Gold ripped +5.03% on soft ADP. Real yields cracked.

Catalyst check. Wednesday August 5. ADP Employment at 8:15 AM ET: +95k, materially below the +150k consensus. First materially-below-consensus ADP of the current cycle. Weekly MBA Mortgage Applications early morning: +2.3 percent (unremarkable). Post-FOMC Fed communications continued to align with the dovish press-conference read from July 29. Gold ripped +5.03 percent from $4,073 to $4,278, one of the largest single-session gold gains of the year. All dates verified against BLS, MBA, and Fed calendars.

The tape

Wednesday delivered a decisive Fed-dovish repricing session, driven almost entirely by the soft ADP print. The move showed up most decisively in gold; other cross-asset moves were more modest.

• Gold: $4,278, up $205 (+5.03 percent) from Tuesday's $4,073. Largest single-session gold gain since April 2024. Intraday high $4,281, close near the high.
• DXY: 99.65, down 22 pips from Tuesday's 99.87. Modest softening.
• 10-year yield: 4.6195 percent, down 0.6bp from Tuesday's 4.6253. Essentially flat.
• Brent CFD spot: $78.93, up $0.58 (+0.74 percent) from Tuesday's $78.35. Marginal recovery.
• USD/JPY: 157.61, down 10 pips from 157.71.
• EUR/USD: 1.1557, up 25 pips from 1.1532.
• GBP/USD: 1.3469, up 18 pips from 1.3451.

Why gold moved and everything else didn't

The gold move relative to the other cross-asset responses is the specific story of the day. A soft ADP print typically moves DXY 30-50 pips, 10Y 3-6bp, and gold 0.5-1.5 percent. Wednesday produced DXY -22 pips (in-line), 10Y flat, but gold +5 percent (~3-10x the expected magnitude). This decoupling is the identifying signature of a real-yield-driven gold move rather than a broader dollar-driven move.

The mechanism: nominal yields held flat while breakeven inflation rose materially on the market's read that softer labor data reduces the Fed's ability to hold rates for long. Breakevens rising while nominals hold produces a real-yield decline, which mechanically supports gold. Under the gold real-yield framework, this is the exact configuration where gold moves outsized versus other rate-sensitive assets.

The Fed reaction function read

Soft ADP is one datapoint but a meaningful one. Combined with Tuesday's soft JOLTS (7.42M vs 7.55M) and the July 29 FOMC's dovish Q&A tone, the labor-market data-flow is now trending decisively toward supporting a cut rather than a hold. CME FedWatch September cut probability rose from approximately 15 percent Tuesday to 28 percent Wednesday; September hike probability dropped to near zero.

For the Fed reaction function under Chair Warsh, the specific signal is that the "prices too high" framing from the prepared statement is losing empirical support as inflation moderates and labor softens simultaneously. Warsh's own dovish Q&A becomes the more credible signal; the hawkish prepared statement looks more like negotiated language than policy commitment.

Setup update

Working thesis reweighted. Dovish hold at 45 percent (up from 40). Persistent-split at 25 percent. Hawkish-tilt drops to 10 percent (from 15). Language-following at 15 percent. Post-meeting Fed communications aligning with the dovish read is exactly the resolution that the blackout-window framework anticipated as the strongest signal about internal committee dispersion.

Confirmed if: Friday's NFP prints materially below consensus (below 100k or negative). 10Y holds 4.55-4.65. Gold holds above $4,200. DXY sub-100 durably.

Invalidated if: NFP surprises firm (above 200k) despite the soft ADP. Post-meeting Fed hawks (Bostic, Kashkari) deliver contradictory speeches. Gold gives back most of Wednesday's gain within 2 sessions.

Watch tomorrow: Weekly Initial Jobless Claims at 8:30 AM ET and Weekly Continuing Claims. Both are the highest-frequency labor-market signal available and will provide the second-order read on whether the ADP softness is a genuine labor-market crack or a single-week noise print.

Nothing on this site is investment advice or a recommendation to trade. Setups published here are falsifiable hypotheses, not signals.