House Delta — 2026-09-18

Desk Plan — House Delta — 2026-09-18

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-09-18-pm.md PM: DELTA

Shared Market Read

• Event risk today: LOW — no Tier-1. No Tier-1 events at all through 2026-09-25; the Fed's 25bp hike (to 3.75–4.00%, Warsh signaling at least one more this year) is in the past, and the BOJ hiked to a 31-year-high 1.25% overnight (fully telegraphed — sell-the-fact, yen weaker at ~157.8). Today's scheduled reads are all second-tier: 9:15 AM Industrial Production (+0.3% MoM cons.), 10:00 AM CB Leading Economic Index (+0.1% cons.), two Fed speeches (Bowman 9:30, Schmid ~11:45), Baker Hughes 1:00. Firm-wide status: GO. Next Tier-1 — and my next hard-skip — is PCE Sep 30 (then NFP Oct 2, CPI Oct 14). No earnings binary today.
• Session character expected: Mechanically-driven, not informationally-driven — and that is the defining fact of the session. Today is quadruple witching: quarterly index futures (opening-based settlement), index options and single-stock options all expire, ~$6.2tn of US options exposure, tracking toward June's record $7.7tn. The briefing is explicit that the open is firmer and tech-led but that expiry mechanics can distort the first 30 minutes and the closing auction, that witching-day open prints can be mechanically stale, and that a choppy open must not be read as a directional signal. Structure may not form cleanly until the expiry dust settles after the first 30 minutes.
• VIX regime: 15.27 (−0.17 vs prior close 15.44) — LOW (<18), and nowhere near the 30 hard-skip. This is the regime DELTA's mechanism is strongest in (dev PF 2.59 in VIX < 20; the family-inverted dependency vs. momentum agents — high macro vol adds noise to the supply-chain transmission rather than amplifying it). The vol backdrop is favorable; the mechanics of the session are not.
• Key levels: Index context (briefing, Alpaca-verified, prior close 2026-09-17): QQQ 716.89, above SMA20 712.73, ATR(14) 8.42, prior-day range 713.32–718.03; SPY 762.64, below SMA20 764.31, ATR(14) 6.66, prior-day range 759.98–763.57 — tech leading, value lagging, consistent with the briefing's "tech-over-value, contained range" shape. Gap on DELTA's own instrument-level inputs: the briefing carries no NVDA / ASML / AMD / TSM price at any verified level — it is Alpaca-owned at the index/ETF level and gives only a directional semis read. Today's scanner pre-market tape supplies directional single-name prints on the complex — INTC +7.67%, ARM +8.57% (Sep 17 close), ALAB +9.06% premarket, with NVDA/MU/AMD/MRVL named in the bid — but no NVDA, AMD, ASML or TSM pre-market print in either document. Those prints are adopted as direction and sector dispersion only, not as owned price levels. No instrument-level price is adopted and none is invented; all four pair inputs require live confirmation at the cash open. Web-sourced single-name figures were reconciled against the briefing's verified index references and, having no instrument-level counterpart to reconcile to, are carried as direction only — not as a basis for a directional thesis.
• Macro backdrop: A genuine post-decision risk-appetite relief trade layered on a still-elevated long end — semis/AI bid on Jensen Huang's Sep 17 comment that Nvidia expects to roughly double the volume of chips it sells next year (Asia's chip complex rallied on it: Samsung/SK Hynix led, Nikkei +1.4–1.7%); oil down a third straight session (WTI back under $100 on Saudi East-West pipeline-restoration expectations, overriding fresh Saudi–Houthi strikes, with the stress in refined products not crude); the 10-yr holding ~4.94% after losing ~10bp Thursday, backing off its 19-year/2007 high of ~5.04%. The single transmission check the briefing names: semis vs. yields — the relief holds only while the 10-yr stays off 5%.

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

Call: No trade expected — the engine runs (not a skip day, no Tier-1, LOW event risk, firm GO), but the session is structurally incompatible with this mechanism twice over, and neither reason is a threshold I can move. p_trade 0.15, direction long, conviction low. This is the second consecutive session with the pre-market lean on the LONG side — the side this book can actually take — and for the second consecutive session the shape of that lean is the problem, not the calendar.

Why the asymmetry in the frontmatter: "no trade expected" but p_trade up from 09-17's 0.10. These are not in tension, and I want the record honest about which is which. p_trade is the probability the engine logs ≥ 1 real trade — not whether it should. Quadruple witching mechanically widens first-hour dispersion in both directions, and my entire signal sits in the 09:30 → 09:55 first-hour window — the exact window the briefing says is distorted by opening-based index-futures settlement. So the probability that some leader mechanically clears ±1.5% before 09:55 is genuinely higher today than on a normal LOW-VIX session: a wider tape is a wider tape, and a stale/mechanical open print can clear a floor it did not earn informationally. That raises p_trade. It does not make the resulting trade good — it makes it less trustworthy. The 0.15 is a probability of a mechanical trigger, not an endorsement.

• Primary reason — structural incompatibility #1: the day's live driver is complex-wide by construction, which is exactly this mechanism's stand-down at its strictest. My edge is an idiosyncratic design→manufacturer cascade: news about NVDA (demand, capacity, product cycle) telegraphing downstream to ASML over hours. Today's driver is Huang's sector-level "chip volume doubles next year" comment — a statement about the whole AI-chip complex, already priced broadly into Samsung, SK Hynix, INTC (+7.67%), ARM (+8.57%), ALAB (+9.06%). That reprices leader and laggard simultaneously on one systematic factor. It is systematic beta by construction, not a supply-chain cascade. And there is a direct, recent read on this: yesterday (09-17) was the same shape — a complex-wide post-FOMC semi bid, the briefing's "only complex with a genuine two-day directional bid" — and the leaders printed AMD +0.98% / NVDA +0.39%, both inside the ±1.5% floor. A strong sector bid did not lift the leaders to the floor yesterday. That is the cleanest available evidence that sector bid ≠ a qualifying idiosyncratic leader move, and it is one session old.
• Primary reason — structural incompatibility #2 (new today, and the sharper of the two): expiry mechanics corrupt my measurement window. My signal is not a close-to-close feature; it is the leader's first-30-minute return off the 09:30 cash open, measured at 09:55. Quadruple witching distorts precisely that window — opening-based settlement of quarterly index futures, mechanical/stale open prints, ~$6.2tn of expiring exposure. The briefing's own instruction — "do not read a choppy open as a directional signal" — is, for my book, a statement that my primary input is untrustworthy today. A first-hour move on a witching open is not cleanly attributable to supply-chain information, which is the only thing my edge is built on. Even a genuinely idiosyncratic leader impulse would be unmeasurable as idiosyncratic inside a mechanically-driven open. This is the first session in Phase 3 where the measurement of my own signal is impaired by construction, independent of what the leaders do.
• The honest tension, stated rather than sanded. LOW VIX (15.27) is my best regime by the backtest; no Tier-1; firm GO; and the complex is bid two days running — on paper the most constructive setup I have had since 08-03. And the structural stand-down is not yet battle-tested on the up side: NVDA has not cleared +1.5% in the first hour in 24 consecutive Phase 3 sessions (last clear: 08-03, NVDA +2.32% → ASML LONG, +$184.68, a win). So the likeliest reason today joins the likeliest reason every day for the past five weeks: the floor is never reached — the stand-down has never actually had to refuse a real qualifying up-impulse, because none has arrived. I should not dress up a five-week structural absence as a disciplined refusal. The distinction I am pre-registering: if a leader clears +1.5% today, the default read is that it is witching-mechanical and/or complex-wide, and the trade is stood down — the one way that inverts is a laggard-leg (ASML or TSM) headline carrying news independently of the leader, which is not present in any input I hold this morning.
• Setup present: no — structurally incompatible, on two independent grounds. (1) The driver is complex-wide (Huang, sector-wide), not an idiosyncratic NVDA/AMD cascade. (2) The first-hour window my signal lives in is mechanically distorted by expiry. Either alone stands the session down; today I have both, and the second one is new.
• Sizing vs. event risk: n/a — no position expected. Standard 0.75% single-position risk is moot. If — against both shape reads — a leader clears +1.5% on a move that reads as genuinely idiosyncratic (laggard-leg news carrying independently of the leader), standard sizing applies unchanged: event risk LOW, and the hard 15:30 flat + 2.0× ATR(14, 5-min) stop carry zero overnight and zero event exposure by rule. One sizing note specific to today: 2.0× ATR is the stop, and today's ATR inputs (QQQ ATR(14) 8.42; my own 5-min ATR) may still be carrying post-FOMC elevation — a wider stop is not a reason to trim size, but it is a reason not to read a witching-inflated first-hour range as signal quality. No parameter is adjusted — the stop is the stop.
• Session-character fit: poor, and worse than yesterday for a reason specific to me. A mechanically-driven witching session is a flow/expiry session, not a pairs-convergence session. Yesterday's incompatibility was story (complex-wide repricing). Today's adds mechanics: the session's cleanest structure is deliberately absent from exactly the window my edge samples. The one path to a clean LONG remains a leader that decouples from the sector bid and builds an idiosyncratic first-hour impulse — and on a witching open I may not be able to distinguish that from a mechanical print even if it occurs. That ambiguity is itself disqualifying.
• Key levels (mine only): the only load-bearing number is the leader's first-hour return off the 09:30 cash open, checked at 09:55, vs. the ±1.5% floor — NVDA ≥ +1.5% → buy ASML at the 10:00 bar open (primary, edge-carrying cell, live 2W/0L, last live trade 08-03) and AMD ≥ +1.5% → buy TSM at the 10:00 bar open (secondary; degraded sub-book n=8, WR 25.0%, PF 0.191). A leader ≤ −1.5% is the disabled-SHORT branch → no trade by rule, not a missed LONG. No instrument-level price is adopted for any of the four legs (see Key levels above — the briefing carries none and I invent none).
• Invalidation / no-trade reasons: (1) the first-hour window is expiry-distorted by construction — the measurement itself is impaired today, and any leader clear is presumed mechanical until a laggard-leg catalyst proves otherwise (the new, sharpest gate); (2) a first-hour leader up-move that reads as complex-wide semi/AI beta (Huang-driven sector bid / short-covering) rather than an idiosyncratic supply-chain cascade — the standing structural stand-down, binding at its strictest because the repricing sits on my own instruments and because 09-17 showed a sector bid failing to lift the leaders to the floor; (3) leaders opening broadly down ≥1.5% = the disabled-SHORT side (ENABLE_SHORT = False) — no trade by rule; (4) any intraday VIX ≥ 30 closes consideration; (5) the remaining risk-derived stand-down is unchanged. Next hard-skip: PCE Sep 30 (NFP Oct 2, CPI Oct 14).
• Up-branch / calibration status (carried forward, unchanged): the 09-04 pre-registered flag — a third NVDA up-graze inside ~0.3pp turns regime into a TEMPER calibration question — remains queued and untouched, and is not advanced today unless a leader actually clears the floor. The ledger stands: NVDA +1.23% (09-02), +1.02% (09-03), AMD +1.20% (08-31); 09-08 NVDA −1.55% (down-side, disabled); 09-10 NVDA −1.44%; 09-14 AMD −0.27% / NVDA −0.21%; 09-17 AMD +0.98% / NVDA +0.39% — the best up-side texture since 09-03 and still inside the floor. A witching-mechanical clear would not be counted toward that flag, because a mechanically-stale print is not an informational graze. No unilateral gate change.
• Book state (carries forward untouched): cumulative −$442.33 across 10 LONG trades, 4W/6L, WR 40.0%, PF 0.494 (SPRT n=10, LLR −0.159, CONTINUE). AMD→TSM sub-book n=8, WR 25.0%, PF 0.191 — 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×, nowhere near the 1.5 gate), pending TEMPER cadence-ratification; read as a signal/regime watch, not a parameter fix. NVDA→ASML 2W/0L remains the quality cell; its trigger has not recurred since 08-03. Rolling 6-month PF monitor is unchanged and not escalated.
• ✔ Operational item — carried forward, desk-top priority: the per-agent 60s run-trades watchdog, and how much this plan leans on p_trade. learnings.md (desk-blind, 2026-09-17 entry) records DELTA starved by the per-agent 60s watchdog on three sessions on record — 09-09, 09-10, 09-17 — with run-trades logging "delta failed or exceeded 60s — skipped" on a cold cache, and DELTA alphabetically first in the glob, so not a cutoff-order effect. 09-17 was subsequently backfilled standalone (python agents/delta/paper_trade.py 2026-09-17, unset PYTHONPATH) and returned no_trade (AMD +0.98% / NVDA +0.39% vs. ±1.5%), so that scored row is correct. This is the reason I state the p_trade caveat explicitly rather than letting the number stand bare: my forecast is scored against trades.csv — if the 60s watchdog trips again today on a cold cache, the run produces no row at all, and my 0.15 is scored against an absence that reflects the pipeline, not the tape. In that failure mode p_trade is not just wrong, it is unscoreable, and the session's leader first-hour returns are lost — which on a witching session is exactly the observation I would most want. A missing observation must never be read as evidence of a near-miss, and it must never be narrated as a strategy stand-down (the 2026-09-14 learnings entry records exactly that misstatement in a shared EOD readout — the absent artifact is a pipeline state, not a mechanism choice). Record inconsistency flagged, not resolved: learnings.md's 09-17 entry names 09-09, 09-10, 09-17 as the starvation set, while this desk's own 09-17 plan carried an intelligence/runner-health-2026-09-15.md REVIEW verdict for a runner_failed on 09-15 (missing daily_2026-09-15.md) as a separate instance. Whether 09-15 is a fourth starvation case or a different failure of the same runner is not settled by the inputs in front of me, and I am not going to pick a version to make the count tidy. Standing asks, both now supported by multiple instances: (a) backfill any remaining standalone holes; (b) the ledger-freshness post-run check — every active stack must have today's performance/daily_.md after run-trades, else flag. Already logged in learnings.md 2026-09-10; the repeats are the escalation case for wiring it in rather than re-noting it.
• Watch items (informational only — no action today, and none of these is an entry signal): (1) whether NVDA/AMD decouple from the sector bid and build a genuine idiosyncratic first-hour up-impulse, distinguished — as always — by the laggard-leg (ASML/TSM) news carrying independently of the leader; on a witching open I will also need to judge whether any such move is distinguishable from mechanical flow, which is the honest uncertainty today. (2) whether the 10-yr re-tests 5.00% and snaps the semis relief — the briefing's named transmission check, and the switch that decides whether today's tech-led bid extends or fades. (3) whether the low-VIX, no-Tier-1, complex-bid configuration finally produces a clean ≥1.5% first-hour up-impulse on a leader — the same watch every session since 08-03, and the one that the open event-day diagnostic run (skip disabled, still unperformed, moderate priority, not something DELTA runs unilaterally) exists to interrogate.

Plan Filed

• Filed: 2026-09-18 04:34 PDT (pre-open)
• Frontmatter forecasts complete for every active member: yes (delta — p_trade 0.15, direction long, conviction low)
• Briefing event risk confirmed: LOW — no Tier-1 through 2026-09-25; quadruple witching is the session's structural feature
• Verdict: No trade expected — the engine runs (LOW event risk, no Tier-1, VIX 15.27 LOW regime, firm GO), but the session is structurally incompatible with the mechanism on two independent grounds: (1) the live driver is complex-wide (Huang's sector-level chip-volume comment), which is systematic beta, not the idiosyncratic design→manufacturer cascade the edge requires — and 09-17's strong sector bid left both leaders inside the ±1.5% floor; (2) quadruple witching distorts the 09:30–09:55 first-hour window the signal is measured in, so the measurement itself is impaired today. p_trade 0.15 reflects the higher chance of a mechanically triggered clear, not an endorsement of it; direction long, conviction low.

No trades taken.

House Delta — Desk Reflection — 2026-09-18

Cadence rule: Weekly/Friday or same-day-on-trade (WS5 per-desk cadence). Today is the Friday trigger. Period covered: 2026-09-11 (last desk reflection) → 2026-09-18. Scheduled sessions in window: 09-14, 09-15, 09-16, 09-17, 09-18.

Single-member desk — no cross-member synthesis to run. The desk reflection is DELTA's own analysis at desk scope; the per-member compatibility file (agents/delta/reflections/daily_2026-09-18.md) is written separately and remains first-person.

Desk Performance Summary

Desk total for the session: $0.00 — two hard skips, two no-signal sessions, and two runner_failed holes (see below).

Period session map:

SPRT state unchanged from 09-11 — zero new trades, so zero LLR movement (n=10, W4/L6, −0.159). Cumulative P&L unchanged at −$442.33. AMD→TSM sub-book unchanged: n=8, WR 25.0%, PF 0.191 — not triggered (no new AMD LONG evidence; AMD's closest was +0.98% on 09-17, still 0.52pp short of the floor).

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

Binding gate this period: the LONG up-branch — leader first-hour return ≥ +1.5% off the 09:30 cash open, read at 09:55. Reading: REGIME, not calibration — upheld, and confirmed from a previously-rare angle.

1. No up-side graze anywhere near the floor — the window's strongest observation is 09-17, and it argues for regime, not calibration. On the most constructive single-day backdrop in weeks — post-FOMC relief, semis bid two days running, VIX LOW, firm GO, the briefing calling the complex "bid" — the leaders printed AMD +0.98% / NVDA +0.39%, both inside the +1.5% floor. An energetic, widely-owned semiconductor bid could not lift either leader to the LONG threshold. That is a structural-absence reading: a sector-level relief bid reprices leader and laggard together on one systematic factor and does not produce the idiosyncratic first-hour leader impulse this LONG edge needs. It is not a threshold being repeatedly grazed. The 09-04 watch-to-escalate (a third NVDA up-graze inside ~0.3pp → calibration conversation) still has not fired: the closest up-value all period was 0.52pp short, and NVDA never exceeded +0.39%.
2. The only other observed sessions are clean non-events or correct skips — no threshold repeated. 09-14 both leaders were negative (regime, nothing close); 09-16 is a correct pre-registered FOMC hard-skip. Across the three observed sessions, not one value approached +1.5% from below. There is no repeated single-threshold miss in the data — regime, wait.
3. The honest caveat: the two most informative up-side sessions are unobserved. 09-15 and today (09-18) both carry runner_failed — the run-trades per-agent 60s watchdog timed out on delta (alphabetically first in the glob, so not a cutoff-order effect; runner_status_2026-09-18.tsv shows delta failed_or_timeout while all nine other members completed), producing no trades row and no performance daily for either day. Today matters most: it is the session where the plan raised p_trade (0.15) specifically because quadruple-witching mechanics mechanically widen first-hour dispersion — i.e. the one session in five weeks where an up-clearing leader was deemed genuinely more likely, and the engine lost the observation. A runner_failed state is a REVIEW state, not a no-trade/no-setup day, and not evidence of a clean skip. It is not scored as a near-miss, and it is not calibration evidence either way — but it is a loss of exactly the informative observations this regime-vs-calibration question needs. The regime verdict above rests on three observed sessions, all cleanly regime; it is not diminished by the holes, but it is not the same epistemic position as if 09-18 had confirmed the no-trade.

Implication — no action on the gate stack; escalate the pipeline, not the strategy. No TEMPER calibration conversation is warranted on the LONG floor (no repeated graze; the strongest bid day produced only +0.98%/+0.39%). The AMD→TSM stop-calibration diagnostic (2026-09-05, verdict MIXED/MARGINAL read NO — stop-multiplier is not the fix) remains pending TEMPER cadence-ratification via the 2026-09 Desk Review; nothing in this window changes it. What this window does escalate is operational: delta was runner-starved on four of the five sessions in the window (failed_or_timeout on 09-14, 09-15, 09-17, 09-18; only 09-16 completed), with 09-14 and 09-17 backfilled standalone but 09-15 and 09-18 remaining genuine holes. The ledger-freshness post-run check (every active stack must have today's performance/daily_.md after run-trades, else flag) and the backfills are the desk's top operational priorities.

Open TEMPER Items (Carried Forward)

Reconnaissance Ledger Items

• Backfill 09-18 and 09-15 (python agents/delta/paper_trade.py , unset PYTHONPATH) — the two genuine holes. 09-18 is the priority (today; quad-witching; the no-trade is likely but the leader first-hour reads must be captured, not assumed). 09-15 pre-market was unambiguously down (structural stand-down) so its backfill is lower-stakes but still owed.
• Implement the ledger-freshness post-run check — every active stack must have today's performance/daily_.md after run-trades, else flag. This is the generalizable fix for the silent-observation-loss class, now at 2 confirmed holes (09-15, 09-18) this period on top of the historically documented ones (09-09, 09-10).
• Event-day diagnostic remains a standing obligation (see Open TEMPER items).

Desk Summary for Orchestrator

House Delta: zero trades this period — 09-14 no_signal, 09-15 runner_failed (hole), 09-16 Tier-1 FOMC hard-skip (correct), 09-17 no_signal, 09-18 runner_failed (hole — today, runner_status shows delta failed_or_timeout while all other members completed). Period net $0.00, cumulative −$442.33 (10 trades, 40% WR), SPRT CONTINUE unchanged (LLR −0.159). Binding gate: LONG up-branch (leader first-hour ≥ +1.5%) — verdict REGIME, not calibration, confirmed from a sharper angle than last period: the strongest sector bid in weeks (09-17, post-FOMC relief, semis bid, VIX LOW, firm GO) lifted leaders only to AMD +0.98% / NVDA +0.39% — a sector-wide bid that cannot reach the idiosyncratic LONG floor is structural absence, not a grazed threshold; the 09-04 watch-to-escalate and the AMD→TSM stop-calibration verdict (read NO) both stand unchanged. One thing watched: the pipeline, not the strategy — delta was watchdog-starved on 4/5 sessions this window, leaving 09-15 and 09-18 as unobserved holes, the latter being the quad-witching session where the plan deliberately raised p_trade because the floor was genuinely more reachable; the runner_failed state is a REVIEW, not a clean skip. No live gate touched; only a TEMPER cadence-ratification of the standing AMD→TSM diagnostic and an operational pipeline fix are owed.

Reflection Filed

• Filed: 2026-09-18 (EOD, desk cadence — Friday trigger)
• Next session: Mon Sep 21 (S&P 500 quarterly rebalance effective date — index-fund districts trade the Friday close, but Sep 21 is the effective open; structural-flow awareness, not a Tier-1). No Tier-1 until PCE Sep 30 (my next hard-skip; then NFP Oct 2, CPI Oct 14). Watch: whether the LOW-VIX post-FOMC relief tape finally produces a clean ≥ +1.5% first-hour up-impulse on NVDA→ASML, or whether the complex-wide-bid texture (which reprices leader and laggard together) continues to deny the idiosyncratic impulse my LONG needs — and, operationally, whether the run-trades watchdog clears delta next session so the observation is actually captured.