Settlement rules

Settlement rules

04:12. Awake before the alarm, mouth like a coin, phone already lighting the ceiling. Consensus 42k. Unemployment 4.6. Average hourly earnings three-tenths. I have written those three numbers on a sticky note, which is a thing I do now, because writing them by hand makes them feel like they came from somewhere.

04:40. Let’s be precise about what I’m about to trade. A survey of economists, most of whom have not employed anyone since their last research assistant, forecasting the output of a survey of business establishments whose response rate has been sliding for a decade — from around sixty percent to something in the low forties, depending on which survey and which year you squint at. The gap gets filled by a model named after the fact that firms are born and die. It’s called birth-death. It is not a metaphor. It’s an ARIMA extrapolation with a body count.

Somewhere out there, actual people worked actual hours in October. That happened. It’s over. It’s knowable in principle and unknowable in practice, and in eleven hours I’m going to take a position on the difference between the unknowable thing and a sticky note.

05:15. The summer they fired the statistician is the moment I mark as the break. August 1, 2025: payrolls print soft (73k), May and June get revised down by a combined 258,000, and by the afternoon the woman running the Bureau of Labor Statistics — Erika McEntarfer — is out of a job. Not the model. Not the response rate. The woman. You can argue about her tenure, her methodology, whatever — I don’t care about the politics this morning. I care about the incentive gradient it drew on every career GS-14 who has to decide, next quarter, whether to flag an anomaly.

Nobody had to falsify anything after that. The chill does the work.

06:00. So the truth moved. It didn’t vanish, it just got a paywall. ADP relaunched its series with the Stanford Digital Economy Lab and started publishing weekly (the NER Pulse), and for about six weeks in late 2025 that private number was the only functioning window into American employment. Revelio scrapes profiles. Homebase counts clock-ins at restaurants. Indeed counts postings, which measure the desire to hire rather than hiring, a distinction the entire commentariat forgets every single month. Chase has card data. Equifax and TransUnion have the delinquency files, and honestly, TRU’s income-verified arrears tell me more about the marginal American household than anything with a seal on it.

The funny part is that ADP is a listed company. The measurement of the labor market is a ticker with a dividend. You can buy the thermometer.

And the Board buys the same subscriptions I do. That’s the part that should terrify people more than it does. The reaction function is now downstream of vendors, and the vendors have quarterly revenue targets and churn metrics and a product team deciding how to smooth a series. My edge used to be understanding the plumbing. Now my edge is which Bloomberg terminal add-on the firm pays for.

07:20. Coffee number two. I pull up the prediction markets, which is the newest and most honest form of self-harm available to me.

There’s a contract paying out on core CPI at three-tenths. It’s at 61 cents. There’s one on payrolls above 50k, one on the unemployment rate rounding up. Read the settlement rules — actually read them, nobody does — and you find the source of truth is the release itself. Not the economy. The release. Whatever BLS publishes at 8:30, revised or absurd or eventually corrected by 900k, is the resolving event.

We have done this before. LIBOR was a poll of bankers’ estimates of a rate at which nobody was actually borrowing, and we hung several hundred trillion dollars of notional off it, and it worked beautifully for thirty years because the number was the contract. The number never had to be true. It had to be published. Then one day people noticed the difference and the whole thing had to be dismantled and replaced with something anchored to transactions.

Macro data is drifting toward the same ontology. The preliminary benchmark revision announced in September 2025 took 911,000 jobs out of the year ending March 2025 — a year that everybody had already traded, hedged, and voted on. Did anything change? Not one hour of work was un-performed. The past simply got rewritten, and the contracts had already settled.

08:24. Positioning is cowardly. Two-year barely moved on the week, the December strip is priced for a cut nobody has conviction about, and crypto is where the fear actually shows because it never closes. BTC has been slipping through the Asia session on thin spot volume, perp funding pinned negative, ETH doing that thing where it underperforms on the way down and underperforms on the bounce. Somebody liquidated a nine-figure long at 03:40 and I felt it in my own book, which is an insane sentence about the relationship between digital coins and American employment statistics, and yet.

Bitcoin is the fastest available expression of a macro opinion. That’s its actual product-market fit. Not payments, not the reserve asset thing, not the gold cosplay. It’s the only casino open when the doubt arrives.

08:29. Fifty seconds. Hands cold. I hate this and I have arranged my entire life around it.

08:30. Twenty-eight thousand. Below consensus. Unemployment ticks. And on page three, in the table nobody opens, the prior two months come down by sixty-one thousand, which means the level was wrong before it was wrong today, which means the sticky note was fiction on arrival.

Algos take the headline. Two-year rips eleven basis points. BTC does $1,400 in ninety seconds. The Kalshi contract goes to a dollar because the release resolved, cleanly, definitively, and forever.

08:31. I’m up. Whatever happened to those people in October is still happening, unmeasured, and I got paid on the spread between two guesses.