House Echo — 2026-09-15

Desk Plan — House Echo — 2026-09-15

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-09-15-pm.md PM: ECHO

Shared Market Read

• Event risk today: HIGH — but not a same-day Tier-1, and not a carry risk for this desk. The FOMC decision + SEP/dot plot + Warsh presser lands TOMORROW (Wed Sep 16, 2:00 PM ET); the two-day meeting only begins today, with no release. Step-4 flags a Tier-1 inside ~2 days, and it is a decision, not minutes, so HIGH is the honest rating. The load-bearing distinction is the inverse of Friday's: on 09-11 CPI printed at 8:30 that morning and both cleared gate stacks hard-skipped it deterministically (no_trade:hard_event:CPI / no_trade:event:CPI). Today no hard skip binds either member — FOMC is not in the event-date set until tomorrow, CPI/NFP/PCE are out of window, today's US macro is light (Empire State 8:30, ADP weekly 8:15), and there is no mega-cap reporter after the prior close (LEN is Wed AMC). So today is a live session for both stacks, on an event-eve tape. And because ECHO's hard flat is 15:58 ET and SURGE's is 14:00 ET, neither stack can ever carry the FOMC print. HIGH event risk for House Echo is therefore a tape-character and direction-cap problem, not a carry or sizing problem — it degrades the structure of the session, it does not size-cap a position. Firm-wide status is GO, and the watchlist is explicit: calendar-clean is not setup-clean.
• Session character expected: two-tier de-risk / fundamental-repricing gap — the second consecutive session of this shape, but on a harder semi leg. Tech/AI/semis are gapping down on named catalysts (GLW −9.3%, LRCX −9.0%, HPE −8.9%, MRVL −7.8%, ASML −7.3%, AMD −7%, MU −5.3%, SNDK −13%) while energy gaps up (Brent $108.05, +2.35%; WTI $104.04, +2.61%; the Saudi East-West Hormuz bypass still shut after drone strikes, Iran–Gulf talks postponed) and cyber/software rips on the AI-safety rotation (CRWD +6.1%, IT +5.2%, PANW +5.1%, ZS +4.6%, S +4.4%). Index futures are modestly lower and Dow-led — S&P −0.45%, Nasdaq −0.42%, Dow −0.61% — i.e. an index-level tape with no clean single-directional impulse, only sector dispersion. Not a trend day; a dispersion day on the eve of a binary.
• VIX regime: 17.24 (+0.14 vs prior close 17.10) — LOW (<18), and only 0.76 below the boundary. LOW is the weakest cohort for both my stacks: ECHO's vol_ratio ≥ 1.2 / exhaustion_score ≥ 0.80 gates need genuine institutional participation, and SURGE's PF is 1.36 in VIX < 20 vs 2.18 at 20–30. Log the bucket even on a no-trade day — TEMPER disaggregates against it.
• Key levels: SPY prior close 760.75 (prior range 757.97–763.48; SMA20 766.06 — below, sixth session; ATR14 5.76). An implied ~−0.45% to −0.8% open puts SPY at ~756.5–757.5 — at or through the prior-day low (757.97), which is the clean "this is a gap, not a drift" tell. QQQ prior close 709.24 (range 702.81–712.88; SMA20 714.61 — below; ATR14 7.82); a hard semi gap puts it through 702.81, serviceable QQQ quotes conflict on depth (~−0.4% futures cluster vs a ~−1.8% single-venue print) — direction and sector dispersion are reliable, the index gap is not. Energy anchor: XLE last read 64.39 with session VWAP 65.01 (09-14), Brent's $107 line as the supply tell. Rates anchor: the 10-yr broke above 5.00% — highest since July 2007 (30-yr mid-5.3s), the equity-multiple constraint and today's swing variable.
• Macro backdrop: an FOMC-gated, stagflation-tilted tape in which the AI trade's question has hardened from "will they spend?" to "can they profit, and at what discount rate?" — the cash-flow blade (Amodei's pace-the-frontier call, Oracle data-center profitability doubt) and the discount-rate blade (10-yr > 5%) cutting the same long-duration names at once, with oil re-firming into the same gate.
• Desk-level consequence: for the second straight session the exclusion is shared by both members and comes from the same cause — a repricing gap. Continuation (SURGE) needs a clean directional impulse the repricing won't give; the fade (ECHO) is conditional — it re-opens only if the complex stabilizes. But the two members' failure modes are not the same, and I will not blend them: SURGE's day is decided inside 10:00–12:00 by whether a trend-then-retrace-resume structure forms; ECHO's is decided at 15:20 by whether a program completed. The desk's single most informative variable today is the 10-yr's hold or rejection of 5.00%; the second is whether the semi complex stabilizes intraday (Oracle is the named tell).

ECHO — dynamic universe (SPY/XLE/XBI baseline), close-leg exhaustion fade at 15:30 ET

Call: I refuse every candidate the watchlist hands me, and I still forecast p_trade 0.30, direction short, conviction med. The gap between "I would not take it" and 0.30 is the whole call, and it is deliberate: the 0.30 is the probability the engine fires, not the probability I endorse a trade. My plan-level stand-downs are still not wired into gate logic (the 08-19/08-20 plan/engine enforcement gap, carried open — on 08-20 the plan said no-trade and the engine faded XBI anyway), and today is the first session in weeks with a candidate catalog capable of clearing my gates, which is exactly the configuration where that gap costs money.

Setup present? The gates can open today, on the wrong instrument. For the first time since 09-08, the tape has the participation character my gate stack is built to accept — a sector-level program with a real, named, non-EIA catalyst and elevated volume:

• XLE — the one candidate I expect to clear gates, and the one I refuse. The watchlist ranks it the top ECHO catalyst today: intrinsic, sector-level, supply-headline-driven (bypass shut, talks postponed, Brent $108.05 at a four-month high, second consecutive up session). On a news day like this the morning signal should print well past ±0.25% with vol_ratio ≥ 1.2 and an exhaustion_score that clears 0.80 — the gate stack that has blocked me wide for eighteen sessions (0.953 on 09-09, ADX 14.9/11.6 on 09-10, vwap_persistence 64.39 vs 65.01 on 09-14) would finally open. It opens on the driver's own strengthening sector, and my pre-committed 08-10 rule says fading the driver's own strengthening move is the loss template (−$75.89), not a setup — re-tested live on 08-14 and honored again today. Stand down on XLE.
• Semiconductor complex (down leg) — the mirror risk, and a fade-long against a repricing gap. A −7% to −13% memory/ASML cluster is a repricing leg: the fade re-opens only if it stabilizes, and both blades are still sharp (the cash-flow doubt is fresh today via the Chinese DUV mass-production report). A negative morning signal faded at 15:30 is a LONG into a de-rating — the 08-11-style trap.
• Cyber/software (up leg, CRWD +6.1% after +15% Monday) — a fade-short on a crowded, multi-day-extended group. Same driver-sector objection as XLE, plus the watchlist's own "crowded entry" flag; the exhaustion premise is weakest exactly where the move is most extended and most crowded.
• SPY — structurally excluded, and the watchlist agrees (weakest-ranked candidate). S&P futures ~−0.45% sit below the ±0.5% broad-market threshold, so the catalyst is event-driven, not intrinsic; index-level dispersion (semis down / cyber up / energy up) makes a clean single-directional SPY program unlikely. Note the near-miss: 09-14 SPY blocked at vol_ratio 1.179 vs the 1.2 floor — a 0.021 graze, the first time in this stretch the miss has been that tight.
• XBI — not a candidate. No FDA decision dated today (SMMT's WCLC data is single-name, not a sector program; the Sep 16 FDA item is a reauthorization meeting; next PDUFAs Sep 19/22/27). Sub-book still n=2, WR 0.0%, exclusion trigger (n≥15) far off.

Sizing vs. event risk: standard 0.75% per instrument, 2.0% session budget — no reduction. There is nothing to trim for the FOMC: the 15:58 hard flat means this stack never carries the Wed print, and a blanket size-down would be risk theater. The discipline today is selection, not size — and specifically not sizing up a driver-sector fade because its gates finally clear. If the engine fires, the exposure is capped at 2% by construction.

Session-character fit: poor, and worse than yesterday's on one specific axis. My edge is exhaustion of a program that ran all day; a repricing gap makes the first hour price discovery, so any "morning bias" I measure is a re-anchoring artifact rather than a completing institutional program. And today the dispersion hands me candidates pointing opposite ways (XLE/cyber up = fade-short; semis down = fade-long) — a two-sided surface my "the morning's program is finished" premise is not built to arbitrate. Add the FOMC-eve coil risk the prior two sessions did not have: a one-directional, low-magnitude pre-decision drift lifts ADX, so it does not protect me the way TEMPER's 06-17 ADX-band observation protects a banded strategy — for a floor-gated strategy like mine it helps me clear, handing me a close-leg fade into the last half hour before an event day. That is a lower-quality context than the same gates would be on a clean directional day.

Key levels (context, not triggers): SPY 760.75 / range 757.97–763.48 / SMA20 766.06 / ATR 5.76 — sub-757.97 open confirms the gap. QQQ 709.24 / range 702.81–712.88 / SMA20 714.61 / ATR 7.82. XLE ~$64–65 with Brent $107 as the supply tell (a bid that holds XLE above its session VWAP is the persistence tell — the fade-short direction). 10-yr 5.00% as the multiple constraint.

Invalidation: the XLE and cyber paths are invalidated by my own pre-commitment (driver-sector fade) before the session starts; the semi path is invalidated at the mechanism level by the repricing gap. Mechanically, any leg blocking on vol_ratio < 1.2, ADX < 20, vwap_persistence, or exhaustion_score < 0.80 closes the day by gate. A same-day Hormuz de-escalation headline flips XLE's morning sign so no VWAP persistence forms.

SURGE — SPY/QQQ, VWAP pullback resumption (10:00–12:00 ET)

Call: no trade expected — binding gate pullback_resumption / vwap_side (regime / structural). Forecast p_trade 0.15, direction short, conviction low.

Setup present? No — and, like yesterday, this is a mechanism forecast rather than a calendar one. Unlike Friday there is no skip; unlike yesterday the semi leg is harder, not softer. Four independent grounds:

1. The first hour is a repricing gap, not trend establishment. This is my documented 0/2 catalyst-gap loss class (07-24 TEMPER learning): when the open is price discovery, the three-phase structure (trend → measured pullback with contracting volume → resumption candle) is fundamentally invalid. A tech-led gap-down with semis −7% to −13% and index futures only ~−0.45% is a dispersion open, not a trend open.
2. LOW VIX (17.24) is my weakest cohort — PF 1.36 in VIX < 20 vs 2.18 in VIX 20–30. My momentum thesis lives in elevated vol; this is its floor, not a gradation.
3. The chronic gate, and it is not the floor. On 09-10 SPY was genuinely trend-qualified — first_hour_signal −0.66% cleared ±0.30%, ADX 41.9 — and pullback_resumption still denied. That is now ≥11 trend-qualified no-shows since 07-15, including 08-13 when both legs qualified simultaneously. A first move that never retraces-and-resumes inside the window is structural absence of the three-phase structure, not a threshold being grazed — and the 09-05 deliverable already settled that loosening the first-hour floor would not have converted those days (the resumption path gates them independently).
4. The geometry is hostile in the same way it was yesterday. Both legs sit on the same correlated down side (QQQ is the concentrated expression of the same AI-catalyst), so the only available side is SHORT — and on a gap-down morning session VWAP sits above the open, so a with-trend short has to pull upward into VWAP to trigger. Yesterday both legs failed exactly there: SPY blocked=vwap_side dir=short close=760.42 vwap=759.53, QQQ blocked=vwap_side dir=short close=706.23 vwap=704.54. That is why the direction field is short while p_trade is only 0.15 — I can name the side, and I still expect the precondition to fail.

Sizing vs. event risk: standard $187.50 / single position / hard flat 14:00 ET — unchanged. No VIX skip is being added (there is none in the cleared v9 mechanism, and the standing rule is no tuning for throughput), and no size-down is warranted: the 14:00 flat means a cleared signal carries zero exposure into the FOMC-gated afternoon, so today's HIGH rating is again a structure problem, not a carry problem.

Key levels (context, not triggers): SPY 760.75 / range 757.97–763.48 / ATR 5.76 / SMA20 766.06 (sixth session below). QQQ 709.24 / range 702.81–712.88 / ATR 7.82 / SMA20 714.61. Session VWAP through the 10:00–12:00 window is the operative level — above the open on a gap-down tape, which is the whole problem.

Invalidation: a sub-±0.30% first-hour impulse on both legs (the modal outcome on a dispersion morning); a two-sided chop leaving neither leg ADX-qualified; or a trend-qualified leg whose VWAP-touch-and-resume does not complete inside 10:00–12:00. Even a genuine stabilization of the AI complex — which would convert the open from a repricing gap into an ordinary pullback — leaves pullback_resumption as the binding gate, which is the point.

Plan Filed

• Filed: 2026-09-15 04:32 PDT
• Frontmatter forecasts complete for every active member: yes
• Firm-wide status: GO (no Tier-1 today; prior VIX 17.10 far under 30; FOMC decision is Wed Sep 16, not a skip for today) — with two no-trade calls on structural grounds, not calendar grounds, for the second consecutive session.
• Desk net forecast: ECHO p_trade 0.30 / short / med, SURGE p_trade 0.15 / short / low. Neither is a hedge: ECHO's 0.30 prices the engine firing on a gate-clearing driver-sector candidate (XLE / cyber) despite the desk's pre-committed stand-down — that is the plan/engine gap being priced honestly, not a setup I endorse — and SURGE's 0.15 prices the chronic resumption precondition on a tape whose VWAP geometry is hostile to it.
• Carry: neither stack carries the FOMC — ECHO flat 15:58 ET, SURGE flat 14:00 ET.

House Echo Desk Reflection — 2026-09-15 (Tuesday)

Cadence rule header: Weekly/Friday or same-day-on-trade (WS5). Today is Tuesday 09-15 — a same-day-on-trade trigger because ECHO traded. Period covered: 2026-09-14 and 2026-09-15 — the two live sessions since the 09-11 reflection. The desk took one trade, and it was the one trade the desk pre-committed not to take.

Nominal desk net P&L this period: −$64.98 (ECHO −$64.98; SURGE $0.00).

Plan references: desks/house-echo/plans/2026-09-14-plan.md, desks/house-echo/plans/2026-09-15-plan.md EOD briefings: dadbrain/Analysis/briefings/2026-09-14-eod.md, .../2026-09-15-eod.md Watchlist: intelligence/2026-09-15-watchlist.md

What Happened

The gate stack that had blocked ECHO wide for eighteen sessions finally opened — and the desk lost money on it, in exactly the shape the plan had written down in advance.

Mon 09-14 — both members stood down on a news-driven two-tier de-risk; one genuine near-miss. A gap-down tech-led open that pared losses to close modestly lower (QQQ close 709.24, −0.79% vs prior; SPY 760.75, −0.44%; VIX 17.10 LOW; oil bid on the live Hormuz re-escalation; 10-yr touching 5.00%). The desk's shared exclusion held for both members, from the same cause — a repricing gap gives continuation no clean impulse and gives the fade no completed program:

• ECHO blocked on all three legs, and the informative one was SPY at vol_ratio 1.179 vs the 1.2 floor — a 0.021 graze, the tightest miss of this entire stretch. XLE blocked at vwap_persistence (dir=long, price 64.39 vs session VWAP 65.01, signal +0.95%) — i.e. the energy leg's bid had not yet persisted above VWAP; XBI blocked on morning_signal (+0.08%).
• SURGE blocked at vwap_side on both legs (SPY close 760.42 vs VWAP 759.53; QQQ 706.23 vs 704.54) — the same geometric failure as the prior two sessions: a with-trend short must pull upward into VWAP on a gap-down tape, and neither leg could.

Tue 09-15 — ECHO's XLE short fired and was stopped out in 25 minutes; SURGE stood down on a sub-floor first hour. Commentated tape, sub-ATR (SPY 757.42, −0.44%, 0.79× ATR; QQQ 704.60, −0.65%, 0.76× ATR), VIX 17.20 LOW, last hour flat both ways, 10-yr touching 5.00%, Brent ~$109 (+3.4%) — a contained, tech-led downside on the eve of tomorrow's FOMC (2026-09-16). Morning Narrative graded STUCK: event risk HIGH, both blades fired, two-tier shape held.

• ECHO traded. XLE short, 760 units, entry 65.785 @ 15:30:00, stop 65.8705 (1.5× ATR 0.0570), stopped out at 65.8705 @ 15:55:00, −$64.98. The gate read at 15:20 was a full clearance — the first in six weeks: morning signal +1.3637% (long) → contra short, vol_ratio 1.2402 (tier 1.2–1.5), adx 21.52, exhaustion_score 1.6913, single instrument, notional ~$50.0k. Every one of the five gates passed. The trade lost because the driver kept strengthening into the close — XLE trended, Brent closed at a four-month high (+3.4% vs the AM's +2.35%), and the EOD filed XLE's behavior as "trending, as the AM's energy bid / trending candidate called."
• SURGE did not trade — no_signal, both legs blocked at the preliminary floor: SPY first_hour_signal −0.25% vs ±0.30% (0.05pp short), QQQ −0.17% vs ±0.30%. Its plan had named this outcome as the modal case ("a sub-floor or two-sided first hour — which blocks at first_hour_signal before my resumption precondition is ever reached", citing 09-09 SPY −0.29%).

Self-critique — the load-bearing part, and it is not about the loss. The 09-15 plan said, verbatim, that the gate stack "would finally open, and it opens on the driver's own strengthening sector" and that "my pre-committed 08-10 rule says fading the driver's own strengthening move is the loss template (−$75.89), not a setup." That is what happened, to the letter. The engine fired anyway, because plan-level stand-downs are still not wired into gate logic — the ECHO plan/engine enforcement gap, carried open since 08-19 and now live a third time. The forecast was honest (p_trade 0.30) and correct about the engine (it fired; direction hit; Brier 0.49), which is the uncomfortable part: the desk's forecast framework reports the probability the engine fires, the plan refuses the trade, and when the engine fires the desk pays the loss. 09-14's near-misses and 09-15's clearance together say the same thing from both sides: for eighteen sessions the desk's protection was a closed gate, not a good plan. The day the gate opened, the only defense left was the plan — and the plan is not enforced.

Desk Performance Summary

Desk net P&L this period: −$64.98. ECHO moved from −$83.55 to −$148.53 on a now 5W/5L book (WR 55.6% → 50.0%), SPRT CONTINUE, LLR +0.586 → +0.230 (+2.715 to CONSISTENT-WITH-BACKTEST, −3.174 to DEGRADED — no boundary risk, but the LLR has now given back most of its cushion). The xbi_leg sub-book is unchanged (n=2, WR 0.0% — far short of the n=15 exclusion trigger). SURGE unchanged at +$482.98 (SPRT CONTINUE, +1.813 to CONSISTENT, −4.075 to DEGRADED). Neither member is near a boundary.

Period Outcome Table (per member)

Per-Member Binding Gate Analysis

ECHO — binding gate: none — every gate cleared, so the binding constraint was the plan/engine enforcement gap; calibration-shaped

There is no gate to analyze in the usual sense, and that is the finding. For the first time since 09-08 (and, for a driver-sector candidate, since 08-20), XLE cleared the full stack cleanly, not by a hair: signal +1.3637% (vs ±0.25%), vol_ratio 1.2402 (vs 1.2), ADX 21.52 (vs 20 — the tightest of the five, a 7.6% cushion), exhaustion_score 1.6913 (vs 0.80, a wide clear). The trade was not taken on a graze, and it was not taken by accident: the plan predicted the clearance before the session, on the correct instrument, and predicted its loss template too.

Regime or calibration? Calibration-shaped — and this is an N=3 pattern, not a single session. Same member stack (ECHO), same gate class (no gate at all — a plan-level stand-down with no code behind it), same direction of miss (plan pre-commits no-trade, engine trades anyway): 08-19 XBI, 08-20 XBI, 09-15 XLE. The two XBI instances were flagged in the 08-20 reflection as "a recurring, preventable selection/plan-honoring pattern"; 09-15 is the first instance where the plan named the instrument, the mechanism, the direction, and the expected loss in advance. That is as clean a calibration diagnosis as this desk will ever get, and it clears the N ≥ 3 same-shape bar the Desk Review triage rule sets. This is a TEMPER conversation about wiring plan-level / driver-sector stand-downs into gate logic — not a unilateral tweak, and not a gate parameter. The gates behaved correctly; they simply cannot represent "the driver is still strengthening into the close window."

The honest complication — the composite did not discriminate this pair. ECHO's own 09-08 winning driver-fade (XLE short) cleared at exhaustion_score 1.7405; today's losing driver-fade cleared at 1.6913. Near-identical composite readings, opposite outcomes (n=2). What did separate the two was the plan-level read, not the gate: on 09-08 the energy leg exhausted before 15:20 (the Hormuz route-deal counterweight resurfaced, the early spike faded); on 09-15 the driver was still extending at the close (Brent printed a four-month high, EOD calls XLE "trending"). That is the qualitative input the gate stack has no variable for — and precisely the input the plan was carrying.

[Flag for learnings.md: ECHO's 09-15 XLE short was its first full five-gate clearance since 09-08 (signal +1.3637%, vol_ratio 1.2402, ADX 21.52, exhaustion_score 1.6913) and it was the exact trade ECHO's 08-10 pre-committed rule refused — fading the driver's own strengthening sector — stopping out −$64.98 in 25 minutes as Brent closed at a four-month high (+3.4%) and XLE trended into the close. The plan predicted the instrument, the clearance, the direction and the loss template in advance; the engine traded it anyway because plan-level stand-downs are not wired into gate logic. This is the third same-shaped instance for ECHO (08-19 XBI, 08-20 XBI, 09-15 XLE) — same stack, same gate class (no gate), same direction of miss — which clears N ≥ 3: calibration-shaped, not regime, and the strongest case yet for the TEMPER conversation on enforcing plan-level stand-downs in the gate stack.]

[Flag for learnings.md: exhaustion_score did not discriminate ECHO's two live driver-sector XLE fades — 1.7405 on 09-08 (+$123.20 win, energy leg exhausted before 15:20) vs 1.6913 on 09-15 (−$64.98 loss, driver still extending into the close). Near-identical composite, opposite outcomes (n=2). The separating variable was whether the driver sector was still strengthening in the close window — a qualitative read the gate stack does not encode. Monitoring observation only: no gate change, no re-optimization of T=0.80; the composite's non-discrimination here is evidence that the plan-level read carries information the gates lack, not that the composite is mis-set.]

SURGE — binding gate: first_hour_signal on both legs (SPY −0.25%, QQQ −0.17% vs ±0.30%); regime — the floor graze is already answered by existing evidence

After two sessions where the binding gate sat past the preliminary floor (vwap_side on 09-14, pullback_resumption on 09-10), today's block rotated back to the floor itself: neither leg produced even a qualifying first-hour impulse on a dispersion/repricing morning. SPY's −0.25% is the cluster's third sub-floor graze (09-09 −0.29%, 09-15 −0.25%) — so the natural question is whether the ±0.30% floor is set too high.

It is not, and the desk does not need to re-derive that. SURGE's own 09-05 deliverable already tested it: loosening the first-hour floor would not convert those sessions, because the days that fail the floor are trend-absent days where the three-phase structure has nothing to build on — and the trend-qualified days that do clear the floor (09-10 SPY −0.66%, ADX 41.9; 09-08 −0.42%, ADX 22.1) then die at pullback_resumption. Blocking at the floor and blocking at the resumption are the same structural absence measured at two different stages. Regime, no fix, no TEMPER conversation warranted on the floor. This is SURGE's eleventh consecutive no-trade session on the current stretch (every session since the 08-31 stop-out), and the forecast side remains its well-calibrated side: p_trade 0.15 / direction short / conviction low → Brier 0.0225, and the plan's stated modal outcome (sub-floor first hour) is literally what printed.

Cross-Member Synthesis

Narrative finding: the desk's shared story for three weeks has been "the repricing tape closes the surface for both Family 2 expressions." Half of that story broke today. 09-14 and 09-15 were the same session character (two-tier, tech-down/energy-bid, sub-ATR, FOMC-eve), yet on 09-14 ECHO's XLE blocked at vwap_persistence with price below VWAP, and on 09-15 the bid persisted, participation completed, and the whole stack opened. So the surface is not closed — it is conditionally open, and the condition (does the energy program complete by 15:20?) is exactly the one the desk has been watching for two weeks. The desk's protection during the 18-session no-trade stretch was a closed gate, not a good plan; the day the gate opened, the plan was the only defense and it is not enforced. For SURGE the opposite is true — its gates are wide-open on this tape and the structure never completes, which is why its no-trade streak is structural rather than protective. The two members' no-trade streaks look the same in the P&L and mean opposite things; that distinction is the finding of this period.

Open TEMPER Items (Carried Forward)

• ECHO — plan/engine enforcement gap: ESCALATED this period. Now N=3 same-shaped (08-19 XBI, 08-20 XBI, 09-15 XLE) and the first instance where the plan named instrument + mechanism + direction + loss template in advance. On 09-14 the desk was defended by a 0.021 vol_ratio graze; on 09-15 nothing defended it. This clears the Desk Review's N ≥ 3 triage bar outright — flag as a candidate for the next Desk Review, sizing-(b) candidate (a new mechanism: gate-level enforcement of plan stand-downs), not a parameter recalibration. No gate touched.
• ECHO — exhaustion_score binding-gate watch. New data: 1.6913 (loss) vs 1.7405 (win) on the same instrument, same direction, same driver. Monitoring only — T=0.80 is in-sample and must not be re-optimized. Flagged for learnings.md.
• ECHO — XBI exclusion review (WR < 50% over 15+ live trades): sub-book n=2, WR 0.0% — NOT triggered. Carried.
• SURGE — session-character / catalyst-gap filter conversation (escalated 08-31, stand-alone THL): open. This period adds a back-to-the-floor rotation rather than new evidence; 09-10's mechanism-gate-confirmed read (the resumption precondition, not the signal floor, is binding) still stands as the strongest data point. Carried as sizing-(a)-shaped to the next Desk Review.
• SURGE — first-hour floor question: explicitly closed for now — the 09-05 deliverable already showed a looser floor converts none of the blocked sessions. Do not re-open on graze counts alone.
• SURGE — plan/engine enforcement gap (shared with ECHO): dormant — every stand-down honored. Carried open.
• SURGE — event-day skip diagnostic (resolved 2026-07-17, MIXED); 2023 PF 0.95 watch; ADX re-tune watch (moderate-trend 22–30 outperforms strong-trend ≥30): all carried, no new data this period.
• Plan-scoring observation (desk-level, for TEMPER/score_plans.py): ECHO's p_trade 0.30 was the engine-fire probability, it was correct (engine fired, direction hit), and the desk lost $64.98 on it (Brier 0.49). While the plan refuses trades the engine still takes, honest engine-fire forecasting and desk P&L will keep diverging. Not a scoring bug — a design tension worth one line in the next plan-calibration review.

Desk Summary for Orchestrator

Desk-cadence reflection covering 09-14 and 09-15 — desk net −$64.98, one trade (ECHO's XLE short). ECHO traded for the first time since 09-08 and lost: XLE short 760, 65.785 @15:30 → stopped 65.8705 @15:55, −$64.98. The gate read was a full clearance — the first in six weeks (signal +1.3637%, vol_ratio 1.2402, ADX 21.52, exhaustion_score 1.6913) and it fired on the one candidate the plan pre-refused — fading the driver's own strengthening energy sector, XLE's pre-committed loss template — as Brent closed at a four-month high and XLE trended into the close. Binding constraint was not a gate; it was the plan/engine enforcement gap, now N=3 same-shaped (08-19, 08-20, 09-15); cumulative −$148.53 (5W/5L, WR 50.0%), SPRT CONTINUE (LLR +0.230, no boundary risk). 09-14 was a no-trade with a 0.021 vol_ratio graze on SPY. SURGE did not trade either session — 09-15 blocked at the first_hour_signal floor on both legs (SPY −0.25%, QQQ −0.17% vs ±0.30%), the stated modal outcome; regime, no fix (the 09-05 deliverable already showed a looser floor converts nothing); 11th consecutive no-trade session; cumulative +$482.98, SPRT CONTINUE (LLR +1.131). Two flags for learnings.md: the N=3 plan/engine enforcement pattern, and the exhaustion composite's failure to discriminate ECHO's two live driver-fades (1.7405 win vs 1.6913 loss). Next: FOMC tomorrow (Wed 09-16, 2:00 PM ET) — decision + SEP + Warsh presser; neither stack can carry it (ECHO flat 15:58, SURGE flat 14:00).