House Delta — 2026-09-22

Desk Plan — House Delta — 2026-09-22

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-09-22-pm.md PM: DELTA

Shared Market Read

• Event risk today: MEDIUM — no Tier-1. Bucket A is clear through 2026-09-29; next hard-skip for this desk is PCE Sep 30 (then NFP Oct 2, CPI Oct 14; FOMC Minutes Oct 7). Today is domestically macro-silent — the only US release is the Richmond Fed Manufacturing Index (10:00 ET, second-tier) and there is no 8:30 AM print — but two live vectors run all day: (1) a heavy hawkish Fedspeak slate (Hammack 8:50, Williams permanent voter + Barr + Paulson clustered 10:05–10:10, Barkin 1:00, Hammack again 2:00) with long-end yields already near multi-decade highs (10-yr ~4.95–4.97%, 30-yr ~5.28%); (2) the geopolitical/diplomatic clock — Trump addresses the UN General Assembly today with a possible Iran sideline, oil whipsawing two-way around $100 Brent, and the Trump–Xi summit Wed Sep 24. Firm-wide trade status: GO.
• Session character expected: mixed / two-tier, direction unknown — AI/tech leadership against a broad tape capped by rates and oil; the cleanest intraday structure in the morning window, headline-driven chop into the afternoon. Per the briefing's standing discipline (and its EOD feedback loop), I am not pre-assigning a one-day polarity on a Fedspeak/geopolitical coin-flip day: the honest call is contained, two-tier, direction unknown. The refusal is posture, not a lean.
• VIX regime: 14.81 (−0.06 vs 14.87) — LOW (<18), ~15 points from the 30 hard-skip. This is DELTA's strongest regime by the backtest (dev PF 2.59 in VIX < 20; the family-inverted dependency — high macro vol adds noise to supply-chain transmission rather than amplifying it).
• Key levels: index context (Alpaca-verified, prior close 09-21): QQQ 741.41, above SMA20 714.65, ATR(14) 9.50; SPY 773.52, above SMA20 764.65, ATR(14) 7.12 — both above their 20-day, tech-led and leadership-concentrated. Trailing ATR is still elevated off Monday's ~1.9–2.3× ATR move (QQQ +2.78%, SPY +1.56%); expect wider-than-average ranges to persist. DOCUMENTED DATA GAP — not filled: neither the briefing nor the watchlist carries a verified NVDA / ASML / AMD / TSM price level, and the watchlist explicitly flags its pre-market tables as a stale Monday (09-21) snapshot. I therefore adopt no instrument-level price for any of the four legs, invent none, and treat every semis figure I cannot reconcile to an authoritative source as directional-only. The only load-bearing number for this desk is each leader's first-hour return off the 09:30 cash open, read at 09:55, vs. ±1.5% — and that is a live measurement, not a pre-market level.
• Macro backdrop: an AI-optimism bid carried from Monday (Meta's post-Muse surge lifting the AI/chip complex; semis named among pre-market leaders) against a hawkish Fed (Warsh hiked Sep 16, more priced) with long-end yields near multi-decade highs and oil >$100 — a tug-of-war whose live risks today are Fedspeak and the UNGA/Iran clock. The asymmetry the briefing names is the one that matters here: the high-multiple AI complex is simultaneously the leadership leg AND the rate-fragile leg — and it is the exact complex this desk trades.

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

Call: setup possible — p_trade 0.25, direction long, conviction low. Not a stand-down day. The engine runs (no Tier-1, VIX 14.81 LOW, firm GO, MEDIUM event risk is not a hard skip), the pre-market shape is a tech-led gap-up with semis among the leaders, and trailing ATR is elevated after Monday's ~2× ATR session — so the mechanical odds that some leader prints ±1.5% in the distorted 09:30→09:55 window are genuinely higher than the 0.15 the desk has been anchoring on, and I am raising the number rather than repeating the anchoring error. What holds conviction to low is that the named driver is complex-wide, not idiosyncratic: the Meta Muse read-through bids the whole AI-chip complex (INTC/AMD/AVGO/ARM/ALAB named), and complex-wide beta reprices leader and laggard on one systematic factor — by construction not the design→manufacturer cascade my edge samples. direction long is not a day-polarity call; it is simply the only live branch (SHORT is disabled by rule).

Setup present: possible, but conditional on an idiosyncratic clear. My signal is not a pre-market level — it is the leader's first-hour return off the 09:30 cash open, read at 09:55, vs. ±1.5%: NVDA ≥ +1.5% → buy ASML at the 10:00 bar open (primary, edge-carrying cell, live 2W/0L, dormant since 08-03) and AMD ≥ +1.5% → buy TSM at the 10:00 bar open (secondary; n=9, WR 33.3%, PF 0.62 — returned to profit 09-21 with +$375.06, its first win since 08-06, on AMD's +3.69% first-hour $1T-milestone rally). Leader ≤ −1.5% = the disabled-SHORT branch (ENABLE_SHORT = False) → no trade by rule, not a missed LONG. Why this is live rather than blocked: Monday settled the desk's five-week regime question — the floor was reachable and profitable when a leader-specific move finally arrived, so the prior up-approaches were a live-setup warm-up, not a grazed threshold. Why it may still not fire: the last five weeks' blocker (complex-wide bid denying an idiosyncratic impulse) is exactly the shape today's narrative describes, and a leader-clear on complex beta is the trade the desk's own reading does not want.

Sizing vs. event risk: standard 0.75% single-position risk, unchanged — no parameter adjusted. Event risk MEDIUM does not trigger a hard skip; my hard 15:30 flat + 2.0× ATR(14, 5-min) stop carry zero overnight and zero event exposure by rule. One honest, pre-registered exposure: entry at the 10:00 bar open sits immediately ahead of the 10:05–10:10 Williams/Barr/Paulson cluster — a qualifying LONG is therefore pre-positioned into a known hawkish-headline window, and the AI complex is the most rate-fragile leg. That is a reason for low conviction, not a sizing change and not a stand-down: the ruleset has no event buffer here, and I am not inventing one unilaterally.

Session-character fit: mixed — the best-fitting window of a poorly-fitting day. The briefing expects the cleanest structure in the morning, and my read completes at 09:55 before the Fedspeak cluster — but the afternoon is headline-driven chop the desk does not sample. A gap-up tech-led open after a ~2× ATR day is also a give-back-prone tape (extensions overshoot and snap back), which is hostile to a first-hour impulse continuing through the 15:30 flat. Net: the session does not block the mechanism the way a rebalance-open or a quad-witching open did; it simply does not help it.

Key levels (mine only): the only load-bearing number is each leader's first-hour return off the 09:30 cash open, read at 09:55, vs. ±1.5% (NVDA→ASML primary; AMD→TSM secondary). No instrument-level price is adopted for NVDA, ASML, AMD or TSM — the briefing and watchlist carry none at a verified level and the watchlist's pre-market tables are a flagged stale 09-21 snapshot, so all four pair inputs require live confirmation at the cash open. This is a stated data gap, not a level I have decided to skip.

Invalidation / no-trade reasons: (1) leaders opening broadly down ≥1.5% = the disabled-SHORT side (ENABLE_SHORT = False) → no trade by rule; (2) any intraday VIX ≥ 30 closes consideration; (3) a hawkish Fedspeak read at 10:05–10:10 that spikes yields and strips the AI complex of the open's bid — that is a quality read on any open impulse, not a parameter change; (4) hard skip: PCE Sep 30 (then NFP Oct 2, CPI Oct 14). No additional stand-down is pre-registered. Note the distinction the desk has held all cycle: a pipeline absence (runner starvation) is a REVIEW state, not a no-trade, and will never be narrated as a strategy stand-down.

Book state (carries forward): cumulative −$67.27 across 11 LONG trades, 5W/6L, WR 45.5%, gross PF 0.923 (SPRT n=11, LLR +0.219, CONTINUE — first positive LLR movement of the cycle, +2.725 from CONSISTENT-WITH-BACKTEST / −3.164 from DEGRADED). NVDA→ASML 2W/0L remains the quality cell. AMD→TSM n=9, WR 33.3%, PF 0.62 — first win since 08-06; the 09-05 stop-calibration diagnostic read NO (widening the stop does not rescue the cell: PF 0.073 → 0.119 across 2.0–4.0×), pending TEMPER cadence-ratification; a signal/regime watch, not a parameter fix. Rolling 6-month PF monitor FIRED 09-21 (0.923 < 1.2 — TEMPER binding condition #2, first trigger), driver concentrated in the AMD→TSM cell: escalated to TEMPER as a calibration conversation, not a unilateral gate tweak. The p_trade under-forecast series (09-21 = 4th instance: 0.15 forecast → +$375.06 win → Brier 0.7225) is the reason today's number is raised to 0.25.

✔ Operational item — desk-top priority, carried forward: the per-agent 60s run-trades watchdog. Fifth flag of the streak landed 09-21 (after 09-09/09-10, 09-15, 09-18); the orchestrator backfill recovered the observation (+$375.06 is in trades.csv and the daily log), but the 18:00 EOD desk summary still read DELTA as no-performance-file. The ledger-freshness post-run check is still not wired. My forecast is scored against trades.csv — if the watchdog trips again today, no row is produced and my 0.25 is scored against a pipeline absence rather than a tape outcome. A missing observation is never evidence of a near-miss. Standing asks, both multi-instance: backfill remaining standalone holes; wire in the freshness check.

Watch items (informational only — not entry signals): (1) whether a leader decouples from the complex-wide bid and clears the floor on a genuinely idiosyncratic move — on a gap-up tech-led open that distinction is the entire read; (2) the 10:05–10:10 Fedspeak cluster and whether a hawkish read presses yields and the AI leg, plus each UNGA/oil headline as a two-way tape-mover; (3) whether the AMD→TSM cell's return to profit is N=1 or the start of sub-book recovery — the desk's live rolling-PF question.

Plan Filed

• Filed: 2026-09-22 04:55 PDT (pre-open)
• Frontmatter forecasts complete for every active member: yes (delta — p_trade 0.25, direction long, conviction low)
• Briefing event risk confirmed: MEDIUM — no Tier-1 (Bucket A clear through 2026-09-29); Richmond Fed Mfg 10:00 is the only US release; live vectors are the hawkish Fedspeak slate and the UNGA/Iran/oil clock; VIX 14.81 LOW; firm-wide trade status GO; normal Tuesday session
• Verdict: Setup possible — not a stand-down day. p_trade 0.25, direction long, conviction low. The engine runs and the pre-market shape (tech-led gap-up, semis among leaders, elevated trailing ATR after a ~2× ATR Monday, LOW VIX = best backtest regime) genuinely raises the odds a leader clears ±1.5% in the first hour vs. the desk's recent 0.15 anchoring; conviction stays low because the named driver is complex-wide beta, not the idiosyncratic design→manufacturer cascade the edge requires, and a 10:00 entry sits immediately ahead of the 10:05–10:10 Fedspeak cluster. Standard 0.75% single-position sizing, no parameter adjusted; binding gate = leader first-hour return read at 09:55 vs. ±1.5%. Documented data gap: no verified NVDA/ASML/AMD/TSM level in either firm input (watchlist pre-market tables flagged stale) — no instrument-level price adopted, all four legs confirmed live at the cash open.

House Delta — Desk Reflection — 2026-09-22

Cadence rule: Weekly/Friday or same-day-on-trade (WS5 per-desk cadence). Today is a Tuesday, but DELTA logged a real trade, so the desk reflects same-day. Period covered: 2026-09-21 (last desk reflection) → 2026-09-22. Scheduled session in window: 09-22 (Tue).

Single-member desk — the desk reflection is DELTA's own analysis at desk scope; the per-member compatibility file (agents/delta/reflections/daily_2026-09-22.md) is written separately and remains first-person.

Desk Performance Summary

Desk total for the period: +$226.88 — a second consecutive winning trade, and the second consecutive session where the LONG up-floor cleared.

Period session map:

Cumulative moved −$67.27 → +$159.61 (net of commission, n=12, WR 50.0%, gross PF 1.183) — the book crossed back above zero for the first time in Phase 3. SPRT LLR +0.219 → +0.597 (CONTINUE, +2.347 from the CONSISTENT-WITH-BACKTEST boundary, −3.542 from DEGRADED). AMD→TSM sub-book: n=10, WR 40.0%, PF 0.88 (up from n=9 / 33.3% / 0.62 on the back of two consecutive wins). NVDA→ASML still clean at 2W/0L (+$264.42).

Per-Member Binding Gate Analysis (delta — the whole load)

Binding gate this period, part 1: the LONG up-branch (leader first-hour return ≥ +1.5% off the 09:30 cash open, read at 09:55) — it cleared for the second consecutive session and the trade won again. REGIME, not calibration; no gate change.

1. Two consecutive decisive clears are now on the board. After five weeks of near-approaches that never got there, the floor has been cleared twice in two sessions: AMD +3.69% on 09-21 (2.19pp of room) and AMD +1.86% today (0.36pp of room). The desk's standing "regime, wait — the setup is live and reachable, the sector is simply not producing idiosyncratic leader moves often enough" reading is now backed by outcomes, not just by argument. Nothing in the near-miss record suggests a threshold being systematically grazed from below; the two clearings bracket the width range and both printed profits.
2. But be precise about the width: today's clear is the thinnest of the cycle. +1.86% is 0.36pp of room — materially different in kind from Monday's +3.69%. It is a clear, not a graze (a graze is an approach that fails), and the mechanism behaved consistently with its own thesis: a smaller leader impulse produced a smaller laggard follow-through (+$226.88 vs +$375.06, HARD_FLAT both times). I am recording the width because the reachability of the floor is now confirmed twice while its magnitude distribution is not a stable property — if the next several clears all come in at +1.6–1.9%, that is a different book from one that clears at +3%+ and it should be tracked as a shape question, not assumed away.
3. No calibration conversation is warranted on the up-floor. A floor that has cleared twice, at two different widths, both profitable, is doing its job. The 09-04 "watch-to-escalate" flag on NVDA up-grazes remains structurally moot.

Binding gate this period, part 2 — and the load-bearing finding: the rolling 6-month PF monitor fired again (PF 1.183 < 1.2 on 12 trades), but TEMPER binding condition #2's own escalation wording is NOT satisfied by this reading, and cannot currently be satisfied by any reading.

1. What the condition actually says. delta_v6b_2026-06-24.md, Phase 3 Caveat §2 (P12): "Track rolling 6-month PF. If rolling PF drops below 1.2 for two consecutive 6-month windows, escalate to formal strategy review." The escalation trigger is two consecutive windows, not one sub-1.2 print.
2. What the implementation can and cannot see. paper_trade.py's rolling_pf_check() takes a trailing-180-day cutoff from the latest trade date, requires ≥ 10 trades in the window, and prints an alert whenever PF < 1.2. It carries no window state, no window index, and no consecutiveness counter — it cannot distinguish a first window from a second consecutive one, and it will print the identical alert every session the window reads below 1.2. This is not a hypothetical: the current window's cutoff is 2026-03-26, and all twelve Phase 3 trades (2026-07-06 → 2026-09-22) fall inside it. The book is 2.8 months old. No trade has ever rolled off a 6-month window. So 09-21's 0.923 and today's 1.183 are two readings of the same window, not two consecutive windows.
3. Honest verdict on the trigger: the desk's 09-21 framing ("first trigger") and any today-based "second trigger" framing cannot be substantiated from the implementation, and at this book age "two consecutive 6-month windows" is undecidable rather than merely unmet. Two things follow, both desk-blind and both proposals, never unilateral changes: (a) the escalation semantics are a monitoring-instrument question for TEMPER — either the monitor needs window-boundary state (alert on a fresh window / count consecutive populated windows that each breach), or the condition should be restated in readings rather than windows if readings were the intent; (b) absent that state, a rolling-180-day PF that contains the entire book is not a "rolling" signal at all — it is cumulative PF under another name, and it should not be read as stronger evidence than the monitor that was pre-registered for exactly this question.
4. What the pre-registered arbiter says: CONTINUE, and the evidence moved the right way. SPRT (n=12, W6/L6, LLR +0.597) is 2.347 below the CONSISTENT-WITH-BACKTEST boundary and 3.542 above DEGRADED. The rolling PF itself rose 0.923 → 1.183, entirely on the two consecutive AMD→TSM wins; the degraded cell improved 0.62 → 0.88 (n=9 → n=10, WR 33.3% → 40.0%). Two consecutive readings either side of a threshold that sits 0.017 away from the current value means the alert's on/off state is a single-trade coin-flip right now. That is precisely why the trigger needs window semantics rather than daily readings — and it is why I read today's alert as calibration-shaped in instrument terms, with the underlying evidence still on the regime/wait side, not as confirmation of degradation.
5. No rescue is being declared on AMD→TSM. N=2 wins is not a sub-book recovery; PF 0.88 over n=10 is still sub-1.0, and the 09-05 stop-calibration diagnostic (read NO — widening the stop does not rescue the cell) and the pending cadence-ratification via the 2026-09 Desk Review both stand unchanged. The right statement is the narrow one: the cell has produced two live wins in two sessions after a nine-trade stretch that included a three-loss run, and its life-to-date PF is still below 1.

Forecast calibration — 5th instance of the one-sided under-forecast series, but the smallest miss of it.

1. p_trade 0.25 → traded, direction hit, Brier 0.5625. The desk raised its number from 0.15 (the anchoring value it had repeated through the late-August/September dormancy) to 0.25, explicitly because the 09-21 under-forecast was the fourth of the series and because the pre-market shape (tech-led semis, elevated trailing ATR, LOW VIX — the strategy's best backtest regime) genuinely raised the odds. The engine fired anyway. So this is the 5th documented instance (07-22, 08-13, 08-03, 09-21, 09-22) — but it is the smallest miss (Brier 0.5625 vs 0.7225 on the three worst), the adjustment moved the right direction, and a 0.25 on a +1.86% thin clear was a defensible number rather than an anchor error. The candor point is the one the series has always carried: on no-Tier-1 / LOW-VIX / firm-GO days the realized fire rate over the last two sessions is 2/2, and my stated probability is still running below that. This is a desk-owned p_trade-calibration conversation for TEMPER, not a gate change.
2. Session-character and event-risk calls were both right this time. The desk called MEDIUM — no Tier-1, with the two live vectors named as the Fedspeak cluster and the UNGA/Iran-oil clock, and explicitly refused to pre-assign polarity ("contained, two-tier, direction unknown"). The session delivered exactly that: QQQ +0.81% into a record Nasdaq close, SPY −0.01%, Dow −0.36%, both legs sub-ATR (SPY 0.37×, QQQ 0.80×), VIX down to 14.21; the Fedspeak slate produced no hawkish shock (Barkin read as a bounded/mid-cycle path; yields contained, 10-yr ~4.96–4.97%); and oil resolved down (Brent back below $100) on Hormuz diplomacy. The one pre-registered exposure the plan named — a 10:00 entry sitting immediately ahead of the 10:05–10:10 Williams/Barr/Paulson cluster — did not bind, and the decision not to change sizing for it was correct in hindsight, exactly as framed (a quality risk, not a stand-down).

Operational — the desk's top standing item recurred for the sixth time, and the plan pre-registered the risk this morning.

1. The per-agent 60s run-trades watchdog starved DELTA again at today's 1:30 PM job (delta failed or exceeded 60s — skipped), runner health REVIEW, no daily artifact written by the job. The orchestrator's direct backfill (python agents/delta/paper_trade.py 2026-09-22, same script, same ASSAY shared cache) recovered the session faithfully — the +$226.88 is a genuine simulation result, not a substitute, and there is no observable downstream difference in the data. But the 18:00 EOD desk summary again read DELTA as "No performance file (no trade)" despite performance/daily_2026-09-22.md existing. This is the 6th flag of the streak (09-09, 09-10, 09-15, 09-17, 09-21, 09-22), and it is worth stating plainly what it costs: my forecast is scored against trades.csv, so on a tripped day the 0.25 would have been scored against a pipeline absence rather than a tape outcome — a missing observation is not evidence of a near-miss. The backfill is the mitigation; the ledger-freshness / stale-output post-run check remains unwired, and the EOD line was still wrong today with the artifact present.
2. One new, small instrumentation gap found today (found while writing this reflection, not before): the near-miss logger populates its per-pair blocking notes only on no-trade sessions (log_delta_no_trade). On a traded session the row's notes carry only strategy=v6b, so the non-firing pair's first-hour value is not recorded. Concretely: I cannot tell whether NVDA approached ±1.5% today or was nowhere near it — the primary edge-carrying cell is the one whose proximity I cannot measure on exactly the days the desk is live. That is a one-line logging fix (always emit nm_by_pair when a trade is taken), and until it exists the desk should read "no signal on the other pair" as unmeasured, not as distant.

Implication — no action on the gate stack; the rolling-PF escalation wording/state is a TEMPER instrument question; keep the AMD→TSM watch open; fix the freshness check and the traded-day near-miss logging. The strategy cleared its floor twice in two sessions and both trades were profitable. Nothing here justifies a unilateral parameter change, and nothing here yet justifies declaring the degraded cell repaired.

Open TEMPER Items (Carried Forward)

Reconnaissance Ledger Items

• Watchdog root fix + ledger-freshness post-run check — still the desk's top operational priority at 6 flags. Today is the sharper case: the artifact existed and the 18:00 summary was still wrong, so the check must validate the ledger, not merely the file.
• TEMPER conversation — rolling-PF trigger semantics + window state. New this period and the desk's substantive escalation: the binding condition cannot fire distinctly at the current book age, and a trailing window containing the whole book is cumulative PF under another name. Pair with the pending 09-05 stop-calibration ratification.
• TEMPER conversation — p_trade calibration series (5th instance).
• Traded-day near-miss logging — emit nm_by_pair on trade rows.
• Event-day diagnostic (skip-disabled) remains a standing obligation.

Desk Summary for Orchestrator

House Delta: the LONG up-floor cleared for the second consecutive session and the desk won again. AMD first-hour +1.86% (> +1.5% floor by 0.36pp — the thinnest clear of the cycle) → TSM LONG 55 @ 446.94, HARD_FLAT 451.06 @15:30, +$226.88 WIN. Cumulative −$67.27 → +$159.61 (n=12, WR 50.0%, gross PF 1.183 — first time above zero in Phase 3); SPRT LLR +0.219 → +0.597 (CONTINUE). Binding gate: the LONG up-floor — REGIME, not calibration, now confirmed by two clearings at two widths; no gate change. The load-bearing escalation this period is the rolling 6-month PF monitor (PF 1.183 < 1.2), and the honest finding is that TEMPER binding condition #2's own wording — "two consecutive 6-month windows" — is undecidable at this book age: the trailing-180-day cutoff (2026-03-26) contains all 12 Phase 3 trades, so yesterday's 0.923 and today's 1.183 are the same window, and rolling_pf_check() carries no window state. Proposed to TEMPER (never applied): add window-boundary/consecutiveness state to the monitor, or restate the condition in readings — the pre-registered SPRT arbiter says CONTINUE. AMD→TSM improved to n=10 / PF 0.88 on two consecutive wins but is not declared repaired. One thing watched: the per-agent 60s watchdog starved DELTA for the 6th time (backfill again recovered a genuine +$226.88), and the 18:00 EOD summary again read DELTA as no-performance-file with the artifact present — the ledger-freshness check is still unwired.

Reflection Filed

• Filed: 2026-09-22 EOD (desk cadence — same-day-on-trade trigger)
• Next session (Wed 09-23): event risk LOW by calendar — no Tier-1 through 2026-09-29; next hard-skip PCE Sep 30 (then NFP Oct 2, CPI Oct 14; FOMC Minutes Oct 7). The week's live two-way is the US–China summit (~09-24) on trade/AI-export-controls/rare earths — date not verified: the 09-22 EOD notes the wire called it "Thursday" while prior briefings carried Wed Sep 24, so it is a watch, not a gate, pending primary-source confirmation. Watch: whether the AMD→TSM cell's two-session winning run is N=2 or the start of sub-book recovery, and whether the primary NVDA→ASML cell (dormant since 08-03) becomes measurable again once traded-day near-miss logging is fixed.