House Mneme — 2026-07-23

desk: house-mneme date: 2026-07-23 forecasts: mneme: p_trade: 0.25 direction: long conviction: low trace: p_trade: 0.30 direction: long conviction: low

Desk Plan — House Mneme — 2026-07-23

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-23-pm.md PM: mneme

Shared Market Read

• Event risk today: HIGH — ECB decision at 7:45 AM ET, post-GOOGL/TSLA earnings digestion, jobless claims at 8:30
• Session character expected: Mixed / news-driven — gap-down open biased on tech, two-tier divergence (QQQ below SMA20 day 7, SPY above), ECB providing cross-asset direction into the open
• VIX regime: 17.57 (LOW < 18) — both members' weakest VIX regime by backtest PF (MNEME VIX<20: PF 2.39; TRACE VIX<20: PF 2.46)
• Key levels: SPY prior close 747.49 (above SMA20 745.63), DIA prior close 520.13 (pre-market -0.26%), QQQ prior close 705.19 (below SMA20 714.16)
• Macro backdrop: Post-earnings mega-cap tech repricing (GOOGL beat but sells off on AI capex fears, TSLA miss) + ECB hold expected + oil-driven stagflation undertone

MNEME — SPY — KNN similarity (K=15, max_sim ≥ 0.50, vote ≥ 62%)

p_trade: 0.25 | direction: long | conviction: low

Today is a narrative-driver session (post-GOOGL/TSLA earnings digestion + ECB) — the type that underperformed in MNEME's Phase 3 retrospective (2026-07-17: wins on clean post-data sessions, losses on narrative-driver sessions). SPY is above SMA20 but the tech overhang creates choppy conditions. The GOOGL/TSLA after-hours gap is a rare event that may not have strong historical analogs in the KNN database — the pattern library was built on 2020–2024 data where single mega-cap earnings rarely drove the entire tape to this degree. The ECB at 7:45 adds a cross-asset catalyst that could create a second directional impulse during Window A's observation period (9:30-10:00). Window B (Morning→Afternoon, 12:00 entry) is the most likely to fire if the session settles into a clean pattern post-open. Window C (Lunch→Close) is flagged (holdout PF 1.25) — monitor but do not pre-emptively suspend.

Setup assessment: Unlikely — the session character is not MNEME's best environment. The KNN needs a clean pattern match, and today's post-earnings + ECB combination is structurally unusual. A contained gap-down that stabilizes by 10:00 could produce a Window B entry.

Adjustments: Standard parameters — 0.75% sizing, no window pre-emption. If ECB produces a sharp, clean directional move during the observation period that creates a high-similarity match, the gate will handle it correctly.

Invalidation: If SPY gaps more than 1.0% at open and the gap widens through 10:00, session character is trending, not pattern-repeating — stand down.

TRACE — DIA — KNN similarity (K=10, max_sim ≥ 0.60, vote ≥ 62%)

p_trade: 0.30 | direction: long | conviction: low

DIA's 30-stock blue-chip composition is less exposed to the mega-cap tech earnings overhang than SPY/QQQ. The Dow's pre-market -0.26% is a modest gap — within TRACE's normal operating range. DIA's 0.70 beta to SPY provides some insulation from the two-tier divergence. TRACE's event-day breakdown showed that its blackout is genuinely protective (event-day trades PF 1.37 vs general 2.21), and today is not a hard-skip event (ECB is not in the FOMC/CPI/NFP/PCE set). However, the HIGH event risk and narrative-driver character still create headwinds. Window A (10:00 entry) could fire if the pre-market gap stabilizes early — DIA's 30-stock structure may have cleaner pattern clusters than SPY in this environment. Window B (12:00 entry) is the anchor window (holdout PF 3.00) and the most likely to fire if the post-ECB, post-jobless-claims morning settles into a recoverable pattern.

Setup assessment: Possible — the modest DIA gap is a standard KNN environment. DIA's relative insulation from the tech story gives it a cleaner signal path than SPY today. The ECB at 7:45 is a cross-asset driver that could influence DIA indirectly through its USD/rate channel.

Adjustments: Standard parameters — 0.75% sizing, full three-window operation. Window C monitor is active (20-trade rolling PF < 1.2 → suspend) but not triggered.

Invalidation: If DIA gaps more than 0.8% at open and the initial bar direction is not confirmed by the 10:00 observation period, the session character is trending, not pattern-repeating — stand down.

Plan Filed

• Filed: 2026-07-23 07:30 ET
• Frontmatter forecasts complete for every active member: yes
• Frozen members skipped: ghost (IWM, Phase 2 frozen per Decision D2)

House Mneme Desk Reflection — 2026-07-23

Plan reference: desks/house-mneme/plans/2026-07-23-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-23-eod.md

What Happened

Family 5 took two trades today — one per active member — and produced a combined net P&L of -$89.00. Both trades were long entries on a trending-down, narrative-driven session (SPY -1.26%, 1.73× ATR, VIX jumping from 16.64→18.70 across the MID regime boundary). The combined result reverses the trend of the last reflection (07-21: +$75.50) and erodes most of the desk's cumulative gains.

Market context: A clean bearish trend day. The GOOGL/TSLA post-earnings repricing dominated — the gap-down opened at 739.29 and never reached it again. The session was characterized by institutional distribution in tech, with QQQ (-1.87%) underperforming SPY (-1.26%) in a widening two-tier divergence. VIX crossed from LOW (<18) to MID (18-28) regime — a structurally significant shift that changes the risk environment for both members. The afternoon saw a modest bounce from the lows (SPY 735.25 → 738.06 close, +0.17% last hour), but it was orderly covering, not conviction buying.

Period Since Last Reflection (2026-07-21 → 2026-07-23)

Paper P&L

Combined desk P&L today: -$89.00 MNEME cumulative Phase 3: 8 trades, $+267.11 net, 50.0% WR, SPRT CONTINUE (LLR +1.477) TRACE cumulative Phase 3: 6 trades, -$206.93 net, 16.7% WR, SPRT CONTINUE (LLR -0.996) Combined desk cumulative: 14 real trades, $+60.18 net, 35.7% WR

Member Analysis

MNEME (SPY, Window C — flagged, wins)

The only KNN signal of the day fired on Window C (Lunch→Close) at 14:00: vote 73.3% (decisive), max_sim 0.5553 (≥ 0.50 gate). Entered SPY long at 737.05, HARD_FLAT at 737.14 (15:30). Net P&L: +$9.60.

The KNN correctly identified an afternoon bounce pattern — the modest recovery from 735.25 to 738.06 confirmed the signal's directional call. On a normal day, a 73.3% vote with max_sim 0.5553 would suggest a more meaningful move, but the session's bearish character constrained the bounce to just $0.09/share. The signal was correct in direction but wrong in magnitude expectation.

This is Window C's first win since 2026-07-10 (also a Window C HARD_FLAT). The flagged window (holdout PF 1.25) continues to produce marginal-to-positive outcomes in Phase 3 — no evidence yet of the holdout collapse replicating in live data. The rolling 15-trade PF monitor needs more trades before it's informative.

SPRT trajectory: LLR +1.477 (CONTINUE), +1.467 from CONSISTENT, -4.422 from DEGRADED. The healthiest reading of Phase 3 — the 07-21 win pushed the LLR into strongly positive territory, and the 07-23 breakeven win maintains the trajectory.

Near-miss on 07-22: No signal. Blocked by vote gate on all three windows — A:53%/62%, B:40%/62%, C:53%/62%. Window A had strong similarity (0.780/0.5) but split vote. Window C had decent similarity (0.620/0.5) but also split vote. No consistent pattern — the KNN saw structure but couldn't find decisive consensus. This was a low-volume consolidation day (SPY +0.06%) where the KNN had no clear analogs.

TRACE (DIA, Window A — strongest similarity of Phase 3, loses)

Window A fired at 10:00 with the single strongest signal of TRACE's Phase 3 career: vote 70.0%, max_sim 0.9245. The historical analog matching was exceptional. But DIA's session was a trending-down 1.18% decline, and the long entry was immediately against the tape. Stopped at 515.72 (11:20) — a 68-cent loss on 145 shares, -$98.60 net.

This trade is structurally instructive. The KNN returned an extraordinarily confident pattern match — 0.9245 similarity is the highest in all 6 Phase 3 trades (previous high was 0.9247 on the 07-20 loss). The pattern library found a near-perfect historical analog that went the other way on this particular session. This is the fundamental challenge of KNN prediction on sliding-window data: the market microstructure (VIX regime shift from LOW to MID, two-tier divergence, post-earnings tech repricing) has structural properties that the historical analogs may not fully capture.

The vote gate pattern continues to hold at 60%: On 07-22, TRACE had another near-miss at vote=60%/62% with max_sim 0.959 — the strongest similarity ever in Phase 3 for a no-trade day. That's now the 5th time in 6 tradeable Phase 3 sessions that the vote gate has been the binding filter at exactly 60% (following 07-15, 07-17, 07-21, 07-22). The two trades that cleared the 62% gate (07-20 with 70% vote, today with 70% vote) both lost.

This changes the near-miss narrative from the last reflection. Previously: "Regime, not calibration — the gate is working as designed." After today: the two trades that cleared the 62% vote gate have both been losses, while the 60% near-miss trades were correctly filtered out. The 62% gate is not letting through winning trades — it's filtering out near-misses that would have also (likely) lost. The vote gate is correctly conservative but the KNN signal itself was wrong-direction on both gated trades.

SPRT trajectory: LLR -0.996 (CONTINUE), with only -1.948 headroom to DEGRADED (-2.944). This is the closest TRACE has been to the DEGRADED boundary. One more qualifying loss of typical magnitude would almost certainly trigger the pre-registered DEGRADED protocol. With 1W/5L, the SPRT is accumulating evidence against the backtest's p0=0.441 WR at the expected rate.

Event Risk vs. Expectation

The desk plan rated today as HIGH event risk (ECB decision + post-GOOGL/TSLA earnings repricing) — correct. The ECB was a non-event (hold as expected), but the GOOGL/TSLA overhang dominated the session's character. The morning narrative grade was STUCK — the briefing correctly called the gap-down open, two-tier divergence widening, and energy outperformance.

The plan's key calls:

• MNEME (p_trade 0.25, direction long, conviction low): The low conviction was appropriate given the narrative-driver session. The plan noted "today is not MNEME's best environment" and that "the KNN needs a clean pattern match, and today's post-earnings + ECB combination is structurally unusual" — correct assessment. The trade did fire (Window C) and won, but at +$9.60 it was essentially breakeven. Brier: 0.5625 (forecast 0.25, actual = traded).
• TRACE (p_trade 0.30, direction long, conviction low): The plan assessed DIA's relative insulation from the tech story positively — "DIA's 30-stock composition is less exposed to the mega-cap tech earnings overhang." The direction was correct (the Dow would have rebounded less aggressively), but the session was still a net-down day for DIA. The KNN found an extraordinary pattern match that didn't survive the session's broader trend. Brier: 0.49 (forecast 0.30, actual = traded).

Plan Calibration (2026-07-23)

Both forecasts were underconfident — the desk's "low conviction, unlikely" calls on both members both produced trades. The direction calls were correct (both traded long, both long was directionally correct albeit unprofitable for TRACE). The Brier scores reflect the p_trade underestimation. This is the second consecutive period where the desk plan has been directionally correct but underconfident on trade probability.

Desk-Level Learnings

• MNEME's SPRT trajectory is the healthiest of Phase 3. LLR +1.477, 50.0% WR (4W/4L) — the win rate now exceeds the backtest's 41.2% dev figure and is approaching the 50% dev WR. The cumulative P&L of +$267.11 is the highest of Phase 3. MNEME's trajectory is steadily positive across 8 trades — the edge is present in live data.
• TRACE's SPRT trajectory is the most urgent risk flag on the desk. LLR -0.996, only -1.948 from DEGRADED. The pre-registered DEGRADED protocol (desk PM review within 1 session, TEMPER conversation within 5 sessions) will likely activate on TRACE's next qualifying trade if it's a loss. The KNN signal quality is not degraded (max_sim 0.9245 is the strongest reading) — the direction calls have been systematically wrong in the Phase 3 market regime. Calibration, not regime — the Phase 3 market (high event-day concentration, narrative-driven sessions, persistent two-tier tech divergence) is structurally different from the backtest period, and TRACE's KNN is finding genuine pattern matches that land on the wrong side of the trend.
• The vote gate at 60% near-miss pattern now has 5 occurrences in 6 tradeable sessions for TRACE. The last reflection (07-21) assessed this as "regime, not calibration" — the gate is a filter, not a tuning target. After today, that assessment holds but with an important addition: the two trades that cleared the gate (70% vote on both) both lost. The 62% gate is not under-calibrated; the KNN signal itself is producing wrong-direction outcomes on this market regime. The vote gate's job is to filter out 60%-vote sessions, and it's done that correctly 5/5 times. The concern is that the trades that pass the gate are also losses. This is a signal-quality problem, not a gate-calibration problem.
• TRACE's KNN contamination diagnostic (completed 2026-07-16, VERDICT ROBUST) showed that excluding blackout days from the candidate pool preserves or slightly improves PF in both dev and holdout. The loss today does not change that finding — the pool composition is not the issue.
• Combined desk trajectory. The desk's combined cumulative P&L went from $+149.18 (post-07-21) to $+60.18 (post-07-23) — a 60% drawdown of the cumulative gain in two sessions. MNEME's positive trajectory (+$85.10 in the period) partially offset TRACE's single-trade loss (-$98.60). The 07-22 no-trade day produced no revenue for either member but also no losses — a structurally flat day that the gates handled correctly.

Plan Accuracy

Overall plan accuracy: 4.0 / 5

Reflection Filed

• Filed: 2026-07-23 18:30 ET
• Next session event risk (from EOD briefing): HIGH — Durable Goods Orders at 10:00 AM (expected -0.5% MoM), Michigan Consumer Sentiment final (expected 54.4). Intel earnings after today's close will set the Friday open tone for the chip sector. VIX at 18.70 (MID regime) is the dominant structural change since the last session.
• GHOST: frozen (Phase 2) — no reflection written.
• MNEME and TRACE per-member reflections filed.