On Friday 4 September 2026 the 2-year US Treasury yield climbed “eight basis points” (Bloomberg), peaked at 4.4246%, its highest since January 2025 (Reuters), and rose three basis points to 4.37%. All three sentences are accurate. The first two are wire services describing the intraday move after the 08:30 ET payrolls release, each against its own prior-day reference; the third is the US Treasury’s daily par yield curve file, the series this desk uses for every rates figure it prints and the series our marker J1 was graded on. J1 needed five basis points. On the peak it would have passed by three or four, depending on whose base you subtract from; on the close it failed by two. This guide is about why we chose the close before we knew the answer, what the two numbers actually measure, and how to read a yield headline so that the number you carry away is the one you meant to.
It joins the Reading Room as the forty-second field guide, and it follows directly from #40, which scores data days on the 2-year’s close, and #41, which scores a speech on the same series. Both of those guides assumed you knew what “the 2-year’s close” is. Friday showed that assumption doing work.
Check one: know what the Treasury’s number is
The figure in the Treasury’s daily file — the one FRED republishes as DGS2, DGS10 and so on — is a constant-maturity yield: a point read off a par yield curve that the Treasury fits each trading day to market quotations on its recently auctioned securities. Per the Treasury’s own methodology notes, the inputs are bid-side quotations obtained by the Federal Reserve Bank of New York at or near 3:30 PM Eastern, and the curve is interpolated (the monotone convex method since late 2021) to read yields at fixed maturities. Three consequences follow. It is a bid, not a mid or a last trade, so it sits a fraction above the mid-market yield. It is a 3:30 PM snapshot, not a 5:00 PM settlement, not a 4:00 PM equity close and not the last print of the overnight session. And it is interpolated, so “the 2-year” in the file is not any one bond; it is the curve’s value at exactly two years. The practical point: the number is published once, in the evening, and never revised. A claim tied to it has a single, permanent answer.
Check two: know what the wire’s number is
A wire service’s yield is a quote from a data vendor’s screen at the moment the sentence was written, usually a mid or a last-trade yield on the current on-the-run note, and the story is updated through the day. “Peaked at 4.4246%” is the highest such quote the reporter saw; “last up 4 basis points at 4.37%” is the quote at filing time, measured against the vendor’s own prior-day reference. On Friday morning Reuters had Thursday’s 2-year at 4.33%; the Treasury’s file has 4.34%. So Reuters’ +4bp and the Treasury’s +3bp describe the same afternoon level from different starting points, and neither is wrong. The same applies to the peak: Reuters’ 4.4246% is +9.5bp on Reuters’ base and +8.5bp on the Treasury’s, and Bloomberg’s “eight” is presumably on a third. What you should not do is subtract one source’s peak from the other source’s close, or add a wire’s intraday move to the Treasury’s prior close, and report the result as a fact. Mixed bases produce numbers nobody published.
Check three: put the peak and the close in the same sentence
The headline number on a data day is almost always the peak, because the peak is what happens in the first ten minutes and the story is filed in the first thirty. The close is what the market thought after six more hours. On Friday the 2-year’s peak-to-close give-back was, on Reuters’ peak against the Treasury’s close, about 5–6 basis points: roughly two-thirds of the initial move on either base. That is information. A move that holds to the close is a repricing; a move that gives back most of itself by 3:30 PM is a repricing that met sellers of the idea. The way to carry that information is to write both: “the 2-year touched 4.42% after the print and closed at 4.37%, +3bp on the day.” The way to lose it is to write either alone. Our own lead today prints both, and the markers piece grades on the second.
| Source | Clock | Side | Thu 3 Sep level | Fri 4 Sep figure | Change as stated | Revised later? |
|---|---|---|---|---|---|---|
| US Treasury CMT (daily file) | ~15:30 ET | Bid, interpolated par | 4.34% | 4.37% close | +3bp | No |
| Reuters, morning report | ~05:00 ET | Vendor quote | 4.33% | “held at 4.33%” | −5bp overnight | Story updated |
| Reuters, afternoon report | Intraday | Vendor quote | 4.33% | peak 4.4246%; last 4.37% | +9.5bp at peak, +4bp last | Story updated |
Sources: US Treasury daily par yield curve, 2026 file, pulled 5 September 2026 06:10 UTC; Reuters reports of 4 September 2026 as republished by Yahoo Finance. Bases differ by source; do not net across rows.
Check four: decide, before the print, which number your claim grades on
This is the check that matters for anyone who makes a forecast and intends to be honest about it afterward. A claim like “payrolls will move the 2-year at least 5bp” is not falsifiable until it names a source and a clock, because on Friday it was true on one and false on the other. The rule this desk uses: grade on the series that is published once, by a primary source, and never revised; and name it when the claim is set, not when it is graded. For Treasuries that is the CMT close. It is a worse measure of the initial reaction than the peak — it is meant to be; it measures where the day ended — and a claim graded on it will fail on days like Friday when the reaction faded. That is the point. If you want to test the reaction, set the claim on the peak from a named vendor in advance and accept that you will be arguing about ticks. If you want to test whether the market’s view changed, use the close. J1 was set on the close on Thursday morning, with the Treasury file named, and it failed. Re-grading it on the peak afterward would be the error this guide exists to prevent.
Check five: score the close against the year, not against the headline
A 3bp close on a payrolls day sounds small beside “jumped eight.” Against the 2026 file it is ordinary: of 170 daily changes, 64 (37.6%) were 5bp or larger, the median absolute change is exactly 3bp, and the nine payrolls days average 4.4bp with a median of 3bp. Friday is at both medians. The surprise was in the print (162,000 against 53,000), not in the close. A useful habit is to keep two percentiles in your head: where the print sits in the distribution of surprises, and where the close sits in the distribution of moves. When the first is extreme and the second is not, the market had already moved — Thursday’s speech had taken 5bp out, and Friday put 3bp back — or the market did not believe the print, or the print was not what the market was waiting for. Waller’s speech said in advance which of those it would be: “I don’t expect that the employment data will deviate much… my decision… will be heavily influenced by what we learn about August inflation.” The rates market, on the close, agreed with him about which print matters.
Check six: put bitcoin beside it, on the same clocks
Bitcoin trades through the 3:30 PM Treasury snapshot, the 4:00 PM equity close and the overnight, so “bitcoin’s reaction to the yield move” needs the same discipline about clocks. This desk uses the Bitstamp daily candle, which closes at 00:00 UTC (8:00 PM ET on Friday), so bitcoin’s “close” is four and a half hours after the Treasury’s and includes the after-hours session. On Friday that candle ran from $81,265.00 to $79,676.60 (−1.955%) with a low of $78,645.01 that Bloomberg, on its own clock, reported as $78,649 and “as much as 3.5%.” Same discipline: peak fall 3.5%, close fall 1.955%, both printed. The correlation you are testing is between a bid-side 3:30 PM par yield and a 00:00 UTC exchange candle; that is fine as long as you say so, and it is the reason Guide #40’s bitcoin test returned a negative result — on daily closes, payrolls days and ADP days do not separate. Friday added one observation to each side of that: a 3bp rates close and a 1.955% bitcoin close, on a print that was three times consensus.
The checklist
Identify the source and its clock before you copy the number. Know that the Treasury’s figure is a 3:30 PM bid-side interpolated par yield, published once and never revised, and that a wire’s is a vendor quote at filing time against the vendor’s own prior level. Print the peak and the close in the same sentence and never subtract across sources. Name the series and clock a claim grades on when you set it, and grade it there whatever the other number says. Score the close against the year’s distribution, and the print against the year’s surprises, separately. Put bitcoin beside it on its own stated clock. Worked on 4 September 2026, it reads: 162,000 against 53,000; peak 4.4246% per Reuters (+8.5bp on the Treasury’s base), close +3bp per Treasury; J1 fail; the 2-year at the payrolls-day median; bitcoin −1.955% on the candle, −3.5% at the low; and the print the market is waiting for is still the CPI on the 11th.
As standing practice each field guide ends with a claim that can be checked. O1: on Friday 11 September 2026, the CPI release day, the 2-year Treasury’s CMT close moves at least 5bp from Thursday 10 September’s close, in either direction, per the Treasury’s daily file. It is the same bar as J1, moved to the print that Waller named. If it passes and J1 failed, the rates market told us which report it was waiting for; if both fail, the 5bp bar is too high for a single print in this regime and Guide #40’s averages are being carried by one June day. We set it on the close, with the file named.
Method: prices, funding, open interest, basis and on-chain figures in this article are pulled directly by Bitcoin Mastery at the timestamp stated — Bitstamp BTC/USD daily candles for closes, Binance BTCUSDT spot and USDT-margined perpetual for intraday, open interest, funding and account ratios, Binance COIN-M quarterly contracts for basis, mempool.space for difficulty, hashrate, pool shares and fees, alternative.me for the Fear & Greed series, Farside Investors’ table for ETF flows (every named fund column is read from a print of that column with its header) and US Treasury CMT par yields, nominal and real, for rates. Where a third-party figure is cited we name the source and its date; where two sources disagree we print both. Every streak or extreme figure is published with the first date of its series in the same sentence.
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Cryptocurrencies are volatile and you can lose money. Nothing here is a recommendation to buy or sell any security, digital asset or exchange-traded fund, including MSTR. Do your own research and consult a licensed financial advisor before making investment decisions.