House Mneme — 2026-07-21

desk: house-mneme date: 2026-07-21 forecasts: mneme: p_trade: 0.35 direction: long conviction: low trace: p_trade: 0.20 direction: long conviction: low

Desk Plan — House Mneme — 2026-07-21

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-21-pm.md PM: MNEME

Shared Market Read

• Event risk today: LOW — No Tier-1 events. No macro data. Light earnings before open (SCHW, GM, NVS, GPC) whose effects are baked into the opening print. The dominant narrative is pre-positioning ahead of tomorrow's mega-cap tech earnings (Alphabet, Tesla after close).
• Session character expected: Modestly trending-up / range-bound. Asian markets rallied overnight (Nikkei +3.36% on chip stocks), US futures pointing higher led by Nasdaq. The chip bounce is the intraday catalyst. Without a macro catalyst, the session is likely range-bound with a mild upside bias — the open-high-fade pattern from yesterday (SPY -0.65% from open, QQQ -0.91%) may or may not repeat. Light volume expected as the market waits for earnings.
• VIX regime: 17.62 (LOW — dropped below 18 from yesterday's 18.65 MID). Both stacks are profitable in LOW VIX (MNEME: PF 2.39, 197 trades; TRACE: PF 2.46). VIX below 18 is a neutral-to-favorable regime for both — not a headwind.
• Key levels: SPY prior close 742.15, SMA20 at 744.74, prior-day range 741.56–748.69. Support at 741.56 (prior low), then 740. Resistance at 748.69 (prior high), then the SMA20 at 744.74 as the first test. DIA: no direct level in the briefing; the Dow's 30-stock composition provides a cleaner tape but lower beta to the chip bounce.
• Macro backdrop: Correction pause. The chip rout (SOXX at 535 make-or-break) is stabilizing — Nvidia +2.39% yesterday. The two-tier market (tech sold off, other sectors holding) continues. The market is in a holding pattern ahead of Wednesday's mega-cap earnings (Alphabet, Tesla) and next week's FOMC (July 29). The light macro calendar means company-specific catalysts and sector rotation dominate.

MNEME — SPY — KNN Similarity-Based (Long Only)

Setup present: Possible, low conviction. Three factors inform today's assessment:

1. Gap-up bias from chip bounce is favorable for long-only, but the session character is uncertain. The overnight rally in Asian chip stocks and the bounce in US futures suggest a gap-up open. A gap-up SPY open that holds through the first hour would establish a constructive micro-structure for Window A's 6-bar observation (9:30–10:00). However, yesterday's open-high-fade pattern (SPY opened at 747.00, closed at 742.15) is still fresh in the market's memory — the chip bounce could be a dead cat bounce that fades again. The KNN engine's historical library contains both patterns, and the vote will reflect whichever is more similar.
2. Window B (10:00–12:00 observation, 14:00 entry) is the anchor — today is a clean data structure for it. No macro releases scheduled. The morning session will be driven by the chip bounce narrative and pre-earnings positioning. The 24-bar observation window (3 features × 24 bars = 72-dim) is MNEME's richest feature space, and Window B's holdout PF of 3.76 is the best across all windows. If the market establishes a settled direction by 10:00 AM (after the initial gap-up settles), the model has a clean 2-hour observation to find a historical match. The afternoon entry at 14:00 ET captures the final 2 hours of the session — the period most likely to be influenced by pre-earnings positioning ahead of tomorrow's Alphabet/Tesla closes.
3. Window A (9:30–10:00 observation, 10:00 entry) and Window C (12:00–14:00 observation, 14:00 entry) are weaker candidates. Window A's 6-bar vector is noise-prone on a gap-up day — the first 30 minutes often include the gap print and the initial reaction, which may not be representative of the session's settled direction. Window C is flagged (holdout PF 1.25) and the 12:00–14:00 lunch period on a low-event day is the slowest part of the session — unlikely to produce a clean pattern match.

VIX context: MNEME's LOW VIX bucket (VIX < 20, PF 2.39, 197 trades) is profitable but the edge is less pronounced than the VIX 20–30 band (PF 3.54). The 17.62 VIX reading is comfortable but not optimal.

Recent performance context: Yesterday's Window C trade was a small STOP loss (-$15.00). Cumulative: 6 trades, 33.3% WR, +$182.01 net. SPRT at n=6, LLR +0.199, status CONTINUE. The small sample means the SPRT is still in the early zone — one trade does not change the runway.

Sizing: Standard 0.75% per trade ($187.50 on $25K). No adjustment — the KNN gates are the filter.

Invalidation: If SPY gaps below 741.56 (prior low) at the open, the session is a confirmed trend-down break — the long-only bias is structurally misaligned. If the gap-up exceeds 0.8% (> ~$748 at open) and then fades aggressively in the first 30 minutes (giving back more than half the gap), the tape is signalling a repeat of yesterday's open-high-fade pattern — low confidence in any directional setup. If the morning session is range-bound within 742–745 for the full 2-hour observation (Window B), the 72-dim vector is unlikely to produce a decisive vote.

TRACE — DIA — KNN Similarity-Based (Long Only)

Setup present: Unlikely, low conviction. TRACE's tighter max_sim gate (≥ 0.60 vs. MNEME's 0.50) means fewer sessions qualify in any environment. The gap-up bias is favorable for long-only, but DIA's 30-stock blue-chip composition mutes the chip-bounce tailwind — the Dow has no semis exposure.

Key considerations:

• DIA's cleaner microstructure is a double-edged sword today. The Dow's lower beta (~0.70 vs. SPY) means the chip bounce produces a smaller upside move. The tighter max_sim gate (≥ 0.60) requires a highly specific pattern match — a muted, low-volatility up-day may not produce a distinctive enough feature vector to trigger the gate. The most likely outcome is a no-trade session where the KNN engine correctly identifies that DIA's current microstructure doesn't match any clean historical pattern.
• Window C is near the pre-registered suspension threshold. TRACE's Window C rolling PF is approximately 1.13 on 18 trades (from the recomputed per-window holdout). The pre-registered suspension threshold is rolling 20-trade PF < 1.2. One more loss could trigger automatic suspension. If Window C's vote passes the gate today, the trade should be evaluated with this additional context — but the pre-registered rule is automatic, not discretionary. The low-event, low-vol character of today's session is not ideal for Window C's lunch-to-close observation period.
• TRACE's event-day deliverable (blackout genuinely protective) is irrelevant today — no Tier-1 events, no hard skip triggered. The KNN contamination diagnostic confirmed the edge is robust to blackout-day neighbor presence in the candidate pool (clean-run PF 3.28 vs. baseline 3.08 in holdout). No structural concern.
• Window A (9:30–10:00 observation, 10:00 entry) was the active window yesterday — it produced a trade that STOP'd at -$130.84. The gap-up today could produce a different micro-structure in the first 30 minutes, but the same caution applies: a gap-up open's first 30 minutes are noise-dominated for the 6-bar vector.

Sizing: Standard 0.75% per trade ($187.50 on $25K). No adjustment.

Invalidation: The max_sim ≥ 0.60 gate is the binding filter. If DIA's opening range is tight (first 30-minute range < 0.3× ATR) and flat, the KNN engine will not find a close-enough historical match. If the chip bounce fades and SPY gives back its gains, DIA will follow lower — the long-only bias is structurally misaligned on a confirmed fade. The most likely outcome: no signal from any window, consistent with the tighter gate design.

Plan Filed

• Filed: 2026-07-21 07:00 ET
• Frontmatter forecasts complete for every active member: yes

House Mneme Desk Reflection — 2026-07-21

Plan reference: desks/house-mneme/plans/2026-07-21-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-21-eod.md

What Happened

Family 5 took one trade today — MNEME only — and produced a combined net P&L of +$75.50. The chip bounce that dominated the session (Nikkei +3.36%, SOX rally, Advantest +8.13% in Japan) provided the clean, sustained directional environment that the KNN anchor window is designed for. MNEME's Window B (SPY) won; TRACE (DIA) sat out — the Dow's muted beta and tighter gates produced no qualifying signal.

Market context: A clean trend day. SPY +0.81% day-over-day, +0.25% from open. QQQ +1.84%. VIX dropped from 18.65 to 17.05 (LOW regime). Both indices traded at ~1.0× ATR — normal range but clearly directional. The gap-up held through the session, with a dead-flat last hour. The chip bounce reversed the previous week's tech rout and produced exactly the structural environment where KNN similarity engines perform best: sustained directional drift without narrative cross-currents or geopolitical overlay.

MNEME (SPY, Window B — anchor): Entered long at 12:00 @ 747.50, held to HARD_FLAT at 16:00 @ 748.25. Net P&L: +$75.50. The KNN returned a modest but decisive signal: max_sim 0.6137 (≥ 0.50), vote 66.7% (≥ 62%). The afternoon was a slow, sustained drift higher — no fade, no late-session reversal. The HARD_FLAT captured the full afternoon continuation. Cumulative Phase 3: 7 trades, 42.9% WR (3W/4L), $+257.51 net. SPRT CONTINUE (LLR +0.838) — healthiest reading since Phase 3 began, with +2.106 to CONSISTENT and -3.783 to DEGRADED. The 42.9% WR now tracks the backtest's 41.2% dev figure almost exactly.

TRACE (DIA, no trade): No KNN signal above gate. The near-miss was Window A: vote=60%/62% (−2pp), strong similarity at 0.825/0.60. This is the 3rd time in Phase 3 that TRACE's vote gate has been the binding filter at exactly 60% (following 2026-07-15 Window A and 2026-07-17 Window C). DIA's muted beta to the chip bounce meant the Dow's feature vector was less distinctive than SPY's — the stronger similarity confirms the KNN was looking in the right direction, but the split vote (6/10 neighbours) was insufficient to clear the 62% threshold. The vote gate continues to perform as designed. Cumulative Phase 3: 5 trades, 20.0% WR (1W/4L), -$108.33 net. SPRT CONTINUE (LLR -0.643), with -2.301 headroom to DEGRADED.

Desk-level observation (Pooled Validation, Family 5): Today is not a co-fire day (TRACE sat out), so no clustered observation. The combined desk cumulative P&L stands at $+149.18 across 12 real trades (7 MNEME, 5 TRACE), recovering from the 07-20 drawdown. The clustered co-fire record remains unchanged from 2026-07-20: 3 co-fire sessions (2026-07-13, 2026-07-16, 2026-07-20) with 1W/2L, net -$153.46.

Period Since Last Reflection (2026-07-20 → 2026-07-21)

Paper P&L

Combined desk P&L today: +$75.50 MNEME cumulative Phase 3: 7 trades, $+257.51 net, 42.9% WR, SPRT CONTINUE (LLR +0.838) TRACE cumulative Phase 3: 5 trades, -$108.33 net, 20.0% WR, SPRT CONTINUE (LLR -0.643) Combined desk cumulative: 12 real trades, $+149.18 net, 33.3% WR

Event Risk vs. Expectation

The desk plan rated today as LOW event risk — correct. No Tier-1 events. The chip bounce was a structural catalyst, not an event-driven one. The session played out as a clean trend day, confirming the plan's assessment that the gap-up would likely hold.

The plan's key calls:

• MNEME (p_trade 0.35, direction long, conviction low): Window B was correctly identified as the anchor. The plan called Window B "the best candidate" and noted "today is a clean data structure for it" — correct on all counts. The low conviction was prudent given the mixed session character of the prior week, but the trade fired and won. Brier score: 0.4225 (forecast 0.35, actual = traded).
• TRACE (p_trade 0.20, direction long, conviction low): Correctly assessed as "unlikely" with the tighter max_sim gate filtering DIA on a muted-beta session. The KNN gates confirmed the plan's assessment. Near-miss at 60% vote/0.825 sim shows the filter is working, not missing.

The plan's anticipation of Window B as the best candidate and Window C as weaker was correct. The only miss: the p_trade forecast for MNEME (0.35) was conservatively low for a session that ultimately produced the correct structural environment.

Plan Accuracy

Overall plan accuracy: 4.0 / 5

Desk-Level Learnings

• MNEME's cumulative WR (42.9%) now tracks the backtest's 41.2% dev figure almost exactly. The SPRT LLR +0.838 is the healthiest reading since Phase 3 began — the trajectory is steadily positive, trending toward CONSISTENT-with-backtest territory. The KNN anchor window (B) fired correctly on a clean, directional session and won. The Window C flagged-loss pattern continues (07-20: -$15.00) but the anchor window's health is the desk's primary monitor, and its signal is good.
• TRACE's vote gate (62%) is consistently the binding filter at the same near-miss point (60%) across structurally distinct session types — PPI+Warsh (07-15), chip-rout continuation (07-17), chip-bounce rally (07-21). The consistent clustering at 60% across three different market environments is evidence the 62% threshold is calibrated to the DIA KNN's natural vote distribution, not overfit to one regime. The gate is working as designed. Regime, not calibration — for the vote gate itself.
• TRACE's SPRT trajectory (LLR -0.643) is the desk's primary risk flag. At -2.301 from the DEGRADED boundary (-2.944), one more qualifying loss would likely trigger the pre-registered protocol. TRACE's next qualifying trade is the most consequential in Phase 3. The 2026-07-20 loss (-$130.84) on a headline-driven opening pop is the most visible symptom — the KNN found a strong pattern (sim 0.9247, vote 70%) on a deceptive session, and the loss was the largest single-trade drawdown in Family 5 Phase 3. The SPRT is accumulating evidence at the pre-registered rate; the protocol is designed for exactly this scenario.
• Combined desk recovery from the 07-20 drawdown. The desk's combined cumulative went from $+73.68 (post-07-20) to $+149.18 (post-07-21) — a recovery of $75.50 from MNEME's Window B win. MNEME's anchor window performed exactly as designed, recovering half of the prior period's losses in one session. TRACE's drawdown is the structural concern; MNEME's positive trajectory is the offset.
• The 2026-07-15 near-miss pattern (both members' Window A at vote=60%/62%) is worth watching as the desk-level finding extends. MNEME's near-misses have been at strong similarity (0.846) with marginal vote (60%). TRACE's near-misses have been at varying similarity (0.878, 0.531, 0.825) but all at the same 60% vote. The vote gate pattern is stronger for TRACE (3 occurrences) than MNEME (1 occurrence). This is consistent with K=10 (TRACE) producing a coarser vote distribution than K=15 (MNEME) — a structural property of the different parameter sets, not a desk-level calibration concern.
• [x] Flag for learnings.md: The 2026-07-15 near-miss (both members at vote=60%/62% on different windows) has now been extended by TRACE's 3rd 60%-vote near-miss. The pattern is relevant to Family 5: the 62% vote gate is consistently the binding filter at the same near-miss point (60%) — across different instruments (SPY and DIA), different observation windows (A and C), and different session types (PPI+Warsh, chip-rout, chip-bounce). This consistency across the family suggests the 62% vote threshold is operating at the natural resolution point of the binomial KNN vote distribution (9/15 = 60% for K=15, 6/10 = 60% for K=10). Desk-level finding: the vote gate is calibrated correctly; no tweak needed for either member.
• [x] Flag for learnings.md: TRACE's SPRT at LLR -0.643 (CONTINUE) with -2.301 headroom to DEGRADED is the desk's most urgent risk flag. The pre-registered DEGRADED protocol (desk PM review within 1 session, TEMPER conversation within 5 sessions) is designed for this scenario and will activate if TRACE's next qualifying trade is a loss. The 2026-07-20 loss (-$130.84) on a headline-driven opening pop is the largest single-trade drawdown in Family 5 Phase 3 and the primary driver of the SPRT trajectory. MNEME's healthy SPRT (+0.838) does not offset TRACE's risk position — each stack's monitor is independent by design.

Reflection Filed

• Filed: 2026-07-21 18:30 ET
• Next session event risk (from EOD briefing): HIGH — Wednesday July 22 brings Alphabet (GOOGL) and Tesla (TSLA) mega-cap earnings after the close. The chip bounce's follow-through will be tested by these prints. FOMC rate decision is July 29. TRACE's SPRT position means any qualifying trade on either member carries heightened desk-level risk awareness.
• GHOST: frozen (Phase 2) — no reflection written.