House Mneme — 2026-07-20

desk: house-mneme date: 2026-07-20 forecasts: mneme: p_trade: 0.35 direction: long conviction: low trace: p_trade: 0.20 direction: long conviction: low

Desk Plan — House Mneme — 2026-07-20

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-20-pm.md PM: MNEME

Shared Market Read

• Event risk today: LOW — No Tier-1 events. Light macro calendar: 8:30 AM Chicago Fed National Activity Index, 10:00 AM Leading Economic Indicators (consensus 0.0%). FOMC blackout period (July 18-29) means no Fed speakers. The dominant risk is the carry-over from last week's chip rout and positioning ahead of this week's mega-cap tech earnings (Tesla, Alphabet, IBM, AT&T, TI Wed/Thu).
• Session character expected: Mixed/choppy, waiting for catalyst. The light macro calendar (LEI only) and absence of Fed speakers mean the session will be driven by technical positioning and pre-earnings squaring. The chip rout is unresolved, but the selling pressure may pause as the market digests the three-day selloff. Expect a range-bound session with no clear trending bias. The LEI at 10:00 AM is the only scheduled catalyst — could set the afternoon tone if it surprises.
• VIX regime: 18.38 (MID regime — 18-28). Edged down from Friday's close of 18.77. Both stacks are profitable in all VIX regimes (MNEME VIX < 20: PF 2.39, 197 trades; VIX 20-30: PF 3.54, 100 trades; TRACE VIX < 20: PF 2.46; VIX 20-30: PF 2.12). The MID regime is not a headwind for either stack.
• Key levels: SPY prior close 743.28, SMA20 at 744.97 — SPY closed below the SMA20 for the first time since July 6. The 740 level (prior-day low) is the key support — a break below would signal the trend-down is accelerating. A bounce from 743 toward 744.97 (SMA20) is the natural mean-reversion path. DIA: no direct level in the briefing — DIA's 30-stock blue-chip composition provides insulation from the chip rout, but the broad market direction is the dominant signal.
• Macro backdrop: The chip rout (PHLX Semi -1.63% Friday, third consecutive day of losses, SOXX near 535 make-or-break level) is the dominant unresolved narrative. The two-tier market (tech sold off, eight of 11 S&P 500 sectors positive on Friday) is the defining structural feature — the broad market is not uniformly risk-off, it's rotating out of tech and into energy/cyclicals. The market is waiting for this week's mega-cap tech earnings as the first potential catalyst for stabilization or acceleration.

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

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

1. SPY below SMA20 — a significant technical event, but the character is mixed/choppy, not trending-down. SPY closed at 743.28, below the 744.97 SMA20. The last time this happened (July 6), SPY bounced back above within 3 sessions. The KNN engine's historical library contains sessions where SPY traded below its SMA20 — the question is whether the neighbours voted for a recovery or continued decline. The mixed/choppy character (not a clean trending-down day) means the session may not produce a single-directional pattern that the KNN engine can match cleanly. The chip rout is focused on semis/tech — SPY's broader sector composition provides some insulation.
2. LEI at 10:00 AM — a clean catalyst at the start of Window B's observation window. The LEI consensus is 0.0% (flat, extending the 29-month streak of non-positive readings). This is a MEDIUM-impact release. Window B (10:00-12:00 observation, 14:00-16:00 entry, hard flat 16:00) has the cleanest data structure: the 24-bar observation window begins at 10:00 AM, capturing the full LEI reaction. The 72-dim feature vector is the richest feature space, and Window B is MNEME's anchor window (holdout PF 3.76). If the LEI print is a positive surprise (above 0.2%) or a negative surprise (below -0.2%), the morning session could establish a clear post-data trend that the KNN engine can match historically. If the LEI prints at consensus (0.0%), the session is likely to remain choppy/range-bound, which is unfavorable for pattern matching.
3. Window A (9:30-10:00 observation, 10:00 entry) has a clean window this week — no mid-observation data. Unlike Friday where the 9:15 AM IP print fractured Window A's observation, today has no 8:30 AM or 9:15 AM data. The 6-bar Window A observation captures a clean opening micro-structure: the gap direction (if any) and the first 30 minutes of price discovery. The 18-dim vector is small but the signal-to-noise ratio is actually improved by the absence of mid-window data. However, the 8:30 AM CFNAI (Chicago Fed National Activity Index, LOW impact) drops before the 9:30 AM open — its effect is baked into the opening print, not mid-window noise.

Key considerations:

• Window B is the best candidate today — the 10:00 AM LEI catalyst at the start of the observation window is the cleanest data structure. The 12:00 ET entry gives the market time to absorb the LEI print and establish a settled direction.
• Window A is a weaker candidate — the observation window is clean (no mid-window data), but the 18-dim vector at 10:00 AM may not capture enough signal after a mixed/choppy opening.
• Window C (12:00-14:00 observation, 14:00 entry): Clean observation window, but flagged (holdout PF 1.25). Standard monitoring.
• Sizing: Standard 0.75% per trade ($187.50 on $25K). No adjustment — the KNN gates are the filter.
• The VIX 18.38 MID regime is favorable (MNEME's VIX 20-30 bucket shows PF 3.54), but the mixed/choppy character is a headwind for pattern matching — the KNN engine needs a directional bootstrap to produce a decisive vote.

Invalidation: If SPY gaps below 740 (the prior-day low) at the open, the session is a confirmed trend-down break below the prior range. The KNN engine's long-only bias is structurally misaligned. If the 10:00 AM LEI prints below -0.2% (negative surprise, signalling contraction), the risk-off tone could accelerate through the morning — Window B's observation window would capture this, and the vote would likely not support a long entry. If SPY is range-bound within 741-745 for the first hour (no direction), the 18-dim Window A vector is unlikely to produce a max_sim ≥ 0.50 match.

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 mixed/choppy session character is a particular headwind for DIA's cleaner 30-stock microstructure, which depends on directional resolution to produce clean pattern matches.

Key considerations:

• TRACE's event-day deliverable showed blackout is genuinely protective (event-day PF 1.37 vs. 2.21 general). Today is not a Tier-1 event day, but the LEI at 10:00 AM is a MEDIUM-impact data release. This is a data-heavy session character even though the events are below the Tier-1 hard-skip threshold. TRACE's KNN engine has historically underperformed on data-heavy sessions.
• Window C is near the pre-registered suspension threshold. TRACE's Window C rolling 20-trade PF is approximately 1.125 (from the recomputed per-window holdout: n=18, PF 1.13). The pre-registered suspension threshold is rolling 20-trade PF < 1.2. One more losing Window C trade could trigger the 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.
• DIA's 30-stock composition provides insulation from the chip rout but not immunity from broad risk-off. The Dow's concentration in financials, industrials, and healthcare means the semis selloff has less direct impact. However, SPY's close below SMA20 is a broad-market technical signal that affects all sectors. DIA's lower beta (~0.70) means less downside capture, but the mixed/choppy character means the bounce (if it comes) will be shallower than SPY's.
• The max_sim ≥ 0.60 gate is the primary filter — it should naturally sit out a session of this character. 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.

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 wide (first 30-minute range > 0.6× ATR) and the 6-bar observation window shows high variance across bars, the KNN engine will not find a close-enough historical match. If the 10:00 AM LEI print is a negative surprise (below -0.2%), the broad-market risk-off tone would extend — DIA is not immune to a macro-driven selloff, and the KNN engine's long-only bias would be structurally misaligned. The most likely outcome: no signal from any window, similar to 2026-07-15 and 2026-07-17 where the gates correctly filtered the session.

Plan Filed

• Filed: 2026-07-20 07:00 ET
• Frontmatter forecasts complete for every active member: yes

House Mneme Desk Reflection — 2026-07-20

Plan reference: desks/house-mneme/plans/2026-07-20-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-07-20-eod.md

What Happened

Family 5 took two trades today — one per active member — and produced a combined net P&L of -$145.84, the desk's worst single-day result since pooling. Both members took long entries in an open-high-fade session: the markets gapped up on Iran peace-talk optimism, then sold off through the day. SPY -0.65% from open, QQQ -0.91% from open. DIA (Dow) was the worst-performing major index at -0.6%. VIX closed at 18.65 (MID regime). LEI at 10:00 AM printed -0.2% (below 0.0% consensus), adding to the cautious tone. Oil surged +3.8% to $91.42/bbl on Hormuz tensions.

MNEME (SPY, Window C — flagged window): Entered long at 14:00 @ 744.64, stopped at 744.49 at 14:30. Net P&L: -$15.00. The KNN's 12:00-14:00 observation window captured SPY in post-LEI stabilization around 743-744 and saw a bounce pattern that didn't materialize — SPY continued its intraday fade through the afternoon. Window C is the flagged window (holdout PF 1.25) and the desk plan specifically noted it was the weakest candidate, but it was the only window whose KNN signal cleared both gates today. Window B (anchor window, holdout PF 3.76) — the desk plan's preferred candidate — did not produce a signal. Cumulative Phase 3: 6 trades, 33.3% WR, $+182.01 net. SPRT CONTINUE (LLR +0.199), well within normal operating territory.

TRACE (DIA, Window A — opening→morning): Entered long at 10:00 @ 520.41, stopped at 519.50 at 10:20. Net P&L: -$130.84. The 6-bar observation window (9:30-10:00) captured DIA opening on peace-talk optimism, but that optimism faded within minutes as oil surged and the Dow became the session's worst index (driven by consumer cyclicals Nike -1.49%, Disney weakness). The desk plan specifically said Window A was "unlikely to fire" — but the KNN found a strong enough signal to clear the max_sim ≥ 0.60 and vote ≥ 62% gates. Cumulative Phase 3: 5 trades, 20.0% WR, -$108.33 net. SPRT LLR -0.643, only -2.301 from the DEGRADED boundary at -2.944 — this is the desk's primary risk flag.

Desk-level observation (Pooled Validation, Family 5): Today is the 3rd co-fire day (both members traded). The clustered record: 2026-07-13 (-$53.95), 2026-07-16 (+$46.33), today (-$145.84) — 1W/2L, net -$153.46. The combined desk cumulative P&L has dropped from $+219.52 (post-07-17) to $+73.68 — a 66% drawdown in one session. MNEME's SPRT remains healthy (+0.199), but TRACE's erosion is accelerating.

Period Since Last Reflection (2026-07-17 → 2026-07-20)

Paper P&L

Combined desk P&L today: -$145.84 MNEME cumulative Phase 3: 6 trades, $+182.01 net, 33.3% WR, SPRT CONTINUE (LLR +0.199) TRACE cumulative Phase 3: 5 trades, -$108.33 net, 20.0% WR, SPRT CONTINUE (LLR -0.643) Combined desk cumulative: 11 real trades, $+73.68 net, 27.3% WR

Event Risk vs. Expectation

The desk plan rated today as LOW event risk (no Tier-1 events). Correct — the LEI release was MEDIUM impact and below the hard-skip threshold for both members. The session's character was correctly assessed as mixed/choppy with an LEI catalyst at 10:00 AM.

The plan's key call — that both members were low conviction (MNEME p_trade 0.35, TRACE p_trade 0.20) with unfavorable session character for pattern matching — was correct in narrative but wrong in execution. The KNN engine on both instruments found signals where the PM's narrative assessment said there were none. Both signals were wrong.

The plan specifically noted:

• For MNEME: "Window C (12:00-14:00 observation, 14:00 entry): Clean observation window, but flagged (holdout PF 1.25). Standard monitoring." — Window C fired anyway.
• For TRACE: "The mixed/choppy session character is a headwind for DIA's cleaner 30-stock microstructure, which depends on directional resolution to produce clean pattern matches." — The 6-bar Window A found a pattern anyway.

This is the first session where both members' KNN engines generated signals that materially disagreed with the PM's narrative assessment, and both were wrong. This requires a desk-level observation, not an alarm.

Plan Accuracy

Overall plan accuracy: 3.5 / 5 — Plan accuracy is dragged down by the setup prediction miss. The narrative assessment was correct; the KNN engine's override was wrong. The plan itself was well-reasoned; the issue is that the KNN engine doesn't read the plan.

Desk-Level Learnings

• The KNN engine's independence from the PM's narrative assessment is a double-edged sword. Today, both members' KNN engines found signals in a session where the desk plan argued against trading — and both were wrong. This is the first live evidence that the KNN engine can override sound narrative reasoning with a false pattern match. The plan correctly assessed the session character as unfavorable for long-only KNN entries (open-high-fade, mixed/choppy, LEI catalyst resolving below consensus), but the KNN found historical patterns that suggested a bounce. The patterns were wrong. This is not an indictment of the KNN mechanism — the backtest PF (2.45 MNEME, 3.08 TRACE) is built on being wrong 55-60% of the time and winning big when right. The issue is that both members were wrong on the same session type, and the losses were amplified by co-fire.
• MNEME's Window C firing while Window B did not is the most telling finding. The anchor window (holdout PF 3.76) had a 24-bar observation window encompassing the full LEI reaction (10:00-12:00) — the richest feature space, exactly the catalyst-driven session the plan identified as the best setup. If the KNN could not find a qualifying signal in the richest observation window of the day, the Window C signal (from the flagged, PF 1.25 window) was facing a higher evidentiary burden. It fired anyway and was wrong. This is the 7th Phase 3 trade from Window C for MNEME (PF 1.25 holdout) — consistent with marginal performance.
• TRACE's SPRT is approaching the DEGRADED boundary. LLR -0.643 with only -2.301 of headroom to the DEGRADED boundary (-2.944). TRACE has now lost 4 of 5 Phase 3 trades (20% WR). The pre-registered SPRT (p0=0.441 from backtest WR) was designed to trigger at evidence, not narrative discomfort — and it's working as designed. At the current trend, one more losing trade could push TRACE into DEGRADED territory, triggering the pre-registered suspension protocol: first, the desk PM reviews; then the monitor-flags.md entry triggers a TEMPER conversation within 5 sessions. TRACE's next trade is the most consequential in its Phase 3 history.
• The 2026-07-16 finding about max_sim distance above gate carries forward. TRACE's loss on 07-16 came at max_sim 0.6013 (0.0013 above gate). Today's loss — at whatever similarity cleared the 0.60 gate — is the second consecutive loss on a signal near the gate boundary. The pattern is accumulating: 4 trades, only the win (07-09 at max_sim 0.8795, far above gate) was high-quality. The three losses came at 0.8574 (07-08 — high sim, but wrong direction day), 0.7707 (07-13), and 0.6013 (07-16). Today's similarity at 2.301 from DEGRADED is the most urgent data point — but at n=5, still too small to calibrate.
• Combined cumulative drawdown from peak. The desk's combined cumulative went from $+219.52 (post-07-17) to $+73.68 today — a 66% drawdown in one session. This is entirely driven by TRACE's $130.84 loss. MNEME's $15.00 loss is noise. The desk's health is tracking TRACE's health, and TRACE is weakening.
• [x] Flag for learnings.md: Today is the first live evidence that both Family 5 KNN engines can produce false pattern matches on the same session type (open-high-fade) when the PM's narrative assessment argues against trading. The KNN is not selectively wrong — it is wrong on the same session character across both instruments. If this pattern repeats (one more co-fire day where both fire and lose on the same session character), it warrants a TEMPER conversation about whether the open-high-fade session character is a structural blind spot for KNN similarity engines — not a strategy change, but a research finding for Family 5's backtest diagnostic run.
• [x] Flag for learnings.md: TRACE's SPRT is at LLR -0.643, with only -2.301 to the DEGRADED boundary (-2.944). At n=5 (20% WR), the pre-registered SPRT parameters are working as designed. If TRACE's next trade is a loss, the DEGRADED boundary is likely to be crossed, triggering pre-registered suspension protocols: desk PM review, TEMPER conversation within 5 sessions. Flag proactively — not actionable yet, but the desk PM is watching.

Reflection Filed

• Filed: 2026-07-20 18:30 ET
• Next session event risk (from EOD briefing): NONE — Tuesday July 21, 2026 has no Tier-1 events. Mega-cap tech earnings season (Tesla, Alphabet, IBM, AT&T, Texas Instruments) begins mid-week.
• GHOST: frozen (Phase 2) — no reflection written.