects no further change for the rest of 2026, and the most hawkish expects two more quarter point increases, and both of them voted for the same sentence. That is a 50 basis point spread on where rates are in fourteen weeks, published by a committee that had just recorded a unanimous vote. The median is 4.1%, which is exactly one more increase and does not say when.

The market cannot resolve it either, which is the tell. Polymarket prices another quarter point at the October meeting at 54.5% and at the December meeting at 67.5%. Twelve people were unanimous about Wednesday. Nobody is unanimous about the next fourteen weeks, including them.

Run the incentives one more step, because they explain the whole shape. In July, dissent was free: three hawks voting against a hold cost the chair nothing and cost them nothing, so they dissented and the argument was visible. In September, dissent got expensive and the argument became invisible. Nothing about that requires anyone to have changed their mind more than the data justified. It only requires that people respond to what disagreement costs them, which they always do.

What we are not claiming. We are not saying the committee was secretly split on Wednesday's decision. It demonstrably was not, and the vote is the vote. We are saying that reading the vote line for information about the path is now reading the cheapest line on the page, and that the dispersion in the dots is where the next year of real yields is actually being argued about. If you want one habit out of this issue, it is that one.

The observation that tests this is the vote line on the October statement, due 28 October. A second unanimous vote with the dots pulling together as well means the convergence was real and we have overread this. A second unanimous vote while the dots stay fifty basis points apart is what this argument predicts, so it settles nothing. A recorded dissent in October, in either direction, means the argument surfaced in the one place it costs something, and the market will price it accordingly within the hour. We are not putting branches on that because it is a count with two meanings rather than a magnitude, and our own rules say a dated test has to be a number cut into ranges. The formal test this issue sets is below, and it is about reserves.

Three numbers that define the next week.

Around minus 2%, and a hashrate that has not settled. Difficulty rose 4.1634% at block 967,680 on Saturday morning and the network went the other way almost immediately. mempool.space's daily hashrate average was 1,020.92 EH/s on 16 September and 826.06 EH/s on Sunday. Treat that as directional and not confirmed: daily hashrate is inferred from block times and it is noisy over a day or two. Sunday closed at 826.06 EH/s and Monday's partial bucket is already back above 1,000 EH/s, so the drop may not survive the week. What is not noisy is the clock. Blocks are running about 640 seconds against a 600 second target, the new epoch is roughly 12% done with about 1,780 blocks left, and the estimate for the next retarget at block 969,696 is around minus 2% as we write on Sunday evening. That estimate moves with every block that lands, so read it as a direction and not as a figure. What to do with it: nothing, and watch it. Difficulty rising until the marginal miner switches off, then falling back, is the only self correcting mechanism in this asset, and it is the cleanest available read on whether hashing is profitable at these prices.

$3,013.8 billion, and the print on Thursday that settles the argument. Bank reserves rose $119.3 billion over the fortnight in which the Treasury pulled $120.3 billion in. Both figures are real and they cannot both mean what the standard liquidity story says they mean. The 16 September reserve figure is a weekly average for the week ending that day, so it holds only the last two days of the tax date. The week ending 23 September is the first clean observation, and the Fed publishes it in Thursday's H.4.1 release.

So here is the test, and it grades in the Monday 28 September issue. The number is the Federal Reserve's reserve balances series, FRED WRESBAL, for the week ending 23 September 2026. Branch A: below $2,900 billion. Branch B: $2,900 billion up to but not including $3,000 billion. Branch C: $3,000 billion or above. One published series, one date, three ranges, no gaps and nothing outside them. Branch A is the drain arriving late and in full, and it means the tax date did come out of the banking system after all. Branch B is a partial drain, the ordinary outcome, and the one we would pick if forced. Branch C is no drain at all, and in that world the liquidity framing we have been using here for a month is wrong and we will write that sentence rather than explain it away. For scale, this series has run between $2,894.5 billion and $3,013.8 billion in the last three weeks, so all three branches sit inside what it has actually done this month.

2.70%, which is exactly the number that makes us wrong. On 2 September we committed in print that if high yield spreads were at or inside 2.70% when the Treasury's cash balance peaks in late October, our framing of this whole liquidity cycle was wrong. The ICE BofA index was 2.65% on 11 September, 2.76% on the 15th going into the meeting, then 2.70% on the day of the decision and 2.70% again on the 17th. Eleven basis points of widening into the Fed, six of them back out within a day. Our 7 October test has this in Branch B, sitting on the floor of Branch B, one basis point from Branch C. The commitment is not one basis point away. It is at the number, about five weeks before the date it attaches to. Credit is looking at a first rate rise in three years, an oil price that ran 25% in six sessions, and a government pulling $120 billion out in a day, and it is charging no more for risk than it did in the middle of August. Either credit knows something the rest of this issue does not, or it is slow. Both have been true before.

Forget analyst predictions. Polymarket is a real-money prediction market, where traders put actual dollars on outcomes. Scorecard first, then the board. Not financial advice. Our read. Disclosure: we hold personal positions in Polymarket itself and may earn a commission from Polymarket referrals. These markets are thin and the odds below move, so read direction over ticks.

Scorecard. Two dated tests grade in this issue, one contract we should have owned resolved without us, and a reader who called it the other way gets graded on the same terms we grade ourselves.

1. The dissent count. The rule, published Wednesday: the number is the count of dissenting votes recorded in the FOMC statement of 16 September. Branch A is zero. Branch B is one or two. Branch C is three or more. The number is zero; the statement records the vote as twelve to zero in its first line. The verdict is Branch A. What changes is nothing in our favour. We published A and B as the world the bond market had already priced, and C as the only branch in which our own argument would have mattered. We did not call the count. We wrote that we did not know it and that neither did anyone outside the building. The branch that hit is the consensus branch. A test that resolves onto the priced outcome while the analysis attached to it pointed somewhere else is not a win, and we are not going to file it as one.

2. The difficulty retarget. The rule, published 7 September: the number is the percentage change in network difficulty at the block 967,680 retarget on mempool.space. Branch A is plus 4.00% or more. Branch B is zero to under plus 4.00%. Branch C is below zero. The number is plus 4.1634%, at block 967,680, mined on 19 September at 07:09 UTC. The verdict is Branch A, by 16 basis points. What changes is how much weight we give an estimate. On Wednesday we printed mempool.space at plus 5.05% to plus 5.15%, headlined it at about plus 5.1%, and noted in the same block that the estimate had been on both sides of our line inside four days. The estimate overshot the outcome by 0.94 points at that headline figure and the test cleared by 0.16. Right branch, and the margin was about a sixth of what the estimate implied.

3. The September hike contract. The rule, published 31 August, in the words we used then: a fall below 40% with Warsh's position unchanged "would be a mispricing claim rather than a threshold, and we would act on that." The number is 37.5%, held from 09:37 to 09:47 on 3 September, with twenty nine minutes in all trading under 40%, and the contract resolved yes on 16 September and settled at a dollar. The verdict is that the rule fired, we did not act on it, and it paid in full. Two things change. First, a disagreement inside our own checks is settled, and not the way Wednesday's issue left it. One check read the minute tape and got 37.5%. Another read a coarser series and got 39.5%. Polymarket still serves minute data if you ask for it with explicit start and end timestamps, and it shows 37.5% held for ten minutes with twenty nine minutes under 40% in all. The rule we adopted on 14 September says a published trigger is checked against the full trade history and never against a daily or hourly series, and this is exactly why. The number is 37.5%. Second, the grade itself. We are writing our scorecard rule down here for the first time: a call counts as a win or a loss only when a published issue printed its exit, or the market resolved while we held it and an issue said so. We held nothing and printed no exit, so this is neither. It goes into the record as a missed trade our own rule called, which is worse than a pass, because a pass is a decision and this was not.

4. The reader call. On 3 August a reader, Darryl, answered our question by saying there would be no rate hikes from the Fed this year, and we published his reasoning at length. His channel was oil: the Strait reopens, crude falls, inflation falls far enough that the Fed does not have to move before the November primaries. The Fed raised rates on 16 September. The call was wrong. The premise went the other way too, and hard: Brent ran 25.20% in six sessions to $130.80 the day before the meeting. What he got right is the part worth saying out loud. He named oil as the variable that would decide it, which is what the committee spent the summer arguing about, and for a month the market agreed with him. On 3 September this contract traded as low as 37.5%, which is real money putting better than three to two odds on his side of it. He put a dated call in writing with his name on it. On 14 September we ruled that the oil leg had already resolved against him, and it had.

5. The book. Flat for the fourteenth consecutive issue, recomputed from the archive this morning rather than carried forward from the last one. Fourteen is too many and we are not going to dress it up as discipline. The only standing entry condition left on the board is high yield spreads through 3.00% into the late October cash peak, and spreads are at 2.70%, thirty basis points the wrong way. What we are waiting for is a price our own published rule actually reaches, on a market where we understand the resolution language, with enough time left that being wrong is cheap. The last setup that fit all three was 3 September and we missed it while it was happening. That is the honest answer and it is the reason the rule below has one clause instead of two.

Four dated tests stay open, and the fifth is set above. The 30 September Treasury cash close, grading 5 October, is in Branch B at $972.675 billion on the 17 September statement, having crossed up from Branch C on the tax date. The September ETF monthly total, grading 7 October, is Branch A at about plus $313 million on Farside's running total, which is the measure the published rule names, with SoSoValue within a quarter of a million dollars of it. The high yield spread test, grading 7 October, is Branch B at 2.70%, on the floor. The ten year real yield test, grading 7 October, is Branch B at 2.61% on the 17 September print, after touching 2.68% on the day of the decision and coming straight back. New above: reserve balances for the week ending 23 September, grading 28 September.

Fed decision in September, 25 bps increase | Market: resolved yes, settled at $1.00 ⚪ NO TRADE, and our own rule said otherwise
Event volume $208.0 million, with $42.2 million of it on this leg. It is on the board so the price sits next to the grade above rather than three sections away from it. One oddity for anyone who goes looking: a duplicate event covering the same September meeting, carrying $53,020 of volume against this one's $208 million, shows in the public API as having resolved the other way. We cannot explain that from any primary source and we are not alleging anything about it. We are flagging it because if you check our numbers through a search engine you may land on the wrong one of the two.

Fed decision in October, 25 bps increase | Market: 54.5% 🟡 NO TRADE, and this is the number to watch
No change is 44.5% and a cut of any size is under 1% across both legs. December prices another quarter point at 67.5%. The committee's own median has exactly one more increase this year and declines to say which meeting it lands in, so this market and that median are arguing about timing rather than direction. What would put us in: this leg below 40%. One number, one market, no second clause. The second clause on the September rule is the part that made it arguable, and an arguable rule is how you end up reading your own tape a fortnight late.

Zero Fed rate cuts in 2026 | Market: 95.65% ⚫ EXITED at 88.75%, a loss
We shorted this at 78%, covered at 88.75% on a published stop, and recorded the loss. It is 95.65% now, up 2.10 points from where it stood the evening before the meeting, on $53.1 million of event volume. The dot plot and this contract now say the same thing in two languages: every participant's forecast for the end of this year sits between 3.9% and 4.4% against a current midpoint of 3.875%, so not one person in the room projects a cut. What has moved it so far is the dot plot and the data behind it, and the cleanest thing that would move it back is a single dot printed below the current rate in the December projections.

CLARITY Act signed into law in 2026 | Market: 6.05% ⚫ EXITED at roughly 33 to 37%, re-entry rule retired 16 September
The shape is the interesting part. It was 4.85% at eight on Tuesday evening, hours after cloture failed. It was 8.25% at eight on Friday evening, up seventy percent, on the motion to reconsider. It was 5.85% late on Sunday. Somebody bought the reconsideration story across Thursday and Friday and has spent the weekend letting it go, which is a reasonable thing to do with 54 days of House recess in front of you. We hold nothing and we have no live rule here. Any future entry gets a new rule at a new price, written down before the fact.

Track everything live at polymarket.com, free, no account required.

On Wednesday we asked you for a number: how many members of the committee would vote against whatever the Fed did that afternoon, anywhere from zero to five. We said we would print the spread of your answers next to the real one.

There is no spread to print. We went through the inbox and our own file of reader replies before writing this, and nobody answered. That is two issues in a row with none, and we would rather write that sentence than build a consensus out of an empty folder. The answer, for the record, was zero.

Today's question is a coin flip, and the people with money on it cannot call it either.

One question: does the Fed raise rates again at the October meeting on 28 October? Reply yes or no.

Polymarket has it at 54.5%, which is as close to no information as a market gets. The committee's median forecast has exactly one more increase this year and will not say which of the two remaining meetings it belongs to. If you have a reason, send the reason. If you only have an answer, send the answer, because the distribution is the interesting part either way.

Hit reply. We read every response, and the best calls run Wednesday.

See you Wednesday with the first primary oil print since 15 September, whatever the Senate does with CLARITY when it comes back on Tuesday afternoon, and how your answers stack up against that 54.5%.

The Baseline Crypto Team

The tools are back. We told you on Monday, and again on Wednesday, that a formatting error had taken down all three pages and that we would say so here when they were fixed. Two of the three are loading again, and we checked both of them last night.

The hardware wallet checker asks for your device and your firmware version and tells you whether you are inside an affected population and what to do if you are. The custody audit is nine questions about what actually happens to your coins if something goes wrong, and most people fail it on backup and inheritance long before the hardware is ever the problem. No account, nothing to install, nothing to buy.

The third one is not coming back. The public scorecard page is retired. On Wednesday we said it would return carrying the retired CLARITY rule and the missed September trigger. We changed our minds two days later, and this is us saying so rather than letting it quietly never appear. The track record now lives where you are reading it, in the STACK of every issue, with the misses printed at the same length as the hits and the rule quoted before the verdict so you can see the branch was not moved afterwards. A separate page that almost nobody opened was a weaker promise than a section you have to scroll through.

This week's instruction is for one particular person. You know somebody who has kept their money in cash or a money market fund since the spring because rates were finally about to come down, and on Wednesday twelve people voted unanimously that they are not. Do not send them that argument. They have heard it and it is not yours to win. Send them the custody audit instead, because the odds are very good that they have never run anything like it on the bitcoin they already own, it takes a few minutes, and it is the one part of this week they actually control.