House Mneme — 2026-07-24

desk: house-mneme date: 2026-07-24 forecasts: mneme: p_trade: 0.45 direction: long conviction: low trace: p_trade: 0.35 direction: long conviction: low

Desk Plan — House Mneme — 2026-07-24

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-24-pm.md PM: MNEME

Shared Market Read

• Event risk today: MEDIUM — S&P Flash PMIs (9:45 AM), New Home Sales (10:00 AM), plus continued post-GOOGL/TSLA/INTC earnings digestion. The Intel beat (+5.2% after-hours, EPS $0.42 vs $0.21) provides a sector-level positive counterweight, but the broader tech repricing (AI capex concern) and the oil/rate feedback loop (10yr at 4.7%, Brent near $100) remain unresolved.
• Session character expected: Fade-or-follow-through — Intel-driven positive open may fade as the broader tech headwinds reassert. Friday lower volume typically favours range-bound / pattern-repeating sessions over clean trend days. The primary question: was Thursday's sell-off a one-day event or the start of a broader correction?
• VIX regime: 18.84 (MID — 18-28). Both MNEME and TRACE are profitable in this regime (MNEME VIX 20-30 PF 3.54; TRACE VIX 20-30 PF 2.12). The LOW-to-MID transition at 18.70 is the first regime change since mid-July; this is within the normal operating range for both stacks.
• Key levels: SPY prior close 738.06, prior range 735.25–742.51, SMA20 745.87 (below SMA20 — structural shift). DIA prior close 520.13. The two-tier divergence (QQQ underperforming SPY) widened on Thursday — this is the dominant structural signal. SPY testing below SMA20 for the first time this month.
• Macro backdrop: The week's earnings-driven sell-off meets a Friday session with macro data (PMIs, New Home Sales) and continued earnings digestion. Intel's after-hours beat is the counter-catalyst, but its effect is concentrated in semiconductors — the broader market (SPY, DIA) is less directly affected. The FOMC decision on July 29 (5 days away) and PCE on July 30 are the next major macro catalysts.

MNEME — SPY (K=15, max_sim≥0.50, vote≥62%, long-only)

Live record: 8 trades, 50% WR, $+267.11 net P&L. SPRT: CONTINUE (LLR +1.477, well within boundaries). Last trade (2026-07-23): Window C, +$9.60 HARD_FLAT.

Setup assessment: Unlikely to below-expected. Today is a narrative-driver session (post-earnings digestion, two-tier divergence, Intel counter-catalyst, Friday macro data) — MNEME's historical pattern (per 2026-07-17 reflection) shows that narrative-driver sessions underperform clean post-data sessions for KNN similarity. The VIX 18.84 is at the LOW/MID boundary — MNEME's VIX<20 PF is 2.39 (still profitable, but the weakest VIX band). The SPY structural shift below SMA20 creates a regime that may not have strong historical analogs in the KNN database.

Key conditions for a signal:

• Window A (observes 9:30-10:00, enters 10:00): The first 30 minutes must resolve the Intel-driven open into a contained pattern, not a trending gap. A gap >0.8% that widens through 10:00 → stand down.
• Window B (observes 10:00-12:00, enters 12:00): The anchor window (holdout PF 3.76). If the post-PMI (9:45 AM) / New Home Sales (10:00 AM) data settles the session into a clean pattern by lunch, Window B is the strongest candidate. This is the most likely path to a trade today.
• Window C (observes 12:00-14:00, enters 14:00): Flagged (holdout PF 1.25). Pre-registered suspension at 15-trade rolling PF < 1.2 — not triggered. Only trades if it produces the strongest single-day vote. Low confidence.

Sizing: Standard — 0.75% ($187.50), 1.5× ATR(14) stop, single position per day.

Invalidation: SPY gaps >0.8% at open with widening trend through 10:00 → trending day, not pattern-repeating. VIX spikes above 20 before 10:00 → elevated volatility degrades pattern quality → stand down Window A, re-evaluate at 12:00 for Window B.

TRACE — DIA (K=10, max_sim≥0.60, vote≥62%, long-only)

Live record: 6 trades, 16.7% WR, $-206.93 net P&L. SPRT: CONTINUE but approaching DEGRADED boundary (LLR -0.996, boundary at -2.944). Last trade (2026-07-23): Window A STOP at -$98.60. The live track record is well below backtest expectations — the primary Phase 3 question.

Setup assessment: Unlikely. TRACE's tighter gates (max_sim≥0.60, K=10) produce fewer signals than MNEME, and the live track record shows the KNN is not finding clean matches in current conditions. DIA's 30-stock blue-chip composition provides partial insulation from the mega-cap tech overhang (DIA beta ~0.70), but the broad market weakness (SPY below SMA20) affects all indices. Friday lower volume may reduce pattern quality further.

Key conditions for a signal:

• Window A (observes 9:30-10:00, enters 10:00): DIA's pre-market level at approximately 520.13 (prior close). A contained gap (-0.3% to -0.5%) is the ideal KNN environment. The Dow's sector composition (financials, industrials, healthcare alongside tech) means the bar-by-bar vectors may be cleaner than SPY's today — the Intel chip beat primarily affects the semiconductor sector, which is a smaller component of DIA than SPY.
• Window B (observes 10:00-12:00, enters 12:00): The anchor window (holdout PF 3.00, 18 trades). The 24-bar, 72-dim observation is the richest feature space. If the post-PMI / New Home Sales session settles into a clean pattern between 10:00-12:00, this is the strongest candidate.
• Window C (observes 12:00-14:00, enters 14:00): Pre-registered for suspension at 20-trade rolling PF < 1.2 (holdout PF 1.13, 18 trades). Low confidence — only trades if it produces the strongest single-day vote.

Sizing: Standard — 0.75% ($187.50), 1.5× ATR(14) stop, single position per day.

Invalidation: DIA gaps >0.8% at open with unidirectional trend through 10:00 → trending day. VIX spikes above 20 → elevated volatility degrades pattern quality. The KNN database contamination diagnostic (~18-20% event-day neighbors) adds structural uncertainty on a MEDIUM event-risk session — the vote ≥ 62% gate is the corrective filter.

Plan Filed

• Filed: 2026-07-24 07:30 ET
• Frontmatter forecasts complete for every active member: yes

House Mneme Desk Reflection — 2026-07-24

Plan reference: desks/house-mneme/plans/2026-07-24-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-24-eod.md

What Happened

Family 5 took one trade today — MNEME's Window B long on SPY — and finished -$143.30 as the afternoon fade caught the midday rally. TRACE had no qualifying signal on DIA. The combined result erases the modest desk gain from 07-23 (+$9.60 from MNEME's Window C breakeven) and cuts into the cumulative buffer.

Market context: A textbook two-tier divergence session. SPY flat (+0.11%) while QQQ dropped another -1.11%, deepening the break below SMA20. Low realized vol on both (0.77-0.91× ATR). The day's structure was dominated by two crosscurrents: the oil-driven relief rally (WTI -4.3% on Iran diplomatic talk hopes) rotated capital into rate-sensitive value (real estate, materials, homebuilders), while AI-chip names continued to liquidate — INTC closed -7.9% on a beat, the third consecutive signal in this complex that even good news can't hold. The Michigan Consumer Sentiment final (49.5 vs 54.4 preliminary, released today at 10:00 AM ET) was a significant negative revision painting a deteriorating consumer backdrop. New Home Sales beat (628K vs 606K) and Services PMI (53.6 vs 51.5) provided some positive offset, but the macro picture was settled: the consumer is weakening, AI capex is repricing, and the market is rotating out of tech into value. VIX at 18.58 (MID regime), stable vs prior close of 18.70.

The session's key structural feature: the midday rally (SPY to 743.71) completely faded. By the close, SPY had drifted back to 738.90 — essentially unchanged on the day but down from the noon peak by 0.65%. This fade is what caught MNEME's Window B long entry.

Period Since Last Reflection (2026-07-23 → 2026-07-24)

Paper P&L

Combined desk P&L today: -$143.30 MNEME cumulative Phase 3: 9 trades, $+123.81 net, 44.4% WR, SPRT CONTINUE (LLR +1.208) TRACE cumulative Phase 3: 6 trades, -$206.93 net, 16.7% WR, SPRT CONTINUE (LLR -0.996) Combined desk cumulative: 15 real trades, -$83.12 net, 33.3% WR

Member Analysis

MNEME (SPY, Window B — anchor window, STOP loss)

Window B (Morning→Afternoon, anchor window — holdout PF 3.76) fired at 12:00 with a KNN signal. Entered SPY long at 743.04. The midday rally carried SPY to 743.71 — a 67-cent per-share peak — but the afternoon fade reversed the entire intraday move and more. Stopped at 741.61 (12:45) at 1.5× ATR. Net P&L: -$143.30.

The plan assessed today as "unlikely to below-expected" for MNEME (p_trade 0.45, conviction low), correctly identifying that narrative-driver sessions underperform clean post-data sessions for KNN similarity. The plan also noted: "today is a narrative-driver session (post-earnings digestion, two-tier divergence, Intel counter-catalyst, Friday macro data)." A trade did fire, and it lost — the narrative-driver assessment was correct even though the forecast probability was reasonable.

The mechanics are instructive. SPY rallied from the open (738.50) to the midday peak (743.71, +0.70%), driven by the oil-rotation rally in rate-sensitive names. The Window B observation period (10:00-12:00) captured this upward drift and the KNN found a continuation pattern. But the afternoon completely reversed the gain — SPY faded from 743.71 to 738.90 close (-0.65% from the peak), with the stop triggered at 12:45 at 741.61. The KNN correctly identified a pattern in the morning microstructure, but the oil-driven intraday catalyst faded in the afternoon as the broader AI-chip sell-off reasserted dominance. The two-tier divergence is an intraday force that the KNN's historical pattern library may not adequately weight — the historical analogs from 2020-2024 don't have the same persistent, session-spanning sector rotation dynamic.

This is the first losing Window B trade since 07-15 (also a STOP, -$143.30). Window B has a 4W/3L record in Phase 3 — still positive and consistent with its anchor-window status.

SPRT trajectory: LLR +1.208 (CONTINUE), +1.737 from CONSISTENT (+2.944), -4.152 from DEGRADED (-2.944). The loss pulls the LLR back from the 07-23 +1.477 high but remains comfortably within CONTINUE territory. The healthiest SPRT reading on the desk.

[Flag for learnings.md]: MNEME's Window B (anchor, holdout PF 3.76) was caught by an afternoon fade in a two-tier divergence session. The KNN pattern was directionally correct during the observation window but the intraday catalyst (oil rotation) faded as the tech sell-off reasserted dominance. This is the second losing Window B trade (now 4W/3L Phase 3) — the first time the anchor window has shown vulnerability in live data. The divergence between morning (observed by KNN) and afternoon (where the exit materializes) is a structural risk for any KNN strategy that observes A→B but executes B→C: the session's character can change after the signal fires. Worth tracking whether this divergence-risk concentrates on narrative/macro-driven sessions specifically.

TRACE (DIA, no trade — max_sim gate)

No qualifying KNN signal on DIA today. All three windows evaluated: none exceeded the max_sim ≥ 0.60 or vote ≥ 62% gate. The net was $0.00.

The plan assessed TRACE at p_trade 0.35, conviction low, direction long. The plan correctly noted: "TRACE's tighter gates (max_sim≥0.60, K=10) produce fewer signals than MNEME" and that "Friday lower volume may reduce pattern quality further." Both calls were accurate — DIA's session on a two-tier divergence day was dominated by the Dow's exposure to rate-sensitive sectors (positive via oil/pullback rotation) but tempered by the broader market's risk-off tone. DIA's 30-stock composition shielded it from the worst of the AI-chip carnage but didn't produce a clean enough bar-by-bar pattern for the KNN to find a high-confidence match.

The no-trade outcome is the correct gate behavior — the KNN correctly declined to trade when no pattern cleared the pre-registered thresholds. In a two-tier divergence market with crosscurrents pulling in different directions (value sector rotation vs. tech liquidation), pattern ambiguity is the expected KNN state, and the gates handled it appropriately.

No change to TRACE's SPRT (still LLR -0.996, CONTINUE, with -1.948 headroom to DEGRADED). The boundary remains the most urgent risk flag on the desk.

Event Risk vs. Expectation

The desk plan rated today as MEDIUM event risk (S&P Flash PMIs at 9:45 AM, New Home Sales at 10:00 AM, plus continued post-GOOGL/TSLA/INTC earnings digestion). This was correct — none of these were Tier-1 events. The actual session narrative was dominated by the oil-pullback rotation and the Intel intraday collapse (which was an earnings-aftermath story, not a calendar event). The Michigan Consumer Sentiment final was released today at 10:00 AM but was not flagged in the morning briefing's calendar — it was incorrectly corrected to July 31. The actual release (49.5 vs preliminary 54.4) was a significant negative revision.

The plan's key calls:

• MNEME (p_trade 0.45, direction long, conviction low): The correct direction call (long→long hit) but slightly underconfident on probability (p_trade 0.45, traded=1, Brier 0.3025). The plan's assessment that "today is a narrative-driver session" was correct — and the narrative environment produced a losing trade for the anchor window. The low conviction posture was justified.
• TRACE (p_trade 0.35, direction long, conviction low): Well-calibrated — p_trade 0.35 with no trade produces a Brier of 0.1225. The plan's assessment that "Friday lower volume may reduce pattern quality" was accurate. The direction call (long→n/a) is neutral — DIA was flat/fractional on the day, so the direction lean is untestable.

Plan Calibration (2026-07-24)

First session where the plan produced an above-average Brier (0.45→1) for a traded direction that was correct but unprofitable. The desk plan has been consistently directionally correct across the last 3 sessions (07-21: both long→hit, 07-23: both long→hit, 07-24: MNEME long→hit). Directional accuracy remains high even when outcome quality is poor.

Desk-Level Learnings

• MNEME's Window B suffered its first structural loss in a two-tier divergence fade. The anchor window (holdout PF 3.76, Phase 3: 4W/3L) was caught by the afternoon fade — SPY rallied to 743.71 at noon and closed at 738.90. This is the defining risk for KNN strategies on this market regime: the session diverges between the observation window and the execution window. The historical library from 2020-2024 may not have adequate representation of sessions where a midday catalyst (oil rotation) fades into a persistent tech liquidation by the afternoon. [Flag for learnings.md]: The two-tier divergence creates a specific failure mode for Window B KNN strategies: the morning pattern (observed 10:00-12:00) captures the rotation-driven rally, but the afternoon (14:00-16:00 execution) reverts as the tech sell-off reasserts dominance. The KNN's historical pattern library may not have enough of these structurally-divergent sessions to correctly weight the afternoon fade risk. This is a session-character-specific vulnerability, not a KNN architecture flaw.
• TRACE's no-trade was the correct gate behavior in a pattern-ambiguous session. DIA's crosscurrent day (Dow caught between rotation-positive value sectors and market-wide risk-off) produced no bar-by-bar pattern that cleared the max_sim ≥ 0.60 gate. This is the desired behavior — the KNN correctly identified that today's DIA microstructure didn't resemble anything in the historical library with sufficient clarity. The 0.35 p_trade forecast was well-calibrated.
• The desk's combined cumulative P&L turns negative for the first time. After 07-23's MNEME breakeven +$9.60, the cumulative buffer was $+60.18. Today's full trade loss (-$143.30) brings the combined desk to -$83.12 across 15 real trades. MNEME's cumulative (+$123.81) still covers TRACE's draw (-$206.93), but the margin is narrowing — MNEME is +$123.81 on 9 trades, TRACE is -$206.93 on 6. The desk now needs approximately 2 MNEME wins (avg win ~$112) to return to breakeven.
• Plan calibration continues to improve. The plan's directional accuracy is 3/3 sessions on direction calls (07-21: MNEME long→hit, TRACE long→hit; 07-23: MNEME long→hit, TRACE long→hit; 07-24: MNEME long→hit). The p_trade calibration is converging — MNEME's 0.45 on a trade day (Brier 0.3025) is reasonable; TRACE's 0.35 on a no-trade day (Brier 0.1225) is excellent. The desk PM's instinct to rate both members low-conviction on narrative-driven, two-tier sessions is validated.
• TRACE's SPRT unchanged at LLR -0.996, with -1.948 headroom to DEGRADED. The no-trade day does not improve or worsen the SPRT trajectory — it simply extends the observation period. The pre-registered DEGRADED protocol (desk PM review within 1 session, TEMPER conversation within 5 sessions) remains the most urgent risk flag on the desk. The window for improvement is narrowing: on the current trajectory, TRACE needs 2-3 wins in the next 3-4 qualifying trades to pull the SPRT back toward the CONSISTENT boundary. Every subsequent session without a win increases the risk of DEGRADED activation on the next qualifying loss.
• FOMC+ PCE double-header on Wed-Thu next week (July 29-30). Both members have hard skip rules for FOMC and PCE — Wednesday and Thursday will be no-trade days for the entire desk. Monday (Durable Goods Orders at 8:30 AM) and Tuesday (Consumer Confidence at 10:00 AM) are the only tradeable sessions before the blackout. With VIX at 18.58 (MID regime) and the two-tier divergence persisting, the Tuesday Consumer Confidence print could be a catalyst — the Michigan Sentiment data (49.5, released today) suggests consumer sentiment is deteriorating faster than the narrative acknowledges.

Plan Accuracy

Overall plan accuracy: 4.0 / 5

Reflection Filed

• Filed: 2026-07-24 18:00 ET
• Next session event risk (from EOD briefing): HIGH — Durable Goods Orders at 8:30 AM (Monday). VIX at 18.58 (MID regime, stable). The FOMC-PCE double-header looms on Wed-Thu with hard skips on both days. Monday and Tuesday are the only tradeable sessions this week.
• GHOST: frozen (Phase 2) — no reflection written.
• MNEME and TRACE per-member reflections filed.