On Tuesday 29 September the market thought there was a two in three chance the Federal Reserve raises rates this month. It does not think that any more. The October leg on Polymarket stood at 67.5% before New York Fed President John Williams spoke in Buffalo that afternoon. It is 16.5% tonight, and no change carries 82.5%.

A hike coming off the table is supposed to be good news, and it is supposed to show up in credit first. One fewer rate rise means a lower path for the cost of borrowing, an easier refinancing next year, and a smaller premium demanded of anybody who has to go to the market for money. Spreads come in. That is the mechanism and it is not controversial.

Credit went the other way.

ICE BofA's US High Yield index option-adjusted spread, the extra yield demanded to lend to companies rated below investment grade, closed at 3.02% on 28 September. Then 3.08%. Then 3.12%. Then 3.24% on 1 October. Wednesday's scorecard turned on this series at 3.02% and called it the highest since 7 April. It has widened another 22 basis points since that email went out, 58 basis points across eight sessions since 2.66% on 21 September, and at 3.24% it is the highest since 31 March.

The 1 October move alone was 12 basis points. That is not a record, and we checked before writing that it was not: 27 March put on 21 in a day. But 25 September put on 13, which means this one episode now contains two of the five largest daily widenings of 2026. Investment grade went 0.77% to 0.86% over the same stretch, so the move is not confined to the junk end, it is just far larger there.

Here is the part we cannot explain, and we are going to say so rather than reach. No default, no bankruptcy filing, no CLO or private credit blow-up, no bank event. Nothing happened in the window that accounts for 58 basis points. Spreads widened on no news, which is either the market knowing something it has not articulated, or quarter end plumbing dressed up as a signal. We will tell you which, and the number that decides it is in BEYOND.

Put the two facts side by side anyway. A rate rise got materially less likely, and the price of lending to a company that needs money went up. Whatever is in that second number, it is not the first one.

This newsletter has spent three weeks on the cost of money. The ten-year at a nineteen-year high, and it went higher after we said so: 5.29% on 30 September, which is the highest print since 14 May 2002, so the honest label on that series is twenty-four years rather than nineteen. Real yields at 2.93%, the highest print of 2026. The nine people who voted to hold in July voting to hike in September. All of that was a rates story, and a rates story has a tidy ending where the Fed goes too far and then stops. Last week was not that. Lenders were handed a materially lower chance of a rate rise and charged more anyway, and the question of what they were actually pricing is the FILTER's.

And bitcoin went up through both of them.

The Coinbase daily closes since Williams spoke: 83,638.42, 83,556.14, 84,848.73, 84,504.88, 84,742.22, and 86,507.11 on Sunday. First to last is 3.4%, earned across the same days credit was deteriorating.

Q3 closed while that was happening, which changes what we are allowed to say about it. On Wednesday we wrote that bitcoin was on course for plus 42.60% and said in the same breath that this was a position and not a result, because the quarter had two sessions left. It has none. From the 30 June close of $58,523.93 to the 30 September close of $83,556.14, bitcoin returned 42.77%. Best quarter since Q4 2024 at plus 47.48%, and it ends three consecutive losing ones.

Now the sentence that has to go next to it, because a 43% quarter reads like a triumph and the year does not. On those same closes bitcoin is down 26.75% from 30 September 2025, when it closed at $114,067.71. The three losing quarters this one ended cost 48.69% between them. That is why a 43% gain still leaves a holder from a year ago well behind, and anybody telling you Q3 was a comeback is quoting the quarter and skipping the year.

Three metrics.

Metric 1, Fear and Greed: 70. Wednesday's issue printed 73. Since then: 71, 74, 72, 67, 65, and 70 this morning. Mind the stamp on that series, because it is easy to read a day late: the index posts a value the evening before the date it carries, which makes this morning's 70 Sunday's reading and the 65 Saturday's. So sentiment went up five points on the day bitcoin added 2.1% and closed the window at its high, which is the gauge doing what it is built to do rather than anything to interpret. The level is the part worth holding. At 70 it is three points below where we printed it on Wednesday, after a window that added 3.4%.

Metric 2, ETF flows: September closed at plus $2,647.7 million. That is our own sum of Farside's printed daily totals, the measure we stated in print, and it is the figure our dated test grades on Wednesday, so treat it as a position and not a verdict. The last two sessions of the month were plus $66.2 million and minus $148.7 million. Do not read that outflow as a quarter end flush: 84% of it was a single fund. October opened plus $102.7 million. We are not printing a 2 October total, because IBIT, which is roughly 62% of the complex's net assets, has not disclosed, and the trackers disagree about whether that is a dash or a zero. A total with the biggest fund missing is not a total.

Metric 3, perpetual funding: about 0.003%, which is zero again. Deribit's eight hour rate on the bitcoin perpetual, against 0.00333% when we printed it on Wednesday. At that size it moves every few minutes, and it has not been anywhere that matters in a fortnight. Bitcoin added 3.4% and nobody paid anything to be positioned for it, while lending to a leveraged company got 22 basis points dearer than the number we published on Wednesday.

The STACK grades two dated tests this morning and reports a published rule of ours that fired ninety minutes after Wednesday's email landed, with nobody at the desk to act on it.

📣 THIS SPOT IS OPEN

Baseline Crypto reaches nearly 7,000 Bitcoin-focused readers every Monday and Wednesday morning, with an open rate just under 50% over the last four weeks. Direct sponsor placements are open, premium position, top of the issue, one per send.

Reply to this email for rates.

"Spreads are widening because the Fed is about to hike again. December is at 71.5%. Credit is just telling you what another twenty five basis points does to a leveraged balance sheet."

No. Credit is not pricing the hike, and the proof is that credit widened fastest on exactly the days the hike was being taken away.

Concede the real part first, because there is a lot of it. December sits at 71.5%, so the market genuinely does expect a rate rise before the year is out. And a higher policy rate really does hurt below investment grade borrowers more than anyone else, because a large share of that debt is floating and reprices straight into interest expense. If you wanted a clean story about spreads, this is the one you would reach for, and plenty of people did last week.

Now look at the sequence, which is what the story cannot survive. The October leg fell from 67.5% to 16.5%. Over that same stretch spreads went from 2.93% on 25 September to 3.24% on 1 October. The days on which a rate rise was being removed from the front of the curve were the days credit deteriorated hardest. If spreads were pricing the hike, those two series move together. They moved apart, and they moved apart violently.

There is a second test and it is better than the first. On 1 October the Vice Chair of the Federal Reserve put in writing that he sees upside risks to inflation. In the ordinary course that is a hawkish document and the front end reprices toward a hike. The October leg did not move back up. Spreads widened twelve basis points that day. And the man whose speech had taken October off the table two days earlier was not arguing the other side of this. Williams called inflation unquestionably too high and guided to one further upward adjustment late this year, just not in October, which is precisely why December kept the hike while October lost it. So credit was handed two policymakers flagging the same inflation risk inside seventy two hours, differing only on when, and it widened through both of them. A market that widens through both is not trading the Fed.

So ask what each of these prices is actually paid to express, which is the question this section always ends up asking.

The December leg at 71.5% is paid to express timing. It says not in October, probably in December, and between those two meetings sits a quarter of data. It is a bet on a calendar. A credit spread is paid to express something entirely different: the probability of not being paid back. A borrower who gets into trouble gets into trouble on cash flow, on a refinancing window that will not open, on a customer who stopped ordering. Not on the target range. The policy rate is an input to corporate distress and it has never been the whole of it, and in a week where nothing else moved it was clearly not the thing moving.

Which gives you the reading almost nobody is saying out loud. A hike priced out alongside spreads at their widest since March means lenders have stopped treating the central bank as the main risk in the room. That is neither hawkish nor dovish. It is the market pricing the reason the Fed might back off rather than the fact of it, and the SIGNAL is honest that no event in the window explains the move, which is exactly what it looks like when a market is repricing a condition rather than reacting to a headline.

It also explains the thing in the SIGNAL that otherwise looks absurd, which is bitcoin adding 3.4% through a credit selloff. Bitcoin is long liquidity. It is not long credit quality. It has no exposure of any kind to whether a mid-market borrower makes a coupon in March, and a world where the Fed stops tightening sooner than expected is a world with more liquidity in it. The textbook claim that a risk asset should sell off with high yield assumes the asset sits somewhere on the same capital structure. It does not sit anywhere on it.

The honest hole in our own argument, and we would rather print it than have you find it. If this were a straightforward growth scare, somebody would be starting to price a rate cut. Nobody is. Zero cuts in 2026 is still at 95.95% on the board below, barely changed on the week. A market removing hikes without adding cuts is saying something narrower than recession: it is saying the tightening is probably over and the damage is landing on borrowers rather than on the economy as a whole. That is the version of this we would defend. It is not the version that sells a newsletter, and it is the one the data supports.

One uncomfortable note to finish, because it is ours. The rule we retired in print on Wednesday, the 2 September entry condition, was reaching for precisely this. It said spreads widening through 3.00% into the October peak was the trade. It had the signal right. It was useless anyway, because it never named an instrument. Being early to a correct observation and unable to act on it is not the same as being right, and the STACK has considerably more to say about that in a moment.

Three numbers that define the next week.

3.46%, the level that decides whether this is the worst credit episode of the year or the second worst. That is where high yield spreads peaked on 30 March 2026, twenty two basis points above Thursday's 3.24%. The distinction is not academic, and it is the number that settles the thing the SIGNAL admits it cannot explain. March is the counter-example: spreads hit 3.46% on the 30th and were back under 3.00% by 8 April, seven sessions later, and in hindsight nobody calls it a credit event. If this episode takes out 3.46% with the October hike still priced out, the growth reading in the FILTER stops being an interpretation and becomes the consensus, and it will arrive with an equity drawdown attached. If instead spreads retrace under 3.00% inside a fortnight, last week was quarter end plumbing wearing a macro costume and this issue made too much of it. We will say which, with the series printed, whichever way it goes.

2.88%, the first lower print in the real yield, and the test of whether 2.93% was the top. A week ago Monday's issue called 2.85% the highest since 2008 and Wednesday's noted that 2.91% had already passed it. The series then made 2.93% on 30 September, its 2026 high, and fell to 2.88%. What to watch is whether real yields keep coming down while spreads keep going up. That precise combination, the government paying less to borrow while companies pay more, is the cleanest signature of a growth scare available in daily data, and you can read it off two free FRED series without paying anyone for a terminal. If instead both rise together from here, we are back in a rates story and the FILTER above is wrong.

7 October, when three of our own dated tests grade at once. High yield spreads, the ten-year real yield and the September ETF total all take their readings from 30 September and all resolve on Wednesday. All three are currently sitting in the branch our own argument implied, which is a more dangerous place to be than it sounds, because the test that confirms what you already believe is the one you check least carefully. The September FOMC minutes should land that day too, three weeks after the decision per the Board's stated practice, though that date is our inference from the practice and not a published confirmation, and no release time is published at all. Wednesday is the heaviest grading day this scorecard has had.

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 this morning, and a published rule of ours fired ninety minutes after Wednesday's email went out with nobody at the desk. Rule, number, verdict, consequence, in that order, for each of them.

Test one, the Treasury cash test, set 2 September. The rule: the closing balance on the Daily Treasury Statement, as of the 30 September statement. Branch A is $1,050 billion or above. Branch B is $950 billion to $1,050 billion. Branch C is below $950 billion.

The number: $984.046 billion. The verdict: Branch B.

The consequence is worth more than the grade. Monday's issue had this tracking Branch C at $924.627 billion, which was the 24 September close, and called it $25.4 billion under Treasury's own published assumption with three business days left. It then printed 945.290, 959.572, 936.600, and closed the quarter at 984.046. Treasury finished $34.0 billion above the $950 billion it assumed on 3 August, having been $25.4 billion below it on the 24th. An agency that publishes a quarter end cash target and lands within thirty four billion of it, after being twenty five billion short with three business days to go, is not improvising.

One procedural note, because it is the sort of thing that quietly ruins a scorecard. This rule grades the 30 September close and nothing else. The balance on 1 October was 893.699, which is a fall of $90 billion in a single day once the quarter had turned. Grading off the current balance rather than the dated one flips the answer from Branch B to Branch C, and that page opens on the latest statement rather than on ours, so pick the date. The rule named a date, and this is what the date was for.

Test two, the SOFR test, set in this newsletter on 30 September. The rule: SOFR as published by the New York Fed for 30 September 2026. Branch A is 4.00% or above. Branch B is 3.90% up to but not including 4.00%. Branch C is 3.80% up to but not including 3.90%. Branch D is below 3.80%.

The number: 3.90%. The verdict: Branch B. The 99th percentile printed 3.99%, still underneath the standing repo facility's 4.00%.

We are not going to oversell this, and you should be suspicious of us if we try. Branch B is the branch Wednesday's FILTER implied, and Branch A was the one we named first as the branch that would cost us. We predicted the ordinary outcome and the ordinary outcome happened. That is not a hard call. The honest version is narrower and still worth something: the squeeze story going round the market was wrong, and we said it was wrong with a number and four branches attached, in advance, in public.

What did happen on the day is more interesting than the rate. SOFR volume hit $3,230 billion on 30 September, the highest in the forty five day series and roughly 9% above normal, and the rate did not move at all. The reverse repo facility took in $11.539 billion, up from $851 million two days earlier, which is the usual quarter end scramble for somewhere to park cash rather than a sign of stress. Against the $2.5 trillion it held at its 2022 peak it is still all but drained. Record volume, a thirteenfold jump in reverse repo usage, and no price response. Quarter end pressure got absorbed in quantity instead of price on the one day everybody expected the plumbing to strain. SOFR then printed 3.87% on 1 October, below the band entirely, which is the quarter end bump unwinding exactly as that FILTER said it would.

Test three, and this is the one that matters this morning. Our live entry rule fired, and we did not act.

The rule, published in this section on Wednesday, in full: "What would put us in: this leg below 40%." We printed it at 44.5% that morning, said it was four and a half points away, and described it in that same section as "a rule that names a contract, a side and a price, which is more than the one retired above ever did."

The time it fired. Polymarket's own hourly price history has this leg at 0.435 at ten o'clock on Tuesday evening, 0.415 at four on Wednesday morning, and back at 0.435 before nine. Then the bar stamped nine o'clock on Wednesday 30 September took it from 0.435 to 0.335 with nothing printed in between. So it crossed inside the hour before nine and was already through by then, and there was never a 40 cent print to transact at. It closed Wednesday at 0.335, then 0.235 on Thursday afternoon, 0.155 on Friday morning, and 0.165 now.

Wednesday's issue sent at half past seven. The rule fired at nine. Ninety minutes. We told you in print that morning that this leg was four and a half points from putting us in, and ninety minutes after that email landed in your inbox it was through. We were not there, because the issue had just gone out and nobody was at the desk.

The arithmetic that makes this honest. Buying at the first print below our level, 33.5 cents, puts that position at 16.5 this evening. Down 17 points, a 51% loss. So not following our own instruction saved us a substantial amount of money.

The verdict: that was luck, not judgement, and we are not going to let it read as anything else. On 21 September we graded a different missed trigger as worse than a pass, because a pass is a decision and that was not. The same sentence applies here. The only difference is that this time the accident went our way, and that makes it more dangerous rather than less, because it invites precisely the wrong lesson. A rule you follow only when you happen to be at a screen is not a rule. It is a preference with a number in front of it.

Set this against what we did on Wednesday, because the pair is the whole point. On Wednesday we retired a rule for naming no instrument, and in the same breath praised this one for naming a contract, a side and a price. It did name all three. Every criticism we made of the other rule, this one answered. It still did not execute. Naming the trade turns out to be necessary and nowhere near sufficient, and what was missing here was not the specification. It was anybody to act on it.

The consequence: the rule comes off the board this morning. It needed either a mechanism that does not depend on an issue being written that day, or retirement, and we do not have the mechanism, so it gets retired rather than carried as decoration. We are not replacing it today, because writing a new entry at a new price hours after watching the old one run away from us is the goalpost moving this section grades other people for.

Which leaves the board with no live rule on it at all, and that is the direct consequence of the paragraph above rather than a separate development. The last two rules both came off for defects in how they were written rather than for being wrong about the world.

The three open tests, with live levels, none of them graded today.

High yield spreads on 30 September, grading 7 October. Branch A is 3.00% or above and the 30 September print is 3.12%, which is Branch A. When we set this on 9 September the latest value was 2.68%, sitting in Branch C, and we wrote that Branch A was what our own argument implied and 32 basis points away. It takes its reading from Wednesday's print and it does not grade until Wednesday.

The ten-year real yield on 30 September, grading 7 October. Branch A is 2.80% or above. FRED's print for 30 September is 2.93%, which is Branch A, and it fell to 2.88% on 1 October. The rule names FRED and grades on FRED.

US spot bitcoin ETFs net positive for September, grading 7 October. Tracking Branch A at plus $2,647.7 million, on our own sum of Farside's printed daily totals, which is the measure we stated in print. The month's last two sessions were plus 66.2 and minus 148.7. Nothing about the final week threatens the branch, and the figure gets restated on Wednesday with the arithmetic shown rather than asserted.

Fed decision in October, 25 bps increase | Market: 16.5% ⚪ RULE WITHDRAWN, and not because the rule was wrong
No change is 82.5%. We printed this at 44.5% on Wednesday and at 64.5% a week ago today, so the hike has gone from the base case to a one in six inside eight days. Read the retirement above before reading anything triumphant into the level. The reason this entry is now a blank rather than a position is that the condition we published came and went while nobody was watching, and the honest response to that is to take the rule down, not to claim the level. What would make this tradeable again: a mechanism that fires without an issue being written, which is a production problem rather than a market one, and we will say in print when it exists.

Fed decision in December, 25 bps increase | Market: 71.5% 🟡 NO TRADE, and the calendar story just weakened
No change is 25.5%. On Wednesday we argued that December holding at 75.5% while October collapsed meant the hike was being removed from the front end rather than rescheduled. December is 71.5% now, four points lower, so that argument needs updating in our own words rather than quietly. The hike is no longer merely losing its October slot, it is losing weight across the year. It is still the most likely single outcome for December, which is why this stays a hold rather than becoming a short. What would move it: the 28 October decision, which settles the timing question in an afternoon and this contract five weeks later.

Zero Fed rate cuts in 2026 | Market: 95.95% ⚫ EXITED at 88.75%, a loss, and it is the counter-evidence to our own argument
Short from 78%, trimmed at 85%, covered at 88.75% on a published stop on 3 August, recorded as a loss in every issue since and still one. We printed 96.25% on Wednesday and it is 95.95% now. The reason this entry earns its place this morning is not the thirty hundredths of a point. It is that the FILTER above argues the market is pricing a deterioration in borrowers, and if that were the whole story this contract would be falling rather than sitting at ninety six. A market that strips out hikes without adding a single cut is drawing a narrower conclusion than recession, and this is the number that keeps us honest about it. What would move it properly: one dot below the current midpoint in the December projections, from anybody in the room.

CLARITY Act signed into law in 2026 | Market: 5.0% ⚪ NO TRADE, and it is now the oldest of three retirements
5.5% on Wednesday, 5.0% now, and we hold no rule here because the re-entry condition was retired in print on 16 September when the markup leg it turned on became unreachable. That makes three published rules off this board inside three weeks, and not one of them died of being wrong about the world, which is the pattern worth noticing. What would make this tradeable again: a markup of the substitute text, and a new rule at a new price written before the fact rather than after it.

The book. Flat for the eighteenth consecutive issue, recomputed from the archive rather than carried forward: seventeen issues published after the 3 August exits, so this is eighteen. On Wednesday we told you one published rule was live. This morning there are none, and the section above is why. Eighteen issues of no position is either discipline or paralysis, and we are no longer confident it is the first one. What would change it is a rule written with execution attached rather than a level attached, and that is a thing we have to build before we are entitled to publish it again.

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

On Wednesday we asked one question with a one word answer: which is telling the truth about the next six months, bonds or bitcoin.

Nothing came back. We went through the inbox and our own file of reader replies before writing this, and there is still nothing new since 3 August. That is six issues running.

Wednesday's issue said that if the answer was silence again we would write this paragraph for a sixth time and it would be the same paragraph. It is the same paragraph. We said we would not redesign the question and we have not. What follows is a different question in the same shape.

This issue has one argument in it. Lenders were handed a materially lower chance of a rate rise and charged companies more anyway, 58 basis points more across eight sessions, with no default and no bankruptcy and nothing in the window to point at. Either they know something nobody has written down yet, or that was quarter end plumbing and it will be gone in a fortnight.

One question: which is it? Reply with one word, growth or plumbing.

If you have a reason, send the reason. If you only have the word, send the word.

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

See you Wednesday with three dated tests grading at once, the September ETF total with the arithmetic shown, and whether spreads went for 3.46% or came back under 3.00%.

The Baseline Crypto Team

Two tools. Free, no account, nothing to install, and both checked again before this went out.

The hardware wallet checker takes your device and firmware version and tells you whether you are sitting inside a population with a known problem, and what to do about it if you are. The custody audit is nine questions about what happens to your coins when something goes wrong, and it is deliberately unkind about backups and inheritance, because that is where almost everybody actually fails.

At nine o'clock on Wednesday morning a number we had published a rule against went through the level, and we found out about it days later. We were not asleep at the wheel. There was no wheel. Nothing we had built would have told us, because the whole arrangement depended on a person happening to look at a screen on a morning when that person was doing something else.

You almost certainly have the same arrangement. There is a price at which you would move coins off an exchange, or add, or finally sort out the backup you have been meaning to sort out since the spring. You know the number. You have no alert on it, no standing order, and nothing that happens if the level prints at three in the morning while you are asleep. It is the same defect we just retired a rule for, and ours at least was written down.

So do the smaller version of the fix. Open the custody audit, and before you answer question one, set an actual alert at your actual number on whatever app you already have open. Takes a minute. Then answer the nine questions, which are about the other half of the problem: what executes when you are not there.

Then send it to the person you thought of while reading that. You know the one. The friend who checks the price eleven times a day and has never once set an alert.