House Delta — 2026-08-13

desk: house-delta date: 2026-08-13 forecasts: delta: p_trade: 0.15 direction: long conviction: low

Desk Plan — House Delta — 2026-08-13

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-08-13-pm.md PM: DELTA

Shared Market Read

• Event risk today: MEDIUM (briefing) — PPI + Initial Jobless Claims at 8:30 AM ET, plus pre-market earnings (JD, TPR). PPI is NOT in this desk's hard-skip set (FOMC/CPI/NFP/PCE only; VIX ≥ 30). Hard-skip check CLEAR, firm-wide GO — this desk CAN trade today. Critically, the 8:30 data prints before DELTA's 09:30–09:55 leader-measurement window and 10:00 entry, so the macro fork resolves before any signal is computed — the data is a backdrop, not a live-position risk. The real event risk is the overnight AI-complex earnings reversal (CSCO -4%, COHR -3%, CBRS -16% after hours) which lands directly in this desk's semis universe, and AMAT reports after close (AI/semis-capex barometer) — that creates two-sided semis positioning all session.
• Session character expected: News-driven / two-sided / conditional. The briefing's own fork: if the AI gap-down fades and the complex finds a bid, expect the muted-range (below-ATR) character; if the AI complex keeps bleeding (CSCO/CBRS/COHR gap-down extends), expect Nasdaq-led weakness that broadens. The overnight driver is a down impulse in exactly the names DELTA's leaders (NVDA/AMD) sit in. This is a genuine two-way tape, not a clean direction.
• VIX regime: 14.60 — LOW (<18), DELTA's best backtest regime (dev PF 2.59 in VIX < 20 vs 1.30 in 20–30); the ≥ 30 hard-skip barrier is far away. The single most supportive input — and it licenses trading today.
• Key levels: The briefing carries no NVDA/ASML/AMD/TSM prices (this desk's instruments are not in its coverage). Last cached closes are 08-10 (NVDA 217.49, AMD 469.71, ASML 1,733.90, TSM 418.59); no Alpaca/IEX-verified 08-11/08-12 closes exist for these four names. Indicative spread triggers from the 08-10 cache (NOT verified-current): NVDA +1.5% ≈ 220.75 fires the clean NVDA→ASML LONG, -1.5% ≈ 214.23 is a disabled SHORT; AMD +1.5% ≈ 476.76 fires the degraded AMD→TSM LONG, -1.5% ≈ 462.67 is a disabled SHORT. The runner computes its own triggers from the actual prior close.
• Macro backdrop: benign-expected PPI (consensus +0.2% MoM / +0.3% core) argues for the quiet gap-fade pattern, but the after-hours AI-complex reversal injects a second, live risk leg — if it extends, the semis complex bleeds and the LONG path dies. The tape is two-sided between a dovish-data relief and an AI-complex down-repricing.

PRICE-VERIFICATION LIMIT (stated per dispatch): the pm briefing is Alpaca/IEX-verified but covers none of DELTA's four instruments. The 08-10 cache is stale (no 08-11/08-12 closes). I anchor spread triggers from the cache with that caveat and treat any 08-13 premarket print for NVDA/ASML/AMD/TSM as unverified — the runner measures the actual 09:30–09:55 first-hour returns, which are the only authoritative signal inputs.

DELTA — NVDA→ASML, AMD→TSM (LONG-only, Phase 3)

Forecast: p_trade 0.15 | direction long | conviction low

Verdict: No setup is the base case — this is a down-biased AI tape for a LONG-only desk, and the only live trade path is a first-hour AI bounce that clears +1.5% against the overnight gap-down. p_trade 0.15. The dominant overnight input is a down impulse in the AI complex (CSCO -4%, COHR -3%, CBRS -16% after hours) — exactly the complex NVDA and AMD lead. For a LONG-only desk, a down-gapping semis complex produces the least-favorable tape class:

• Most likely (no_signal): the complex gaps down modestly and the first-hour leader returns land sub-±1.5% in both directions — the recent 08-10 / 08-11 pattern (AMD +0.12% then -1.66% [that one cleared the SHORT bar], NVDA -0.57% then near-flat). Quiet semis means no signal, no trade.
• Second branch (short_disabled): if the AI bleed is significant, NVDA and/or AMD clear -1.5% in the first hour → a SHORT signal, which is disabled (LONG-only by standing TEMPER rule). Logged to the SHORT shadow book (n=5 → n=6), no trade. This is the most likely impulse outcome if the complex keeps bleeding — and it is untradeable.
• Only live trade (LONG): the AI complex fades the gap-down and finds a strong first-hour bid, NVDA clears +1.5% → clean NVDA→ASML LONG (NVDA_ASML 2W/0L live, dev combined PF 1.85 — the highest-quality LONG in the book). This is the briefing's explicit "AI names find a bid on the gap-down" branch, but it is the lower-probability fork on a day the overnight driver is negative.

Calibration math (day-conditional): P(signal) ≈ 0.30 — two-sided, news-driven day with real directional impulse in the complex (vs. the dead-flat 08-10/08-11), so a ±1.5% leader move is more plausible than those quiet days, but the impulse is likely negative (disabled) or sub-threshold. P(LONG | signal) ≈ 0.50 — the up path is the only tradeable one (down = disabled SHORT), and a gap-fade-into-strength is genuinely possible on a benign PPI + relief tape, but the overnight down-driver makes the up-fork a coin flip, not a lean. Net tradable-signal probability ≈ 0.15. Direction long by construction (LONG-only book). Conviction low: base case is no trade; the LONG path requires the AI complex to reverse a negative overnight impulse in under an hour.

Sizing vs. event risk: standard 0.75%, single position — no change. The 8:30 PPI/claims print lands before the measurement window and entry, so the macro fork is absorbed into the open, not carried as a live-position risk. The hard flat at 15:30 removes any overnight into AMAT's after-close print by construction. The event risk that matters is signal quality: on a two-sided AI-reversal day the lead-lag can distort (leaders gap with their laggard → compressed/zero lag → late entry), handled by scrutiny and invalidation, not a size cut. No scale-down warranted and none appropriate — a possible trade is not a risk to pre-cut.

Key levels (indicative from 08-10 cache, NOT verified-current): NVDA +1.5% ≈ 220.75 → clean NVDA→ASML LONG; NVDA -1.5% ≈ 214.23 → disabled SHORT (shadow n=6). AMD +1.5% ≈ 476.76 → degraded AMD→TSM LONG; AMD -1.5% ≈ 462.67 → disabled SHORT. Laggard cash-open gaps are the second read: ASML/TSM gapping with their leader partially closes the spread (late entry); a laggard not gapping is the only genuine lead-lag shape left. Index tone to watch: whether QQQ's soft leg (Nasdaq futures weak overnight) stabilizes or the AI complex's bleed broadens — that decides the no_signal vs. short_disabled branch.

Invalidation (stand down entirely):

• NVDA and/or AMD clear -1.5% in the first hour — that's a disabled SHORT, not a trade; the desk is closed for the day by the LONG-only constraint (the down path is structurally untradeable).
• SPY or QQQ cash-open gap > 1.5% in either direction — impulse absorbed before the measurement window (a fresh mid-session >1.5% move against the LONG direction also stands me down).
• The AI complex keeps bleeding after the open (CSCO/CBRS/COHR gap-down extends, AMAT positioning turns bearish) — confirms Nasdaq-led weakness; the LONG path is dead.
• A hot PPI print (reaccelerating producer inflation) undercuts the benign-CPI relief and flips the tape risk-off — the AI gap-down extends, LONG dies.
• QQQ turns sharply negative in the first 30 minutes while the complex follows — down-tape confirmed, no LONG.

Information cost (documented, not acted on): a down-AI day is the desk's structural blind spot — if either leader cascades -1.5% the disabled SHORT is logged to the shadow book (n=5 → n=6; n=7 if both), but LONG-only keeps costing information in exactly the down-tape the overnight narrative favors. The shadow book is the accounting, the 30-LONG gate (n=8/30) is the governing re-enablement path, and the 07-24 learnings observation (the LONG gate may be structurally unreachable in a sustained semis bear — shadow book as the alternate evidence path) stays live and is directly relevant today. The event-day diagnostic (skip-disabled run) remains open at moderate priority — PPI is not a skip, so today is a normal (non-skip) tradeable day and adds a clean no-skip datapoint either way.

Plan Filed

• Filed: 2026-08-13 04:46 PDT (pre-open, morning dispatch)
• Frontmatter forecasts complete for every active member: yes
• Hard skip check: CLEAR (PPI/Claims 8:30 AM is NOT in DELTA's skip set; VIX 14.60 < 30) — firm-wide GO, desk CAN trade. AMAT after close is not a skip; the 15:30 hard flat removes any overnight into it
• Price-verification limit stated: briefing carries no DELTA-instrument prices; 08-10 cache anchors are stale (no verified 08-11/08-12 closes) — runner measures actual first-hour returns
• Dominant input: overnight AI-complex reversal (CSCO/COHR/CBRS gap-down) — a down-tape bias for a LONG-only desk; base case no_signal or disabled SHORT, only live trade is a first-hour NVDA→ASML LONG on a gap-fade-into-strength

Reflection — House Delta (DELTA) — Daily — 2026-08-13

Period Performance

Single-member desk. DELTA traded once today — Long TSM 57 @ 434.21, stopped 432.01 (11:30, STOP_LOSS), net -$125.54. Cumulative -$307.96 across 9 paper trades. SPRT sequential monitor CONTINUE (n=9, LLR +0.120, +2.825 from the CONSISTENT-WITH-BACKTEST boundary). AMD→TSM sub-book: n=7, WR 28.6%, PF 0.226 — not triggered, but chronically degraded.

What Worked

• Event-risk handling was clean. The 8:30 PPI/claims print (MEDIUM, correctly rated) landed before DELTA's 09:30–09:55 measurement window and 10:00 entry, so the macro fork was absorbed into the open with zero live-position exposure — exactly the plan's design. Nothing on the calendar was missed or overstated; PPI printed even softer than the benign consensus.
• Direction forecast hit. The plan refused to pre-assign single polarity (good discipline) but the LONG-only book is long by construction, and it was long in the right direction — a LONG fired and the tape was genuinely risk-on.
• Sizing was right. Standard 0.75%, single position, unchanged — appropriate for a pre-absorbed macro event. This was a normal stop on a normally-sized position, not an over-bet. The 15:30 hard flat removed any overnight into AMAT's after-close print.
• The plan's own warning was prescient. DELTA named AMD→TSM "the weakest cell... taken mechanically, scrutinized by construction, not celebrated." That is exactly the box it checked today. The mechanism behaved as the plan feared it would.

What Didn't Work

• The session-character call was the wrong side of the tape. The plan wrote around a down-biased AI tape (CSCO/COHR/CBRS overnight gap-down → "Nasdaq-led weakness that broadens"). The actual day was a clean trending-up tech-led rally to a record (QQQ +1.17%, SPY +0.69%) — the AI single names bled AND the broad market rallied on soft PPI. This is the two-tier divergence resolving in the macro's favor, the opposite branch of the fork the plan built its base case around. The plan over-weighted a name-level gap-down as an index-level risk and under-weighted a benign macro print releasing the same morning.
• p_trade 0.15 was under-forecast (Brier 0.7225). A LONG fired and DELTA had it priced at 0.15. The probability mass sat on the clean NVDA→ASML cell ("the only live trade"), but the mechanical runner buys the laggard on whichever leader clears +1.5% — and on a risk-on macro day, the degraded AMD→TSM cell is the one most likely to fire. The plan treated the degraded cell as a tail; the day made it the operative path.
• The degraded AMD→TSM LONG keeps losing. Sub-book n=7, WR 28.6%, PF 0.226 — well under the rolling 1.2 threshold. This is the second consecutive loss pattern on this pair and it fired LONG on the exact day the plan centered the clean NVDA→ASML path. This is trending toward a calibration story (a chronically-degraded cell that keeps firing and keeps losing), not a regime wait.

Confidence Changes

• The lead-lag LONG mechanism itself: unchanged. This loss is a normal stop on a degraded cell — it does not impeach the NVDA→ASML primary edge (2W/0L live, dev combined PF 1.85). No confidence change in the clean cell.
• The degraded AMD→TSM cell: further conviction that it is the structural drag on the book. PF 0.226 at n=7 is not evidence; but the mechanism keeps surfacing it as the operative trade on risk-on days, and it keeps losing. This is the conversation DELTA should bring to TEMPER — not a unilateral tweak.
• Base-case probability calibration: lower trust in "no_signal" forecasts on macro-relief days. The mechanical runner will fire the weakest cell; DELTA's p_trade must price that.

New Hypotheses

Junior Agent Queue

None — DELTA is a single-member desk. Hypotheses above are flagged for TEMPER's calibration conversation, not queued for junior-agent deployment.

Notes for Next Period

• The load-bearing gate question: is the binding constraint regime (setup never came close) or calibration (a repeatedly-close, repeatedly-losing degraded cell)? The near-miss/loss data says calibration: the AMD→TSM sub-book (n=7, PF 0.226) keeps firing and losing well under threshold — evidence worth a TEMPER conversation, never a unilateral gate change.
• Weight the macro leg on data mornings. When a benign inflation print is releasing the same morning a marquee risk-name gap-down lands, do not let the name-level gap become the index base case. Soft PPI → risk-on was today's reality.
• Price the union, not one cell. p_trade must reflect that the runner fires whichever leader clears +1.5%, including the degraded AMD→TSM cell on risk-on days.
• Next session (08-14): HIGH event risk — Retail Sales (Jul) + Michigan at 8:30/10:00, WMT/ABNB pre-market, AMAT after-close sets the semis gap. Not in DELTA's hard-skip set (FOMC/CPI/NFP/PCE; VIX 14.63 < 30); reviewable but pre-absorbed by the 09:30 measurement window.

Reflection Filed

• Filed: 2026-08-13 15:45 PDT
• Next session event risk (from EOD briefing tomorrow's calendar): HIGH — Retail Sales + Michigan, WMT/ABNB pre-market, AMAT after close.