Market referenceEUR/USD
Loading EUR/USD market data…

OANDA reference price and change via TradingView. It may differ from the FOREX.com bid and ask; this is not an executable quote.

August 2026 employment situation · release September 4

Signals, measured.
Predictions, accountable.

A public research notebook for labor data and market signals. Every forecast is timestamped, ranged, and scored after the outcome.

EARLY forecastFROZEN AUG 28
+69K

Nonfarm payroll change

80% range−111K
+218K
Unemployment4.1%80% range 3.8–4.3%
Hourly earnings+0.22%Month over month · range +0.02–0.42%
Earnings, annual+3.37%Year over year · range +3.04–3.79%
Prior revisions−29KPrior two months · range −130K to +84K

J0.1-RT baseline · frozen August 28 from initially published BLS values under a rule registered before evaluation. The timestamped snapshot and ranges are now immutable. This is not a EUR/USD direction or trade recommendation.

EARLY forecast frozen · August 28, 2026 · 4:02 p.m. CT

01 / FROZEN EARLY FORECAST RANGE

The first forecast is locked.

The shaded band is the registered 80% interval. The point, range, model version, and information cutoff are frozen and will be scored after the September 4 release.

02 / LATEST EVIDENCE · AUGUST 28

The evidence changed. The rule did not.

Official calibration and regional services data leaned softer, while large-strike impact was zero. No new eligible Employment Situation release occurred, so the registered rule froze +69K without a discretionary rewrite.

OFFICIAL CALIBRATION−79K benchmark

BLS’s preliminary March benchmark showed total nonfarm employment 79K lower and total private employment 178K lower than the survey estimate.

Calibration context only; monthly CES history has not been rewritten.
REGIONAL LABORServices −3 · jobs −4

Kansas City services fell from 14 to −3 and its employment index from 4 to −4. BLS counted zero workers on qualifying large strikes in the reference pay period.

Regional diffusion and strike context are not national payroll counts.
LABOR & FORMATION55 events tracked

Michigan added a 15-position Conduent notice effective after the reference period. July recorded 578,926 applications and 151,857 high-propensity applications.

Notices, applications, and projected formations are not realized net payrolls.
MARKETS & RATESRates moved higher

August 28 Treasury yields closed at 4.34% for 2 years and 4.73% for 10 years. The latest controlled ETF session remained August 27: SPY +0.66%, QQQ +1.37%, IWM +0.29%, TLT −0.20%.

No August 28 ETF close was invented when the controlled vendor had not posted it.
03 / FIVE-DAY PAPER FORECAST · SCORED

The first directions missed.

Forecast window · August 17–21, 2026

SW0.1’s two published bullish calls were wrong. Both actual returns remained inside their wide registered 80% ranges. The forecasts are shown without rewriting them.

SPYDIRECTION MISS
−1.37%Actual return · forecast +0.55%80% range −1.53% to +4.23% · interval covered
QQQDIRECTION MISS
−2.41%Actual return · forecast +0.86%80% range −2.94% to +5.70% · interval covered
WITHHELDIWM and TLT failed their publication gates and were not scored as issued calls.

One paper window cannot establish a profitable edge. These misses count against the model.

04 / EUR/USD JOBS-REPORT STUDY

Testing the move before the release.

Development and demo only

The primary hypothesis is a pre-release EUR/USD position chosen only from information available by 8:25 a.m. ET, then closed at 8:35 a.m. ET.

01One primary window8:25 → 8:35 ET
02Executable sidesLong ask→bid · short bid→ask
03Stress testsNormal, doubled, and severe costs

FXCM minute archives contain 119 of 127 required event windows, with eight explicit gaps. That supports coarse development—not final execution proof. An Integral/TrueFX eligibility and data-rights inquiry was submitted August 24; access remains blocked until a written response arrives.

Research and demo testing do not authorize a funded September 4 trade.

LONGER-TERM TESTS
S0.1 REJECTEDReturnSharpeMax drawdown25.7%1.39−10.5%It beat SPY’s raw return but lost on risk-adjusted return and drawdown. Costs included.
S0.2 REJECTEDReturnSharpeMax drawdown15.3%1.08−9.9%Volatility scaling reduced exposure, but the model still lost to SPY on every publication gate.
01SPY

U.S. large cap

Benchmark

RESEARCHING
02QQQ

Nasdaq 100

Growth signal

RESEARCHING
03IWM

U.S. small cap

Economic sensitivity

RESEARCHING
04TLT

Long Treasuries

Rate signal

RESEARCHING
05 / METHOD

A forecast cannot move after the fact.

01

Collect

Official market, labor, formation, and layoff data are saved with retrieval timestamps.

02

Test

Candidate models must beat simple baselines before they can become operational forecasts.

03

Register

Every estimate, range, model version, and information cutoff is frozen before release.

04

Score

Outcomes and errors are appended publicly. Losing forecasts are never rewritten.

06 / EVIDENCE BASE

Five controlled source families.

BLS
Jobs & wagesFRED
Macro & ratesCENSUS
Formation & housingWARN
Labor disruptionsMASSIVE
Market prices
Next checkpoints
AUG 28EARLY forecast frozen
SEP 03Final forecast freeze
SEP 04Outcome and scoring