House Mneme — 2026-09-18

desk: house-mneme date: 2026-09-18 forecasts: mneme: p_trade: 0.20 direction: long conviction: low trace: p_trade: 0.12 direction: long conviction: low

Desk Plan — House Mneme — 2026-09-18

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-09-18-pm.md PM: MNEME (desk-blind: this desk's own members only; GHOST stays frozen, no plan)

Shared Market Read

• Event risk today: LOW. No Tier-1 event within the next 7 trading days (through 09-25); next Tier-1 beyond the window: PCE Sep 30, NFP Oct 2, CPI Oct 14. Wednesday's FOMC (unanimous 25bp hike to 3.75–4.00%, Warsh signalling at least one more) is in the past. Firm-wide status GO — neither active member hard-skips today. Today's only scheduled reads are second-tier: Industrial Production 9:15, Bowman 9:30, CB LEI 10:00, Schmid ~11:45 — none a market-binary.
• The structural fact of the session — quadruple witching. Quarterly index futures (opening-based settlement), index options and single-stock options all expire; Citadel Securities estimates ~$6.2tn of US options exposure expiring (some desks cite up to $9.6tn). Firm inputs are explicit and I am adopting them verbatim as a mechanical constraint: elevated-volume flow concentrated around the open and the close; do not read a choppy open as directional; structure may not form cleanly until after the first 30 minutes.
• Session character expected: Mechanically-driven / mixed — a modestly risk-on, tech-led open into expiry flow. Futures point firmer with the Nasdaq complex leading and the Dow/large-cap-value leg lagging. Whether the semis bid (Huang's "chip volume roughly doubles next year" comment, Asian chip rally) survives the open is the first tell; the second is whether the 10-yr (held ~4.94% after losing ~10bp, off its 5.04% 2007-high re-test) stays off 5% — that is the switch that decides whether the tech relief extends.
• VIX regime: 15.27 — LOW (<18) (−0.17 vs prior close 15.44). LOW VIX is each active member's best-backtested regime (MNEME VIX<20 PF 2.39; TRACE VIX<20 PF 2.46). VIX is not a gate input for either stack, but a LOW print on a GO day means the tape is tradeable in principle — vol does not block; structure quality does.
• Key levels: SPY prior close 762.64, prior range 759.98–763.57, SMA20 764.31 (still below the 20-day — yesterday's +1.1% gap-and-hold has not reclaimed it), ATR(14) 6.66. QQQ prior close 716.89, range 713.32–718.03, SMA20 712.73 (above), ATR(14) 8.42 — note yesterday's SPY range was only 0.54× ATR despite the +1.1% move: a gap-up-and-hold, not a trending session. DIA is not in the briefing's price block — confirm the gap at the open for framing only. Cross-asset: WTI back under $100 (third down session, Saudi East-West pipeline restoration), gold firm near $4,390–4,420, silver sharply higher, 10-yr ~4.94%, 30-yr ~5.28%.
• Macro backdrop: Post-Fed relief plus a genuine AI-demand catalyst, layered on a second central-bank tightening (BOJ to a 31-year-high 1.25%, yen weaker anyway), a still-elevated long end, and a live Saudi–Houthi headline channel into oil.

The desk-level structural read — why today is a reduced-fidelity day for both of my members, not a blocked one

Quarterly expiries are in both stacks' libraries (the 2020–2024 history contains ~16 quarterly witching Fridays per instrument, none of them in the live blackout set beyond any that coincide with FOMC/CPI/NFP/PCE). So witching is not a hard skip and I am not inventing one. But the mechanics land on the two specific places my members' signal machinery is most exposed:

1. Window A's observation vector is the mechanically distorted one. A observes 9:30–10:00 (6 bars) and enters at 10:00 — precisely the window firm inputs flag as mechanically noisy and potentially stale, with index futures settling on the opening print. A vote that clears 62% off a mechanically-distorted 6-bar fingerprint is a lower-fidelity match, not a better one. This matters disproportionately for MNEME: A produced 5 of my stack's last 6 live signals (08-17, 08-21, 08-24, 08-31, 09-08), with 08-27 the only B fire in that stretch.
2. Window B's exit lands in the distorted close. B hard-flats at 16:00 — into the witching closing auction, the session's heaviest mechanical flow and where prints can be pinned. An exit there adds mechanical noise to the trade's outcome independent of whether the pattern read was right. B is MNEME's anchor window (holdout PF 3.76) and the window I'd otherwise most trust.

Windows C (both stacks) and A (TRACE) are less directly damaged, but C is flagged on both stacks (MNEME holdout PF 1.25; TRACE holdout PF 1.13) and I do not lean on it.

Net: neither stack is structurally incompatible with today — no element of either gateset says "skip" — but the session's mechanics degrade the fidelity of the exact windows that have been producing this desk's signals. Both stacks are also, separately, in a long sub-bar stretch: MNEME has not fired in 7 sessions (last: 09-08 Window A, vote 80%, scratch) and TRACE has not fired in 7 sessions (last: 09-09 Window C, vote 70%, −$12.78 hard-flat). MNEME's binding gate is the vote threshold, not similarity — over the last 8 sessions its sims ran 0.44–0.92 against a 0.50 floor while its votes ran 27–60% against the 62% gate, with three 60%/62% grazes (09-01, 09-09, 09-15). TRACE's is the same gate with more room: every recent vote 30–60% vs 62%, similarity above its 0.60 floor on most days.

Forecasts are therefore set below each stack's recent fire rate (MNEME ~25% over 24 sessions; TRACE ~8% over 24 sessions — the recent run-rate for both is 0/7) rather than at it, because the witching-fidelity handicap is a genuine, specific negative that I can name. I am not writing a hedged 0.5, and I am not writing 0.0 either: a real fire is possible on a mechanically clean post-10:00 continuation, and a 1-in-5 and a 1-in-8 are the honest numbers.

MNEME — SPY KNN (p_trade 0.20, long, low)

Live and eligible — but my recent signals come from the one window today's mechanics damage most, so the base case is another sub-bar stand-down. No hard skip (FOMC was Wednesday, not today); LOW VIX is my best regime (VIX<20 PF 2.39); my mechanism is healthy (SPRT CONSISTENT-WITH-BACKTEST — 23 trades, 47.8% WR, +$1,041.36, LLR +3.793). Setup present: not cleanly — possible, not likely. The binding gate has been my vote (≥62%), not max_sim: over 09-09→09-17 my sims ranged 0.44–0.91 versus the 0.50 floor, while votes ran 27–60% — three grazes at exactly 60%/62%. What would need to align today is a window whose vote clears 62% on a tape that has visibly established a directional continuation after the settlement dust settles (i.e. past 10:00, not on the open's mechanical prints). Window A is the window I trust least today even though it has fired 5 of my last 6 signals — a 6-bar vector built on 9:30–10:00 witching flow is the lowest-fidelity match available to me. Window B (anchor) is the one I'd trust if a clean post-noon continuation forms, with the caveat that its 16:00 hard flat exits into the witching auction — a mechanically noisy exit, which is an outcome-noise fact I accept rather than a reason to override the gate.

Sizing: standard 0.75% ($187.50 on $25K) — no adjustment. Event risk is LOW with no Tier-1 for 7 trading days, so there is no event-risk reduction to make; conviction is low, so there is no reason to increase. MAX notional $75K unchanged.

Key levels: SPY prior close 762.64; prior range 759.98–763.57; SMA20 764.31 — SPY is still below its 20-day despite yesterday's +1.1%, so a reclaim of 764.31 is the structural tell that the relief has legs. Holding 759.98 (prior range low) keeps the relief bid intact; a break of 759.98 with the 10-yr pushing back toward 5% is the fade tape. ATR(14) 6.66 → a 1.5× ATR stop is ~10 index points.

Invalidation: if SPY gaps >1.5% at the open (relief overrun), I stand down entirely. If the 10-yr re-tops 5.00% and snaps the semis bid, I stand down entirely. And the honest disclosure: a Window A fire generated on the mechanically distorted 9:30–10:00 vector, with no post-10:00 continuation to confirm it, is a fire I would not have chosen — I will log it as it happens and flag it in the reflection rather than pretend the gate was wrong. No gate is amended today: nothing here touches v5's live parameter set.

TRACE — DIA KNN (p_trade 0.12, long, low)

Live and eligible — but the setup has been structurally absent for 7 sessions, DIA is the lagging leg again, and my DEGRADED clock is one qualifying fire from deciding its path. No hard skip. LOW VIX is my best regime (VIX<20 PF 2.46); my 09-05 fetch_today_sip_embargo fix has delivered 3-for-3 full three-window reads and should read cleanly today. Setup present: unlikely. My binding gate is the vote (≥62%) and I have not been close to it: every recent vote ran 30–60% against the 62% gate (≥2pp–32pp sub-bar) while similarity sat above my 0.60 floor on most days (0.80–0.89 on 09-14/09-15 A-window). Genuine similarity, decisive-vote threshold never reached — the setup never came close. Two fires in the last 24 sessions (08-27 stop, 09-09 Window C −$12.78 hard flat), none in the last 7. Today's bid is explicitly tech-led with the Dow/large-cap-value leg lagging — my instrument is, again, the fragile leg, not the leader.

Sizing: standard 0.75% ($187.50 on $25K) — no adjustment. The DEGRADED clock is explicitly not a sizing input (it neither suppresses nor forces a legitimate fire and does not change position size), and LOW event risk with no Tier-1 means there is no reduction to make.

Key levels: DIA is not in the briefing's price block — confirm the open print for framing only (my last live entry was 09-09 at 524.85). DIA must hold its prior-day low to keep any continuation alive; watch whether the Dow's energy members (XOM/CVX) offset its yield-sensitives with the 10-yr at ~4.94% — that offset is what would let DIA participate in a tech-led day at all. Note the same witching caveat as the desk read: a DIA Window A read (TRACE's own A fires exist) is built on the distorted open window, and B's 16:00 exit lands in the auction.

Invalidation: if DIA gaps >1.5% at the open OR the semis bid fades and the 10-yr re-tops 5.00% (the Dow's rate-sensitives take the brunt), I stand down entirely.

The live watch — DEGRADED clock. LLR holds at −2.057, −0.888 from DEGRADED (−2.944). The pre-registered desk-PM-review-within-1-session / TEMPER-within-5 protocol remains near-activated and the next qualifying DIA fire's outcome decides the path — a win relieves the clock, a stop-out advances it toward DEGRADED. A quadruple-witching session is a poor place for that deciding trade to land (outcome noise independent of read quality), which is one more reason my honest p_trade is low — but if a qualifying fire comes, it is taken as the gates dictate and the clock records the outcome. No gate is amended today.

Pooled note (Family 5)

Both active members are LIVE and eligible on a firm-wide GO day with LOW event risk, no Tier-1 for 7 trading days, and LOW VIX — each stack's best-backtested regime. Neither is structurally incompatible with the session; there is no reason in either gateset to write "no trade expected." But this is a reduced-fidelity day for the specific windows that generate this desk's signals: witching mechanics put mechanical flow into the 9:30–10:00 observation window (Window A — MNEME's dominant recent firing window, 5 of its last 6 signals) and into the 16:00 closing auction where Window B (MNEME's anchor) hard-flats. Both stacks are also cold: 0 fires in the last 7 sessions each, with the vote gate as the binding constraint on both. Forecasts: MNEME 0.20 long low / TRACE 0.12 long low — below each stack's recent fire rate, because the fidelity handicap is real, named, and specific to today. If both fire same-day they count as one clustered observation for pooled significance (same family, same-day fires across instruments) — low probability today. Both monitoring clocks sit live and unchanged: MNEME SPRT CONSISTENT (LLR +3.793); TRACE DEGRADED clock −0.888 from boundary. GHOST stays frozen (Phase 2, Decision D2) — no plan, no forecast. Next Tier-1: PCE Sep 30.

Plan Filed

• Filed: 2026-09-18 04:35 PT (07:35 ET, pre-market)
• Frontmatter forecasts complete for every active member: yes (mneme 0.20/long/low; trace 0.12/long/low)
• GHOST: FROZEN (Phase 2, Decision D2) — not planned for, no member file written
• Firm-wide status: GO — both active members LIVE (no hard skip; event risk LOW, VIX 15.27 LOW); quadruple witching is a mechanical-flow fidelity handicap, not a skip

House Mneme Desk Reflection — 2026-09-18

Plan reference: desks/house-mneme/plans/2026-09-18-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-09-18-eod.md

Cadence Rule Header

Weekly/same-day-on-trade cadence. Today is Friday 2026-09-18 — a quadruple-witching session (quarterly index-futures + index/single-stock options expiry, ~$6.2tn US options exposure per Citadel), the first day either member has traded since the 09-11 reflection. This reflection therefore covers the full period since 2026-09-11: Mon 09-14, Tue 09-15 (LIVE, GO — both stood down), Wed 09-16 (FOMC Tier-1 hard skip), Thu 09-17 (LIVE, GO — both stood down), and Fri 09-18 (today) — where both members FIRED (a same-day co-fire). GHOST remains frozen (Phase 2, Decision D2) — no reflection.

What Happened

The desk's 7-session cold streak ended — both members fired today, a same-day co-fire that counts as ONE clustered observation for Family 5 pooled significance. Period outcome: MNEME WIN +$5.39 (Window B scratch), TRACE LOSS −$88.45 (Window A stop-out). Combined desk P&L this period −$83.06 on 2 trades.

The stand-down half of the period was regime-clean for both members. 09-14 / 09-15 / 09-17 were all no_trade:knn_gate — every session stood down on the vote threshold (≥62%), with similarity generally above each stack's floor (MNEME sim 0.437–0.89 vs 0.50; TRACE sim 0.528–0.892 vs 0.60) but no decisive vote. The one near-graze (09-15, MNEME A:60% and B:60%, both 2pp sub-bar) was a double-window same-day graze, not a repeated single-threshold pattern. 09-16 was a clean FOMC Tier-1 hard skip (firm-wide no-GO; the unanimous 25bp hike to 3.75–4.00% was correctly avoided).

Today (09-18) was the fire day. On a firm-wide GO day, LOW event risk (no Tier-1 for 7 trading days; second-tier IP/LEI prints non-moving), LOW VIX (15.27→14.81), and a tech-led two-tier session (QQQ +0.62% above SMA20; SPY −0.13% below SMA20), both members produced genuine, decisive signals and both fired long:

• MNEME fired Window B (anchor, holdout PF 3.76) at 12:00 — vote 80%, max_sim 0.6018 (above the 0.50 floor). Entered 759.0, stopped out at 759.06 @ 12:20, +$5.39, win=True. A 20-minute scratch win: entered below the SMA20 (764.26) on a broad-flat tape, and the stop filled slightly above entry. The plan's witching handicaps did NOT materialize destructively — MNEME did not fire Window A (the distorted open window it has relied on for 5 of its last 6 signals); it fired the anchor on the post-noon tape and was not exposed to the 16:00 closing-auction exit (stopped at 12:20, well before it).
• TRACE fired Window A at 10:00 — vote 70%, max_sim 0.8622 (well above the 0.60 floor). Entered 514.97, stopped out at 514.36 @ 10:45, −$88.45, win=False. This is a Window A stop-out on the flagged lowest-fidelity window — the exact window the plan identified as built on mechanically-distorted 9:30–10:00 witching flow. DIA (the Dow/value leg) closed as the plan pre-assigned it — the fragile leg on a tech-led day (SPY −0.13%, broad index below SMA20) — and TRACE caught that weakness. It is also its sixth Window A stop-out (07-08, 07-13, 07-20, 07-23, 07-27, 09-18) in 10 total trades.

The period's load-bearing read is a split: for the GATE, both members are REGIME; for the FORECAST, the desk underweighted fire probability on a day that both members traded.

• Gate (regime, both): The binding gate all period was the vote threshold. On the stand-down days (09-14/15/17) it blocked with sub-bar votes that never formed a same-shaped N=3 repeated close-graze against one threshold in one window. MNEME's only near-misses (09-15 A:60/B:60) were a single double-graze. When the vote genuinely cleared (09-18: MNEME 80%, TRACE 70%), the gate FIRED as designed. Neither stack's near-miss data says "repeatedly close against one threshold" — it says structurally sub-bar until a decisive pattern formed. No gate is being unilaterally tweaked.
• Forecast (calibration concern, both): The plan forecast p_trade MNEME 0.20 / TRACE 0.12 (low, low) on the "reduced-fidelity witching + 7-session-cold" reasoning. Both fired, so both Brier scores are the desk's worst of the recent stretch — MNEME 0.64, TRACE 0.7744 — versus the 0.0–0.09 the desk has posted on its stand-down days. A GO + LOW-VIX + no-Tier-1 day with both mechanisms genuinely live made a same-day co-fire a more likely outcome than 0.20/0.12 implied. This is a forecast over-conservatism miss worth an honest self-critique and a learnings flag — it is not a gate miscalibration, and nothing about it touches execution.

Combined desk P&L this period: −$83.06 on 2 trades (MNEME +$5.39 win; TRACE −$88.45 loss). Desk cumulative now 34 trades, +$596.31 (MNEME 24 / +$1,046.75; TRACE 10 / −$450.44).

Paper P&L

Desk P&L this period: −$83.06 (2 trades) MNEME cumulative Phase 3: 24 trades, +$1,046.75 net, SPRT CONSISTENT-WITH-BACKTEST TRACE cumulative Phase 3: 10 trades, −$450.44 net, 10% WR (1W/9L), SPRT CONTINUE (LLR −2.41, −0.534 from DEGRADED) — protocol ACTIVATED Combined desk cumulative: 34 trades, +$596.31 net

Event Risk vs. Expectation

• 09-14/15/17 (LIVE, GO — LOW) — correct. No Tier-1; both members stood down on sub-bar votes with proper full window reads. The 09-05 fetch_today_sip_embargo fix delivered full three-window reads on all three live sessions (TRACE's B/C windows no longer no_data), extending the clean observability run.
• 09-16 (FOMC Tier-1, firm-wide NO-GO) — hard skip correct. Unanimous 25bp hike to 3.75–4.00% (Warsh signalling further tightening); SEP + presser. Both members' clean FOMC-skip rows (07-29, prior) re-confirmed; Brier 0.0 both.
• 09-18 (LOW event risk, GO — correct rating). No Tier-1; second-tier IP (≈flat vs +0.3% cons) and CB LEI ("edged down") both non-moving, as the plan forecast. The dominant variable was the quadruple-witching mechanics — contained (both sub-ATR, SPY 0.72× / QQQ 0.79× ATR), mechanically noisy, exactly as the AM pre-assigned.
• Session character (two-tier, tech-led, contained — correct). QQQ +0.62% (above SMA20, firm +0.26% last hour) while SPY −0.13% (below SMA20). The Dow/value leg lagged as the plan pre-assigned — and that pre-assignment is exactly what TRACE's loss paid for. The plan's "don't read a choppy witching open as directional" discipline held (SPY flat vs open).
• Invalidations — none triggered. No >1.5% open gap for either SPY/DIA; 10-yr stayed off 5% (~4.9x); VIX held LOW (15.27→14.81) and even fell.
• Forward risk: no Tier-1 within 7 trading days. Next Tier-1s: PCE Sep 30, NFP Oct 2, CPI Oct 14. The weekend carries the unresolved Saudi–Houthi channel into oil (WTI back under $100) and the still-weak yen — both contained today but live moving parts into PCE week.

Per-Member Binding Gate Analysis

MNEME — binding gate: the vote threshold (≥62%). REGIME for the gate (fired when a decisive pattern formed and won), but the p_trade forecast was over-conservative (Brier 0.64).

MNEME stood down on knn_gate on 09-14 (A:47/B:53/C:53), 09-15 (A:60/B:60/C:40 — the near-graze), 09-17 (A:53/B:33/C:33), hard-skipped 09-16 (FOMC), and fired Window B on 09-18 (vote 80%, max_sim 0.6018) → +$5.39 win.

• The gate is the vote threshold, and it behaved per design. All period, max_sim cleared or near-cleared the 0.50 floor while the decisive-vote threshold was the blocker. The near-miss reads: 09-15 showed a same-day double-window graze (A and B both 60% vs 62%) — but that is two windows grazing once, not one threshold grazed repeatedly (the N=3 same-shape signature of calibration trouble). No unilateral tweak to locked v5 gates.
• Crucially, MNEME's fire came from Window B (anchor), not Window A. The plan's central concern was that MNEME's recent signals (5 of its last 6) come from Window A — the mechanically-distorted 9:30–10:00 observation window on a witching day. Instead, MNEME's first fire in 7 sessions was the anchor window (holdout PF 3.76) on the post-noon tape, and it exited via a 12:20 stop — never exposed to the flagged 16:00 closing-auction flat. The two windows the plan flagged as the reduced-fidelity risks (A observation, B exit) both resolved benignly for MNEME. The Window A thread that dominated recent-fires was not the window that fired.
• SPRT stays CONSISTENT-WITH-BACKTEST (24 trades, +$1,046.75) — a win, no drawdown stress.
• Forecast self-critique (the real finding): p_trade 0.20 → traded → Brier 0.64, the desk's worst in this stretch. Over 09-14→09-18 MNEME's votes ran 33–80%; the day a post-noon continuation actually formed (strong B vote), the mechanism was live and fired. Underweighting fire probability on a GO/LOW-VIX/no-Tier-1 day is a forecast-calibration miss worth carrying — it does not indicate the gate is wrong.

Verdict: REGIME (gate) — the vote threshold was the binding gate and it worked; the setup was absent until 09-18, when it fired and won. The p_trade forecast was over-conservative (calibration note, not gate evidence).

TRACE — binding gate: the vote threshold (≥62%). REGIME for the gate (sub-bar votes blocked 09-14/15/17; a genuine decisive vote fired 09-18), but the fire was a Window A stop-out on the flagged lowest-fidelity window, and the DEGRADED clock is now ACTIVATED at −0.534 from boundary.

TRACE stood down on knn_gate on 09-14 (A:50/B:20/C:60), 09-15 (A:50/B:60/C:50), 09-17 (A:60/B:50/C:30 — 60% near-graze), hard-skipped 09-16 (FOMC), and fired Window A on 09-18 (vote 70%, max_sim 0.8622) → −$88.45 stop-out loss.

• The gate behaved per design. The near-miss data is regime: votes ran 20–60% against the 62% gate on stand-down days while similarity sat above the 0.60 floor on most windows — the setup was structurally absent, not repeatedly grazing a single threshold in a single window. Today a decisive vote (70%) formed on genuinely high similarity (0.8622) and the gate fired. TRACE's knn_gate block is not mis-calibrated.
• The loss is a session-character / window-quality read, and it is TRACE's defining pattern. This is the sixth Window A stop-out: the DIA long fired at the 10:00 open on a 9:30–10:00 observation block that, on witching day, carried mechanically-distorted flow — the exact "lowest-fidelity window" the plan flagged. DIA closed as the fragile leg (SPY −0.13%, broad index below SMA20), and TRACE caught it (−88.45). TRACE's Window A fires keep losing (07-08, 07-13, 07-20, 07-23, 07-27, 09-18 are A stops; its only win was an A HARD_FLAT on 07-09). That is not a single-day graze — it is accumulating same-shaped evidence about TRACE's Week-A window quality, which is a TEMPER conversation, never a unilateral gate tweak.
• DEGRADED clock — now ACTIVATED. At 9 trades LLR was −2.057 (−0.888 from DEGRADED). Today's stop-out advanced it to LLR −2.41, −0.534 from DEGRADED (−2.944), at 10% WR (1W/9L). The pre-registered desk-PM-review-within-1-session / TEMPER-within-5 protocol is now triggered. This is the desk's highest-priority live item. It is explicitly not a sizing or gating input (standard 0.75% sizing unchanged; it neither suppresses nor forces a fire) — it is an escalation clock, and today's loss moved it to the edge.
• Observability held: full three-window reads on all live sessions this period (09-14/15/17/18), extending the post-09-05-fix clean run. The data path is healthy; the DEGRADED pressure is an outcome-quality signal, not a data artifact.

Verdict: REGIME (gate) — the vote threshold was the binding gate all period and fired legitimately on 09-18; the loss is a flagged-window/lagging-leg session read, but the DEGRADED clock is now ACTIVATED (−0.534 from boundary) and the Window A stop-out pattern is the TEMPER conversation to have. No gate tweak; escalation protocol triggered.

Cross-Member Synthesis

Narrative finding — the cold streak broke with a same-day co-fire whose two halves are cleanly split. MNEME's fire validates the gate: after 7 sessions of sub-bar votes, a decisive post-noon continuation formed on the anchor window and it won — the mechanism works. TRACE's fire re-validates its flagged-window exposure: it fired on the mechanically-noisy morning window on the day its instrument was the structurally-weak leg, and the stop-out advanced the DEGRADED clock to within 0.534 of the boundary. The desk forecast a low probability of ANY fire (0.20/0.12) and got a double — a genuine over-conservatism miss. Both members remain LONG-only, same-family, same-day → this co-fire is one clustered observation for the Family 5 pooled read. Combined desk cumulative: 34 trades, +$596.31. TRACE's DEGRADED clock is the defining live item.

Open TEMPER Items (Carried Forward / New)

• TRACE — DEGRADED-clock protocol now ACTIVATED (NEW, escalated). LLR −2.41, −0.534 from DEGRADED (−2.944) after the 09-18 stop-out (10% WR, 1W/9L over 10 trades). The pre-registered desk-PM-review-within-1-session / TEMPER-within-5 protocol is now triggered. Highest-priority live item; escalate to TEMPER per the protocol. It is a monitoring/escalation clock, not a sizing or gating input.
• TRACE — Window A stop-out pattern (NEW, N=6). Six of TRACE's 10 trades are Window A fires; all six stopped out (07-08, 07-13, 07-20, 07-23, 07-27, 09-18); its only win was an A HARD_FLAT (07-09). This same-shaped pattern (binding window = A, direction of miss = stop-out) is now at N=6 and worth a TEMPER conversation about whether TRACE's Week-A observation window (9:30–10:00, 6 bars) is too thin/noise-sensitive relative to its K=10 / max_sim≥0.60 design. Propose, never execute — no unilateral gate change.
• TRACE — Window C suspension monitor. Rolling-20 Window C PF < 1.2 monitor open (last C fire 09-09 −$12.78). No C fire this period (TRACE fired A today).
• MNEME — forecast over-conservatism on GO/LOW/no-Tier-1 days (NEW calibration note). p_trade 0.20 (MNEME) / 0.12 (TRACE) on a day both fired → Brier 0.64 / 0.7744, the desk's worst of the stretch. Flag for learnings: on a GO/LOW-VIX day with no Tier-1 within 7 days and both mechanisms live, an over-conservative p_trade underestimates the co-fire probability. Plan-calibration note for the scoring conversation, not a gate issue.
• MNEME — Window A thread (benignly resolved this period). The plan's central witching concern (A dominates MNEME's recent fires, and A is the distorted window) did NOT materialize — MNEME fired the anchor (B). Window A held (no fire). The A-window dominance of recent fires (5 of last 6 pre-09-18) is not itself a defect while B (anchor) is the one that fires on the strongest days; monitor the A/B mix, no action.
• MNEME — SPRT. CONSISTENT-WITH-BACKTEST, sticky (LLR +3.793, 24 trades, +$1,046.75).
• Family 5 pooled validation. Today's MNEME+TRACE same-day co-fire counts as one clustered observation for the pooled significance read. Pooled cumulative: 34 trades, +$596.31.

Desk Summary for Orchestrator

House Mneme's 09-18 reflection (covering 09-14, 09-15, 09-17 live stand-downs + 09-16 FOMC skip + 09-18 co-fire since the 09-11 reflection) broke the desk's 7-session cold streak with a same-day co-fire. MNEME (SPY, FIRED — WIN +$5.39): Window B (anchor) fire, vote 80%/sim 0.6018, entered 12:00, scratched at 12:20. Binding gate all period = vote threshold (REGIME) — sub-bar votes blocked 09-14/15/17 (single 09-15 double-window 60% graze, not an N=3 pattern), and the gate fired correctly when a decisive post-noon B pattern formed. Notably did NOT fire the flagged-window A; SPRT CONSISTENT-WITH-BACKTEST (24 trades, +$1,046.75). Forecast over-conservative (Brier 0.64). TRACE (DIA, FIRED — LOSS −$88.45): Window A fire, vote 70%/sim 0.8622, entered 10:00, stopped 10:45. Binding gate = vote threshold (REGIME) on the stand-downs, but the 09-18 fire was a Window A stop-out — the 6th in 10 trades — on the flagged mechanically-noisy morning window on the day DIA was the pre-assigned lagging leg. DEGRADED clock now ACTIVATED: LLR −2.41, −0.534 from boundary — pre-registered desk-PM-review/TEMPER-within-5 protocol triggered; escalate. Pooled: today's co-fire = one clustered observation; combined desk cumulative 34 trades, +$596.31. Learnings flag: GO/LOW-VIX/no-Tier-1 days were underforecast (Brier 0.64/0.7744) — the desk's worst of the stretch. Next Tier-1: PCE Sep 30; both members live until then.

Reflection Filed

• Filed: 2026-09-18 (EOD reflection; covers 09-14, 09-15, 09-16, 09-17, 09-18 since the 09-11 reflection)
• Next session event risk: Mon Sep 21 — no Tier-1; next Tier-1s PCE Sep 30 / NFP Oct 2 / CPI Oct 14. Weekend carries the Saudi–Houthi → WTI channel and the weak yen — both contained today, live moving parts into PCE week.
• GHOST: frozen (Phase 2, Decision D2) — no reflection written.
• MNEME and TRACE per-member compatibility files written (agents/mneme/reflections/daily_2026-09-18.md, agents/trace/reflections/daily_2026-09-18.md).
• Learnings flag (desk-blind, own members only): the desk underforecast a same-day co-fire on a GO/LOW-VIX/no-Tier-1 day (p_trade 0.20/0.12 → Brier 0.64/0.7744). TRACE's DEGRADED clock is ACTIVATED (LLR −2.41, −0.534 from boundary) and its Window A stop-out pattern is at N=6 — both are TEMPER conversations (escalation and window-quality, respectively), never unilateral gate edits.