House Mneme — 2026-07-17

desk: house-mneme date: 2026-07-17 forecasts: mneme: p_trade: 0.30 direction: long conviction: low trace: p_trade: 0.25 direction: long conviction: low

Desk Plan — House Mneme — 2026-07-17

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-17-pm.md PM: MNEME

Shared Market Read

• Event risk today: MEDIUM — Busy macro data block (8:30 AM: Import Prices, Housing Starts, Building Permits; 9:15 AM: Industrial Production, Capacity Utilization; 10:00 AM: U. Michigan Consumer Sentiment). No Tier-1 hard skip (FOMC/CPI/NFP/PCE) — both stacks trade through MEDIUM event risk. NFLX gapping down 8%+ on earnings miss adds a tech-sector headwind alongside the ongoing chip rout.
• Session character expected: Gap-down / news-driven. Nasdaq futures -1.6%, SPY futures -0.84%. The macro data block is the first potential pivot — the market may attempt a recovery after the data is absorbed, or the selloff could accelerate. The VIX regime shift from LOW (16.73) to MID (18.08) is the first such shift since July 13 and signals elevated uncertainty.
• VIX regime: 18.08 (MID regime — first shift from LOW since July 13). Both stacks are profitable in MID/VIX 20-30 (MNEME VIX 20-30: PF 3.54, 100 trades — strongest regime; TRACE VIX 20-30: PF 2.12, all regimes profitable). The shift from LOW to MID is actually favorable for MNEME's edge profile, though the gap-down character is a separate headwind.
• Key levels: SPY prior close 750.87, SMA20 at 744.86 (+6pts above). The overnight gap-down (SPY futures -0.84%) puts SPY near the SMA20 support level. A break below 744.86 would be a significant technical event. DIA no direct level — infer Dow 30-stock composition is less exposed to the chip rout but still subject to broad risk-off. SPY/QQQ divergence at its widest in this cycle (QQQ 13.67pts below SMA20, SPY above) — SPY is the constructive side but the gap-down threatens the SMA20 support.
• Macro backdrop: The chip rout (PHLX Semi -5% Thursday, selling accelerating into Friday) is the dominant narrative. The macro data is secondary — the structural rotation out of AI/tech into cyclicals/energy is the defining feature. SPY's broader sector composition provides some insulation vs. QQQ, but the gap-down is broad-based.

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

Setup present: Possible, lower conviction than yesterday. Three factors work against a clean setup today:

1. Gap-down open — structurally misaligned with long-only bias: SPY futures -0.84% puts the open near or below the prior 750 reference. The KNN engine's long-only bias is structurally opposed to a risk-off gap-down. The gates (max_sim ≥ 0.50, vote ≥ 62%) are the correct filter — they should naturally sit out sessions where the pattern library doesn't support a long entry. I expect most of the KNN neighbours to vote against a long entry after a gap-down of this magnitude.
2. Fractured data timing across observation windows: Unlike yesterday where the 8:30 AM data block resolved cleanly before Window A's observation window, today's data is spread across the entire first half: - Window A (9:30-10:00 observation, entry 10:00): The 9:15 AM Industrial Production print drops MID-OBSERVATION (between bars 3 and 4 of the 6-bar window). This is the worst timing — the 18-dim feature vector captures a fractured opening where the first 3 bars reflect the gap-down + 8:30 AM data, and the last 3 bars include the IP reaction. The KNN historical library contains very few sessions with this specific data-timing pattern, making clean pattern matches less likely. - Window B (10:00-12:00 observation, entry 12:00): The 10:00 AM Consumer Sentiment print drops at the START of the observation window. The 24-bar observation includes the sentiment reaction, which is a cleaner data structure than the mid-window IP print. This is the best candidate window today. - Window C (12:00-14:00 observation, entry 14:00): No data events. Cleanest observation window, but the window itself is flagged (holdout PF 1.25).
3. VIX shift is favorable — but the character is the issue: VIX 18.08 puts MNEME in the VIX 20-30 bucket (strongest regime, PF 3.54). However, the gap-down character is a long-only headwind regardless of the vol regime. The KNN engine's vote mechanism should naturally reflect this: if 15 neighbours voted for a long entry after a similar gap-down pattern, few of them are likely to have gone up.

Key considerations:

• Window B (10:00-12:00 observation, entry 12:00) is the best candidate — the full morning session after Consumer Sentiment is the richest feature space, and the 12:00 ET entry is late enough that the gap-down may have been absorbed or reversed.
• Window A is the weakest candidate today — the 9:15 AM IP print mid-observation is noise that the KNN engine likely won't find clean historical matches for.
• Sizing: Standard 0.75% per trade. No adjustment — the gates are the filter, not manual sizing.
• Invalidation: If SPY opens below the SMA20 (744.86) and the first 30-minute range exceeds 0.7× ATR (8.35), the session is likely a trending risk-off day. The KNN gates should filter this naturally. If the 10:00 AM Consumer Sentiment print is sharply negative (< 48.0), the gap-down could accelerate through the morning — Window B's observation window would capture this, and the vote would likely not support a long entry.

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

Setup present: Unlikely, lower conviction than MNEME. DIA's 30-stock blue-chip composition provides some insulation from the chip rout, but the broad risk-off gap-down is a headwind for any long-only strategy. TRACE's tighter max_sim gate (≥ 0.60 vs. MNEME's 0.50) means fewer sessions qualify in any environment, and today's gap-down character is structurally misaligned.

Key considerations:

• TRACE's event-day deliverable showed that event-day sessions underperform (PF 1.37 vs. 2.21 general). While today's data block is not a Tier-1 hard skip, the busy macro calendar creates a similar "data-heavy" session character. The blackout is genuinely protective for TRACE — the event-day-deliverable finding supports this.
• DIA's 30-stock composition (less tech, more financials/industrials/healthcare) means the chip rout has less direct impact, but the broad-based risk-off tone (SPY -0.84% futures) affects all sectors. The Dow's 30 names are not immune to a macro-driven selloff.
• The key level for a potential bounce: if the macro data (Import Prices, Housing Starts) provides a disinflation-reinforcing read, the gap-down could be faded. DIA's lower beta (~0.70) means less downside capture in a selloff but also less upside capture in a recovery — the bounce won't be as sharp.
• Sizing: Standard 0.75% per trade. No adjustment.
• Invalidation: The max_sim ≥ 0.60 gate is the primary filter. If DIA's opening range is wide and the 6-bar observation window is noisy (high variance across bars), the KNN engine will not find a close-enough historical match. This is the most likely outcome today — essentially a high-probability no-trade session on the gates alone, before considering the macro backdrop.

Plan Filed

• Filed: 2026-07-17 07:30 ET
• Frontmatter forecasts complete for every active member: yes

House Mneme Desk Reflection — 2026-07-17

Plan reference: desks/house-mneme/plans/2026-07-17-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-17-eod.md

What Happened

Family 5 took one trade today — MNEME only — and produced a combined net P&L of $+0.00. MNEME fired a Window B long at 12:00 ET that was immediately stopped at entry price (break-even). TRACE sat out — no KNN signal above either gate.

The session was the third consecutive chip-rout day, now testing make-or-break technical levels (PHLX Semi SOXX near 535). The macro data block was busy but the Import Price Index re-inflation surprise (+0.3% MoM vs -0.8% consensus) was the day's primary inflection — it capped the morning recovery attempt. SPY opened at 742.17 (gap-down from 750.87 prior close), bounced to a midday high of 747.25, then faded into close at 743.28. SPY closed below its 20-day SMA for the first time since July 6. VIX rose to 18.77 — MID regime, up from 16.73. The defining structural feature: eight of 11 S&P 500 sectors finished positive, but the tech-weighted indexes were routed. A two-tier market.

MNEME (SPY, Window B — anchor): Entered long at 12:00 @ 746.89 when SPY was at the peak of its midday bounce (747.25 high), stopped at entry price immediately — net $0.00. The KNN returned a decent signal (max_sim 0.6763, 66.7% vote — 10 of 15 neighbours up) but the timing was wrong: the bounce was a technical recovery capped by the Import Price re-inflation surprise, not a trend reversal. The entry bar opened at the high of the bounce and immediately faded, triggering the 1.5× ATR stop at entry. Cumulative Phase 3: 5 trades, 40.0% WR (2W/3L), $+197.01 net. SPRT CONTINUE (LLR +0.469).

TRACE (DIA, no trade): No KNN signal above gate today. The near-miss data: Window A had a high similarity (0.904) but only 40% vote (far below 62%); Window C had a close vote (60% vs 62%) but similarity was below gate (0.531 vs 0.60). TRACE has now logged 3 no-trade days in Phase 3 across event skip, bars outage, and gate failure — the tight max_sim ≥ 0.60 gate is structurally limiting. Cumulative Phase 3: 4 trades, 25.0% WR (1W/3L), $+22.51 net. SPRT CONTINUE (LLR -0.289).

Desk-level observation (Pooled Validation, Family 5): This session had only one active member (TRACE sat out), so no clustered observation today. The desk's combined cumulative P&L stands at $+219.52 net across 9 real trades (5 MNEME, 4 TRACE), with 2 same-day co-fire sessions producing clustered observations: 2026-07-13 (-$53.95) and 2026-07-16 (+$46.33). The clustered record is 1W/1L, net -$7.62 — effectively break-even across the two co-fire sessions.

Period Since Last Reflection (2026-07-16 → 2026-07-17)

Paper P&L

Combined desk P&L today: $+0.00 MNEME cumulative Phase 3: 5 trades, $+197.01 net, 40.0% WR, SPRT CONTINUE (LLR +0.469) TRACE cumulative Phase 3: 4 trades, $+22.51 net, 25.0% WR, SPRT CONTINUE (LLR -0.289) Combined desk cumulative: 9 real trades, $+219.52 net, 33.3% WR

Event Risk vs. Expectation

The desk plan rated today as MEDIUM event risk (busy macro data block, no Tier-1). Correct. The Import Price Index re-inflation surprise (+0.3% vs -0.8%) was the session's inflection point — a genuine macro surprise that capped the recovery attempt. The plan identified the mid-observation timing risk for Window A (9:15 AM Industrial Production print) correctly — Window A didn't fire. The plan correctly identified Window B as the best candidate — it did fire — but the bounce was a technical dead cat, not a trend reversal.

The plan flagged that "the gap-down character is a separate headwind" and set p_trade at 0.30 for MNEME — low conviction. Correct. The session's character (chip rout continuing, re-inflation surprise, gap-down fade) was the wrong environment for a long-only KNN strategy, even with a decent signal.

Plan Accuracy

Overall plan accuracy: 3.75 / 5

Desk-Level Learnings

• MNEME's break-even stop-out is the least damaging outcome for a wrong-timed entry. A stop at entry price costs nothing. The KNN was directionally right (SPY did bounce) but the entry at 12:00 ET captured the peak, not the continuation. This is an inherent limitation of Window B's fixed entry time on a recovery-bounce day — the KNN can't know whether the bounce will continue or fade by 12:00.
• The 2026-07-15 near-miss (MNEME Window A: vote=60%/needed 62%, sim 0.846/0.50) is worth noting. The KNN came 2pp below the vote gate on a session with high similarity. The same thing happened on TRACE Window A the same day (vote=60%/needed 62%, sim 0.878/0.60). Both members had the same near-miss pattern on the same session (PPI + Warsh testimony day) — the KNN was close to firing but the gate held. This is the second consecutive cycle where the vote gate has held on structurally noisy sessions. Evidence the gate is working correctly, not a calibration issue.
• MNEME's SPRT boundary is healthy (LLR +0.469, +2.475 to CONSISTENT) — still well within CONTINUE territory. The break-even doesn't move the monitor.
• TRACE's SPRT remains the desk's primary risk flag. LLR -0.289, only -2.655 to the DEGRADED boundary. Three losses in four trades is a tough start but still within the pre-registered SPRT parameters (p0=0.441 from backtest WR). The 2026-07-16 loss at max_sim 0.6013 (barely above gate) and today's Window C near-miss (vote 60%/sim 0.531, both gates failing) both suggest TRACE's signal quality may degrade near the gate boundary — but 4 trades is too small to calibrate from.
• [x] Flag for learnings.md: TRACE's max_sim gate (0.60) has now produced 3 losses in 4 Phase 3 trades, with the 2026-07-16 loss at max_sim 0.6013 (0.0013 above gate). The Window C near-miss today (sim 0.531 vs 0.60) suggests the gate is correctly filtering low-similarity signals. The 2026-07-16 loss at a barely-qualifying similarity is the first live signal that the max_sim-vs-outcome relationship may carry information for Family 5 that the backtest (n=59 for TRACE) lacked power to detect. Early — flag for TEMPER at the 10-trade checkpoint, not actionable now.
• [x] Flag for learnings.md: The 2026-07-15 near-miss (both members' Window A at vote=60%/needed 62%) is the second data point suggesting the 62% vote threshold is creating a genuine quality filter on structurally noisy sessions. Both members' backtests found near-zero similarity spread within the qualified cohort — the vote gate itself is the binding filter, not max_sim. Worth monitoring whether near-miss days (vote 60-62%) systematically underperform qualified days (vote ≥ 62%) at the 10-trade checkpoint.

Reflection Filed

• Filed: 2026-07-17 18:30 ET
• Next session event risk (from EOD briefing): NONE — Monday July 20 has no Tier-1 events. LEI (10:00 AM) is MEDIUM impact but below the hard-skip threshold. Clean trading session expected.
• GHOST: frozen (Phase 2) — no reflection written.