How to Tell Which Jobs Print Actually Moves Rates — ADP, JOLTS, Claims and Payrolls Scored on the 2-Year Treasury, and What the Same Test Says About Bitcoin — Field Guide #40

A jobs week has a rhythm. Tuesday brings JOLTS, the job-openings survey, six weeks stale. Wednesday brings ADP’s private payrolls estimate. Thursday brings weekly jobless claims. Friday at 08:30 ET brings the Bureau of Labor Statistics’ Employment Situation — nonfarm payrolls, the unemployment rate, average hourly earnings — and everything else in the week is a rehearsal for it. Each of the four produces a headline that says the market “reacted,” and by Friday evening you will have read four contradictory accounts of what the labour market told the Fed.

This guide gives you a scoreboard that does not care about the headlines. The 2-year Treasury yield is the cleanest daily expression of what the market believes about the next several Fed decisions, the US Treasury publishes its close every evening in a file you can download for free, and scoring that file by release date takes ten minutes. The result, for 2026, is that payrolls Fridays move the 2-year 4.6 basis points on average, ADP Wednesdays move it 2.1, and an ordinary session moves it 3.7. One of those four prints is a market event and the other three are not, and you can prove it yourself before the next one lands. The guide ends with the same test run on bitcoin, which does not give a clean answer, and we print that result rather than pretend otherwise.

Why the 2-year, and not the hike odds

The tempting scoreboard is the CME FedWatch probability — “hike odds rose to 66%” — and we showed on Tuesday why it is a bad one: it is a live number, four outlets printed four values from it in 72 hours, and almost nobody prints the clock. The 2-year has three advantages. It has one official close a day, published by the issuer. It aggregates the whole path of policy over 24 months rather than one meeting’s binary. And it is a price, so it cannot be quoted at two values at the same time by two people who are both right. The 10-year is the better instrument for bitcoin — Field Guide #39 explains why the real component of it is bitcoin’s opportunity cost — but for the question “did this print change what the Fed will do,” the 2-year is the instrument.

Check one: pull the file and confirm the row exists

The Treasury’s daily par yield curve is a CSV at home.treasury.gov/resource-center/data-chart-center/interest-rates/daily-treasury-rates.csv/2026/all?type=daily_treasury_yield_curve&field_tdr_date_value=2026&page&_format=csv. The 2026 file, pulled this morning, has a header row and 169 data rows, 2 January to 2 September, with columns for 1-month through 30-year. Print the row count and the first and last dates before you compute anything. The most common error in this whole exercise is not arithmetic; it is scoring a release day that is missing from the file (a holiday, or a day the file has not yet been updated) against the wrong neighbour. Yesterday’s row exists: 2 September, 2-year 4.39, 10-year 4.79. If today’s does not yet, do not fill it in from a broker screen; wait.

Check two: compute day changes from closes, not from levels

The number you want for each session is the close minus the previous session’s close, in basis points, where “previous session” means the previous row in the file, not the previous calendar day. Fridays compare to Thursdays; Mondays compare to Fridays; the Tuesday after a holiday Monday compares to the prior Friday. Then take the absolute value. A jobs print that pushes the 2-year down 6bp because payrolls were weak is exactly as much of a reaction as one that pushes it up 6bp because they were strong, and averaging signed moves would net them to nothing and tell you the market never reacts to anything.

Two lines of code do it: sort the rows by date, and for each row subtract the prior row’s 2-year. For 2026 that produces 168 daily changes with a mean absolute move of 3.7bp and a median of 3bp. That is your base rate, and it is the number every release-day figure has to be compared against. A print that moves the 2-year 3bp has not moved it.

Check three: score by release calendar, and print the row counts

Now overlay the calendar. The BLS publishes payrolls on the first Friday of the month, except when a holiday intervenes: in 2026 the releases fell on 9 January, 6 February, 6 March, 3 April, 8 May, 5 June, 2 July (a Thursday, because 3 July was the observed holiday) and 7 August. ADP publishes two days before, on the Wednesday: 7 January, 4 February, 4 March, 1 April, 6 May, 3 June, 1 July, 5 August and 2 September. Look each date up in your change series.

US Treasury daily par yield curve, 2026 file, 169 rows, own pull 3 September 2026, 06:10 UTC. Absolute day-on-day change in the 2-year constant-maturity close.

Print the counts. Eight payrolls days, nine ADP days, 168 session-to-session changes. A reader who sees “payrolls days average 4.6bp” without the eight is entitled to assume you cherry-picked, and a reader who sees the eight is entitled to note that it is a small sample — which it is. The 5 June print alone (+12bp, the day of the May payrolls report) contributes roughly a third of the payrolls total. Drop it and the payrolls mean falls to 3.6bp, in line with an ordinary day. That is worth saying out loud: the evidence that payrolls days are special rests heavily on one day, and the evidence that ADP days are not special is stronger, because it is consistent — three zeros, nothing above 6bp, nine observations.

Check four: name the confound on the same day

A release day is rarely only a release day. Before you attribute a move to a print, list what else happened in that session. This week’s JOLTS day, 1 September, saw the 2-year rise 5bp (4.34% to 4.39%), which looks like a JOLTS reaction until you recall that it was also the first session after US strikes on Iran, Brent settled nearly 6% above Friday’s close (and traded more than 7% above it at our Wednesday snapshot), and Japan’s 10-year closed at 3.00% for the first time since 1996. We split that day into 0bp of real yield and +4bp of breakeven at the 10-year, which is an inflation shock, not a labour-market one. The JOLTS miss cannot be credited with a move that the oil price explains. When a jobs day coincides with a Fed speaker, a CPI print, a Treasury auction or a geopolitical event, score it, footnote it, and give it less weight.

Check five: separate the miss from the level, and both from the revision

Wednesday’s ADP was reported as a miss — 38,000 against 47,000 — and it was. But 9,000 jobs is a small miss on a series whose month-to-month noise runs to six figures, and the more informative fact in the release was the level: the slowest month since January, with manufacturing down 17,000 and large firms supplying 34,000 of the 38,000. Separately, July was revised from 44,000 to 46,000. Three different numbers — the surprise versus consensus, the level versus recent months, and the revision to the prior month — each get their own headline, and the market prices the combination. A print can miss consensus, come in at a healthy level and carry a big upward revision at the same time, and the 2-year will tell you which of the three it cared about.

The same applies to Friday. Consensus for August payrolls, as cited by Morningstar, Kiplinger and FactSet this week, runs from +50,000 to +65,000, and individual forecasts run from +25,000 (Barclays) to +140,000 (Interactive Brokers). That 115,000 spread means the “miss” will be reported against whichever consensus the outlet uses. Score it against the 2-year instead, and the argument about which consensus was right becomes irrelevant.

What the same test says about bitcoin: not much, and we are printing that

The obvious next step is to run the identical scoring on bitcoin’s daily close. We did, on Bitstamp BTC/USD daily candles, 244 sessions from 2 January to 2 September 2026, absolute close-to-close change. The base rate is a mean of 1.67% and a median of 1.25%. On the eight payrolls days the mean absolute move was 3.12%; on the nine ADP days it was 2.30%. Both are above the base rate, and if we stopped there we would have a tidy story: bitcoin reacts to jobs data, and reacts more to payrolls.

We did not stop there, and the story does not survive. The payrolls mean is carried by 6 February, when bitcoin rose 12.32% in a session — a move nobody attributes to the January jobs report. Take the medians instead of the means and the order reverses: 1.75% on payrolls days, 2.46% on ADP days. Wednesday’s ADP day was a 0.124% move, the smallest of the nine. With eight and nine observations, one outlier and a reversal between mean and median, the honest reading is that this test does not distinguish bitcoin’s response to ADP from its response to payrolls. Bitcoin moves more on jobs days than on other days, probably, and that is all the 2026 data will support.

Own computation, 3 September 2026, 06:10 UTC. Treasury file 2 Jan–2 Sep 2026 (169 rows, 168 changes); Bitstamp daily candles 2 Jan–2 Sep 2026 (244 rows, UTC days). Bitcoin trades on weekends, hence the larger count.

Why does the test work on the 2-year and not on bitcoin? Because the 2-year has almost nothing else to react to. Its daily move is dominated by Fed expectations, and the labour market is the biggest input to those expectations, so a labour print is a large share of a small variance. Bitcoin’s daily variance is ten times larger and is driven by flows, leverage, liquidations and its own reflexive positioning, and a jobs print is a small share of a large variance. The signal is there — it is why we watch Friday — but it is not extractable from eight observations, and a guide that claimed otherwise would be selling you the same false precision it just warned you about in the hike odds.

The five traps, in order of how often we have fallen into them

Trap one: the missing row. Scoring a holiday or an unpublished day against the wrong neighbour. Print the row count and the last date before anything else. Trap two: the signed average. Averaging moves with their signs nets them to zero and proves nothing. Use absolutes for size, signs only for direction on a single day. Trap three: the same-day confound. A 5bp move on JOLTS day that was really an oil move. List the session’s other events before you attribute. Trap four: the outlier that is the whole result. One +12bp day is roughly a third of the payrolls total; one 12.32% bitcoin day is the whole reason the payrolls mean beats the ADP mean. Print the median next to the mean, always. Trap five: the consensus that was not yours. A 9,000-job miss against one consensus is a beat against another. The 2-year does not know which consensus you used.

The worked example, and the marker it produced

Run the five checks on this week. Check one: the 2 September row exists (4.39, 4.79). Check two: the day change is 0bp at the 2-year and 0bp at the 10-year. Check three: that is the third zero on nine ADP days this year, and below a 3.7bp base rate. Check four: no confound — Wednesday had the Beige Book at 14:00 ET, which is a description rather than a datum, and an equity rally led by a single stock. Check five: a 9,000 miss, a January-low level, a +2,000 revision, and a 2-year that priced none of the three. Conclusion: the rates market was not ignoring the labour market on Wednesday; it was waiting for Friday, as it has on every ADP day this year. Which is a falsifiable claim, so we marked it: J1, the 2-year closes on 4 September at least 5bp from Thursday’s close, in either direction, per the Treasury file. If it fails — if Friday’s payrolls also leave the 2-year flat — then the market has stopped listening to jobs data ahead of the 16 September meeting, and this guide will need a sixth check.

This guide extends How to Read the Jobs Report (on what is inside the BLS release itself) and Tuesday’s piece on hike odds; the Treasury-file mechanics are shared with Field Guide #39. All forty guides are indexed in the Reading Room.

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 and monthly returns, Binance BTCUSDT spot and perpetual for intraday and derivatives, Binance COIN-M quarterly contracts for basis, mempool.space for difficulty, hashrate and fees, alternative.me for the Fear & Greed series, Farside Investors’ flow table for ETF flows, and the US Treasury’s daily nominal and real par-yield files for rates. Equity, commodity and dollar closes are from Yahoo Finance’s chart feed and are third-party quotes. 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, every windowed pull carries its row count, and every figure carried forward from a previous article is recomputed from raw inputs rather than copied.

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. Do your own research and consult a licensed financial advisor before making investment decisions.