House Delta — 2026-07-22

desk: house-delta date: 2026-07-22 forecasts: delta: p_trade: 0.15 direction: long conviction: low

Desk Plan — House Delta — 2026-07-22

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-22-pm.md PM: DELTA

Shared Market Read

• Event risk today: MEDIUM — GOOGL and TSLA report after close. The macro calendar is clean (MBA Mortgages 7 AM, State JOLTS 10 AM — both LOW impact). FOMC blackout period continues (July 18–29). EIA crude oil inventories at 10:30 AM ET. The day's real risk is binary post-close earnings; the session itself is a pre-positioning consolidation window.
• Session character expected: Consolidation / pre-earnings positioning — low conviction intraday. The chip bounce that drove yesterday's recovery (+0.25% SPY, NVDA +2.39%) extends into a second day in Asia but the Nikkei stalled (-0.25%). ES futures are modestly lower (-0.14%), Nasdaq futures -0.53%. The session is a holding pattern for tonight's mega-cap prints. The range is likely below ATR for both SPY and QQQ. IBM's -22% pre-market gap-down on a revenue miss is a negative cross-current but isolated to that name.
• VIX regime: LOW (17.44, +0.39 vs. prior close 17.05). Still in the optimal band for DELTA (dev PF 2.59 in LOW VIX). The tick-up from 17.05 to 17.44 is minor but directionally notable — the fear premium from the chip rout has fully unwound; the VIX is at multi-week lows.
• Key levels: NVDA and AMD first-hour return (09:30→09:55 close, 5-min bars) ±1.5% is the binding gate. SPY 748.15 (above SMA20 744.94), QQQ 708.78 (below SMA20 714.58, 6th consecutive day). The chip bounce is pausing — SOXX momentum stalled in Asia. The IBM gap-down has no direct semiconductor read-through but is a caution flag for earnings season volatility.
• Macro backdrop: Earnings season is the dominant force. Today pivots to mega-cap tech — the week's heaviest earnings calendar. The oil spike (+2.78% Brent to $93.54) adds a sector-level cross-current (energy up, tech consolidation) but does not directly affect the semiconductor supply-chain pairs. The yen is at a 40-year low, boosting Japanese exporters (Advantest, Tokyo Electron) but raising FX intervention risk.

DELTA — Lead-Lag Pairs Trader (NVDA→ASML, AMD→TSM, LONG-only, Phase 3)

Setup: DELTA requires a leader (NVDA or AMD) to move ≥ ±1.5% in the first 30 minutes (09:30→09:55 close, 5-min bars). If triggered, DELTA enters the laggard at 10:00 ET in the same direction. Exit: hard flat 15:30 ET or 2.0× ATR(14, 5-min) stop. Single position only — if both pairs fire, the stronger |leader return| wins.

Phase 3 constraint: LONG-only. SHORT signals are computed and logged but not traded. TEMPER rejected short re-enablement on 2026-07-15 — the 30 LONG paper trades at PF ≥ 1.5 gate stands.

Current track record: 4 paper trades (1W/3L), net -$350.78, 25% WR. SPRT: CONTINUE (LLR -0.458, n=4). The 3 losses are all on AMD→TSM (0W/3L). The only win is on NVDA→ASML (7/10). The sub-book kill switch (AMD→TSM) has not triggered but is the binding Phase 3 concern.

Today's call — p_trade 0.15, LONG direction, low conviction:

No trade expected — pre-earnings consolidation character is structurally incompatible with a ±1.5% first-hour move today.

The case against (binding):

1. Pre-earnings consolidation days are structurally the worst backdrop for DELTA. DELTA's edge requires a sharp, conviction-driven first-hour move in a semiconductor leader — the kind that happens on trending days with a clear catalyst or a sharp reversal. Pre-earnings positioning days are low-volatility, range-bound, and lack directional conviction. The session character is explicitly "consolidation / pre-positioning" with below-ATR ranges expected. This is the wrong structural backdrop for a ±1.5% first-hour impulse.
2. Yesterday's chip bounce did not produce a qualifying move, and today is less favorable. Yesterday was the most favorable session for LONG signals in over a week — the chip bounce was the dominant narrative, Asian markets surged, and pre-market chip movers were up 5-6%. Despite that, NVDA's first-hour return was -0.99% (missed ±1.5% by 0.51pp) and AMD was -0.04% (missed by 1.46pp). The bounce was real but not violent enough to trigger DELTA's threshold. Today's backdrop is strictly weaker: the Nikkei stalled (-0.25%), ES futures are negative (-0.14%), and the pre-earnings consolidation character means less directional conviction, not more.
3. Nasdaq futures are -0.53% pre-market, suggesting a flat-to-slightly-negative open with no gap-up momentum. DELTA's LONG signals are most likely when the leader opens flat-to-modestly-up and then rallies hard in the first hour. A flat-to-negative open provides no momentum foundation for the first-hour push. The pre-market data does not support a +1.5% first-hour rally in NVDA or AMD.
4. The VIX ticked up (+0.39 to 17.44) — still in LOW regime (optimal), but the direction is adverse. The VIX trajectory from yesterday's low of 17.05 is heading up, not down. A rising VIX on a consolidation day suggests the market is cautious, not confident — the wrong backdrop for a sharp semiconductor rally.
5. AMD→TSM sub-book is 0W/3L. Even if an AMD signal fires (structurally less likely than NVDA on a pre-earnings day), the sub-book track record is a binding concern. The pre-registered sub-book kill switch has not triggered, but every AMD→TSM signal today carries the statistical weight of three consecutive losses. The signal would need to be evaluated for TSM's first 15-minute price action confirming the direction before entry — a higher quality bar than the strategy standard.

The narrow path to a signal (p_trade 0.15 — possible but unlikely):

• NVDA opens flat-to-slightly-down, then rallies hard in the first 30 minutes on a short-covering or pre-earnings positioning impulse (unlikely on a consolidation day)
• The chip bounce from yesterday carries forward into a second day of buying, overcoming the Nikkei stall and the negative ES futures (requires a catalyst that doesn't exist today)
• VIX holds below 18 and ticks down, not up, during the first hour (the current trajectory is adverse)
• NVIDIA's first-hour return clears +1.5% from the 09:30 open to the 09:55 close (requires a sharp intraday reversal that is structurally improbable on a consolidation day)

If a signal fires (unlikely): Standard sizing (0.75%). If the signal is on NVDA→ASML (the stronger sub-book, 1W/0L), standard entry. If the signal is on AMD→TSM (0W/3L), evaluate TSM's first 15-minute price action before entering — the sub-book track record demands a higher quality bar. The 15:30 hard flat exit protects against any post-entry drift ahead of GOOGL/TSLA after the close.

If no signal fires (most likely, ~85%): No-trade. The pre-earnings consolidation character is structurally incompatible with DELTA's ±1.5% first-hour threshold. This is a "regime, not calibration" day — the setup never came close, no fix needed, wait for a trending session with a clear catalyst.

Plan Filed

• Filed: 2026-07-22 06:50 ET
• Frontmatter forecasts complete for every active member: yes

House Delta Desk Reflection — 2026-07-22

Reflection covering: 2026-07-18 through 2026-07-22 (since prior desk reflection 2026-07-17)

What Happened

One trade taken across three sessions this period. DELTA logged the firm's only trade today — a LONG TSM at 420.73, exited at 422.85 via HARD_FLAT at 15:30 ET, net +$108.12. The only agent action in the entire firm on a pre-earnings consolidation day.

The period in three sessions:

1. 2026-07-20 (Mon) — No signal, correct stand-down. The chip rout continued from the prior week's close. SOXX opened near support at ~535, NVDA and AMD both gapped slightly lower and failed to reverse. Neither leader approached ±1.5%. The plan forecast p_trade 0.10, direction long — correct on both probability and direction (no trade to score). Brier 0.01. Clean.
2. 2026-07-21 (Tue) — SURGE trades, DELTA sits out. SPY +0.25%, the chip bounce was the dominant narrative (NVDA +2.39%). But NVDA's first-hour return was -0.99% (missed the +1.5% threshold). AMD was -0.04% (missed by 1.46pp). The bounce was real but not violent enough — consistent with the plan's assessment. Plan forecast p_trade 0.25, direction long. Correct call. Brier 0.0625. Meanwhile SURGE took a LONG trade on the session (per EOD context) — different edge family, different conditions. SPRT for DELTA remained CONTINUE.
3. 2026-07-22 (Wed) — TRADED: AMD→TSM LONG, +$108.12, the unlikeliest trade of the week. This is the event that warrants the reflection.

The Trade That Shouldn't Have Happened — Brier Analysis

Plan forecast: p_trade 0.15, direction LONG, conviction LOW. The plan laid out five structurally sound reasons against: pre-earnings consolidation character, yesterday's bounce didn't produce a signal, Nasdaq futures negative, VIX ticked up, AMD→TSM sub-book 0W/3L. The conclusion was explicit: "No trade expected — pre-earnings consolidation character is structurally incompatible with a ±1.5% first-hour move today."

What actually happened: AMD posted first-hour return ≥ +1.5%. The catalyst: SMCI surged ~20% on a Q4 gross margin guidance of roughly double prior projections. Super Micro's margin expansion is a direct positive read-through for AMD (SMCI uses AMD and NVIDIA chips in its AI servers). AMD rallied on SMCI-specific news, triggering the AMD→TSM signal. TSM entered at 10:00 ET at 420.73, rode the SMCI lift through the session, and exited at HARD_FLAT at 422.85. Stop never threatened.

Brier: 0.7225 — the worst single-session Brier score in DELTA's Phase 3 history. The calibration model assigned 15% probability, but a genuine trade fired. Direction was correct (LONG on a LONG signal).

Was this genuinely unexpected, or is the calibration model missing something?

Both, and this is the key finding.

The macro regime assessment was correct: the session was a pre-earnings consolidation with below-ATR range (SPY 0.51× ATR, QQQ 0.44× ATR), no directional conviction, a two-tier divergence. The plan's 5 reasons against were individually valid assessments of the macro backdrop.

What the plan missed is an idiosyncratic, company-specific catalyst that overrode the macro regime for a single name (AMD). SMCI's pre-market guidance was unlikely to be in the pre-market scan at plan time (06:50 ET). The 20% SMCI surge created sector-level enthusiasm that lifted AMD independently of the consolidation character. This is a genuinely hard-to-predict event — not a calibration failure in the usual sense, but a catalyst-class blind spot.

The calibration model (p_trade = f(session character, VIX regime, pre-market data, recent trend)) does not currently incorporate sector-specific idiosyncratic catalysts. SMCI margin guidance is not FOMC or CPI — it's an unscheduled, non-Tier-1 event that can nonetheless generate enough single-name momentum to trigger a pair signal. The question for calibration: should p_trade be elevated on any day where a major semiconductor-adjacent name has overnight news, regardless of the macro backdrop?

[Flag for learnings.md: Consider whether p_trade calibration should incorporate an "idiosyncratic semiconductor catalyst" modifier — when a major AI-infrastructure name (SMCI, AVGO, MRVL, or similar) has overnight news with >5% implied move, the probability of a first-hour leader move in NVDA or AMD is elevated even on consolidation-character days, because the single-name catalyst can override the session character. This is distinct from Tier-1 event risk and would need its own detection source.]

AMD→TSM Sub-Book: First Win Changes the Narrative

The trade was on the AMD→TSM pair — DELTA's secondary pair, which entered the period at 0W/3L (WR 0.0%). Today's win moves it to 1W/3L (WR 25.0%). The pair is no longer a zero-WR sub-book.

Critical context: the AMD→TSM LONG was 0W/3L because in those three losses, AMD surged 3–5% in the first hour and TSM failed to follow. The supply-chain transmission from AMD (chip designer) to TSM (foundry) is less exclusive than NVDA→ASML, and the backtest showed it — AMD→TSM combined PF 1.61 vs NVDA→ASML 1.85. The three consecutive losses in live Phase 3 were consistent with this weaker edge.

Today's trade broke the pattern. AMD's first-hour move was on an SMCI-specific catalyst, not on a general semiconductor surge. The SMCI read-through (margin expansion = AI server demand healthy = more chip orders = TSM benefits) is a more direct and specific catalyst than a broad AMD rally. The quality of the catalyst matters: a well-grounded supply-chain narrative produced a better outcome than diffuse sector momentum.

This does not rehabilitate the AMD→TSM sub-book. One win at 25.0% WR is not evidence of edge restoration. But it changes the character of the sub-book discussion: from "loss-making on both pairs" to "the secondary pair shows some sensitivity to catalyst quality."

SPRT Monitor

• Status: CONTINUE — n=5, LLR -0.079
• Prior at n=4: LLR -0.458 (2026-07-17 reflection)
• The win today has moved the LLR from -0.458 toward zero — still negative, but the gap has narrowed meaningfully
• Distance to CONSISTENT boundary (+2.944): +3.023
• Distance to DEGRADED boundary (-2.944): -2.865
• The next 5–15 trades will be the decisive sample for SPRT resolution

Cumulative Desk Status

Brier Period Summary (Jul 14–22)

Period mean Brier: 0.1261. The single 0.7225 outlier pulls the average above the absolute target of <0.10. Excluding today's idiosyncratic-catalyst-driven trade, the 6-session mean is 0.0267 — excellent calibration on the other 6 days.

Forward Outlook

• FOMC July 29: 7 days out. Blackout period continues. FedWatch hike odds at 28% (up from 11% a week ago — oil-driven). DELTA does not hard-skip FOMC day itself unless VIX ≥ 30 (currently 16.64). The FOMC day is tradeable per strategy_v2 rules.
• ECB rate decision tomorrow (Jul 23): Cross-asset event. Oil at $94.20 (one-month high). If Dollar moves on ECB, it creates a macro cross-current that can dampen individual AMD/NVDA first-hour moves — similar to the Jul 13 geopolitical session but less extreme.
• Next week earnings deluge: GOOGL/TSLA after the close tonight. MSFT, META, AMZN, AAPL next week. These hyperscaler prints determine the capex narrative — and that narrative directly affects NVDA's first-hour volatility. A strong GOOGL/TSLA reaction tomorrow sets up the most favorable LONG backdrop since Phase 3 launched.

Desk-Level Learnings

• The AMD→TSM sub-book has its first win — on a catalyst-quality basis, not a regime-change basis. The win came on a company-specific SMCI catalyst, not on a broad semiconductor rally. This should inform how we evaluate future AMD→TSM signals: the catalyst quality and specificity to the supply chain may matter more than the raw |leader return| magnitude.
• Calibration has a catalyst-class blind spot. Non-Tier-1 idiosyncratic semiconductor-adjacent news (SMCI, AVGO, MRVL, INTC company-specific catalysts) can generate first-hour moves even on consolidation-character days. The p_trade model does not currently scan for these. If we continue to see trades fire on such days, p_trade should be adjusted upward when significant overnight news in a semiconductor-adjacent name is detected.
• The desk's only trade in 5 sessions was an outlier — that's the nature of low-frequency stat-arb. 1 trade in 5 sessions, at p_trade averaging 0.16 across the period, is consistent with the strategy's design fire rate. The surprise is not that a trade happened — it's that it happened on the lowest-p_trade day. The calibration error is real and worth fixing.

Filed

• Filed: 2026-07-22 18:30 ET
• Next session event risk: HIGH (ECB rate decision + post-GOOGL/TSLA reaction day + INTC and NOW earnings after close)
• SPRT status: CONTINUE (LLR -0.079)
• AMD→TSM sub-book kill switch: NOT TRIGGERED (1W/3L, WR 25.0%)