House Mneme — 2026-08-14

desk: house-mneme date: 2026-08-14 forecasts: mneme: p_trade: 0.55 direction: long conviction: med trace: p_trade: 0.2 direction: long conviction: low

Desk Plan — House Mneme — 2026-08-14

Briefing reference: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-08-14-pm.md PM: MNEME

Shared Market Read

• Event risk today: MEDIUM — tradeable, NOT a hard-skip. July Retail Sales at 8:30 AM ET (consensus +0.1% MoM / prior +0.2%) and Michigan Consumer Sentiment prelim at 10:00 AM ET (consensus ~54.5 / prior 55.2). Both are second-tier — neither MNEME (SPY) nor TRACE (DIA) carries them in the pre-registered Tier-1/2 hard-skip set (shared/event_calendar.py, tier_filter=2). The runners will evaluate signals today — no no_trade:event_day pre-close. Firm-wide status per watchlist: GO (no Tier-1 in the 2-day window; FOMC Minutes Aug 19 is the next Tier-1). AMAT's beat-and-raise reported after close Thu is already in the price — the gap-down is an open, not a discrete tail, and no window in either stack holds past 16:00 ET.
• Session character expected: Trending-up bias, event-conditioned. The record-high S&P (7,798.99, 27th all-time high) on back-to-back benign inflation (in-line CPI Tue, soft/flat PPI Wed) is the base case. This morning's calibration directive (from the 08-13 PARTIAL grade) is explicit: weight the macro leg. The AMAT -5% sell-the-news gap-down and the two-tier AI divergence (memory names SNDK +13.7% / MU +4.2% strong vs. AI-networking CSCO -8% / CBRS -14% bleeding) are the conditional risk leg, NOT the index base case — exactly the error class flagged on 08-13, where a benign PPI released the same morning and drove the tape to a record while the names stayed contained. Benign Retail Sales → expect the record-high drift to extend; hot Retail Sales OR a broadening AMAT/semis fade → two-sided-to-weak.
• VIX regime: 14.53 — LOW (<18). Both stacks' best backtested regime (MNEME VIX<20 PF 2.39; TRACE VIX<20 PF 2.46). This is live and favorable today, not moot — LOW VIX is where both mechanisms' edge is strongest, and it is a tailwind for any qualifying signal on a tradeable day.
• Key levels: SPY prior close 777.84 (08-13), prior range 774.19–779.37, SMA20 754.50 (above), ATR(14) 8.40. DIA not in the briefing price block — TRACE's recent Phase 3 band ~$515–524 (last entry 07-27 at $523.31). Both indices hold above SMA20; broad uptrend intact and extending.
• Macro backdrop: The disinflation gauntlet's next tradeable leg is the Friday consumer read (Retail Sales) — a benign print keeps the soft-landing risk-on bid intact into a record; a hot print reopens the reacceleration risk. Oil's Hormuz escalation (Brent ~$88, +5% wk) is a background commodity bid that tilts the Dow's cyclicals firmer relative to Nasdaq's semis-heavy tape.

MNEME — SPY KNN (p_trade 0.55, long, med)

Setup present: likely to evaluate — this is a recognizable LOW-VIX continuation day, with the two-tier divergence as a quality risk on whatever fires, not a reason to expect no signal. Today is tradeable (Retail Sales/Michigan are not in my skip set). The 8:30 Retail Sales print lands before all three of my observation windows begin (A 9:30–10:00, B 10:00–12:00, C 12:00–14:00) — the macro noise is largely digested by the time my fingerprints form. The briefing's calibration directive is the key read for me: benign base case, AMAT gap-down as a conditional risk leg. My library has real analogues for LOW-VIX, above-SMA20, record-extension SPY sessions that open with a single-name gap-down contained while the broad index holds — that is a recognizable microstructure day, not a black-swan print. I've been active (3 of last 4 tradeable sessions logged a real trade, including a clean Window A win on 08-10). The 10:00 Michigan survey is a modest headwind — it sits at my Window A entry boundary and inside the Window B observation — but a low-impact second-tier sentiment survey rarely moves SPY enough to corrupt a pattern match; it is a watch, not a blocker.

• Sizing vs. event risk: Standard 0.75% ($187.50), one position — Retail Sales/Michigan are tradeable second-tier catalysts, not sizing inputs; the event filter is a gate and it does not trigger today. No defensive sizing, no shadow book. The live risk is two-tier divergence, and my documented 07-24 learning is that such days reverse between observation and execution windows — that argues for discipline on the exit (respect the 1.5× ATR stop immediately, especially in the afternoon), not for reduced size. My edge is regime-backed (LOW VIX) and the divergence penalty is expressed through the stop, which I honor.
• Session-character fit: Good-to-partial. The base case (benign retail → record extension) is my home turf. Window A (observe 9:30–10:00, execute 10:00–12:00) is cleanest — it is fully pre-Retail-Sales in observation and only brushes the Michigan release at its entry. Window B (observe 10:00–12:00, execute 14:00–16:00) is the most vulnerable to the two-tier reversal shape I lost on 07-24; Window C stays the standing flagged-window watch (holdout PF 1.25, rolling 15-trade PF monitor).
• Key levels: SPY prior range 774.19 (low) / 779.37 (high), prior close 777.84. A sustained hold above ~777 (prior close) into the 10:00–12:00 observation signals a converged record-extension (clean continuation shape); a fade back toward/below 774.19 (prior-day low) is the divergence-fade shape I must respect and stop out of promptly.
• Invalidation (stand down): If the AMAT/semis gap-down broadens into a SPY gap-down > 1% at the open and VIX ticks above 18 on a hot Retail Sales surprise, I sit out the AM session entirely. A hot consumer-print that undercuts the benign disinflation narrative flips the day from my recognizable LOW-VIX continuation library into an event-repricing (spurious-match) class.
• What I'm watching to learn: (1) Whether the AMAT -5% gap stays name-contained (single-name fade = my clean shape) or broadens into semis/tech (Nasdaq-led weakness = the two-tier failure mode) — the briefing's calibration directive says assume contained until shown otherwise; (2) whether the 10:00 Michigan print injects any visible SPY dislocation at my Window A entry / B observation (does a low-impact survey ever corrupt my pattern match in practice?); (3) Window C's rolling PF — the standing flag after the 08-11 stop.

TRACE — DIA KNN (p_trade 0.2, long, low)

Setup present: possible but structurally muted by two stacked constraints. Today is tradeable (Retail Sales/Michigan not in my Tier-1/2 skip). DIA's cyclicals/industrials (CAT, HON, MMM, CVX) are insulated from the AMAT/semis gap-down and benefit from the oil/Hormuz commodity bid (Brent ~$88) — the Dow is arguably the better-positioned firm leg of this session's two-tier tape. That directional case is real. But two things keep my genuine probability of a real trade low: (1) my B/C no_data availability pattern (sharpened 08-11 to "early window works, later windows don't") persists — the 24-bar windows that historically carry my signal keep returning no bars, leaving only Window A viable; and (2) my recent A-window vote profile has run 30–60% (08-13 A 50%, 08-11 A 30%, 08-06 A 40%) — consistently below the effective 62% bar even when max_sim is ample (0.79–0.92). I rate a real trade at 0.2: the A window can evaluate, but both the data path and the recent vote distribution argue against a clean fire.

• Sizing vs. event risk: Standard 0.75% ($187.50), one position — same reasoning as MNEME; Retail Sales/Michigan are not sizing gates. Separate live-risk flag: my SPRT is at LLR −1.350, −1.595 from DEGRADED (CONTINUE). Today is the first tradeable (non-skip) session since that state set in — the DEGRADED protocol cannot start on a no-trade day, but if I log a qualifying trade today and it loses, the pre-registered DEGRADED protocol begins immediately. This is a monitoring headwind, not a sizing input — I do not pre-emptively shrink size on it.
• Session-character fit: Poor-to-moderate. The two-sided tape is directionally favorable to the Dow, but the persistent B/C data gap is the binding constraint on my firing rate, independent of the macro. A clean Window A signal is the only realistic path today — and the 10:00 Michigan print lands exactly at my A entry, an added mid-window noise input on the one window that can fire.
• Key levels: DIA band ~$515–524 (last entry 07-27 at $523.31). Whether the Dow reclaims the band on the firm leg — and specifically whether my 6-of-7 losing open-high-fade shape recurs or the cyclicals hold their oil-backed gains — tells me if my recent losing shape is a two-sided-tape phenomenon or a normal-Dow weakness.
• Invalidation (stand down): If BOTH the A window returns no_data AND my recent low-vote pattern holds (vote < ~50%), no qualifying signal can form — I close without a trade. Absolute invalidation: if DIA gaps > 1% and VIX breaks 18 on a hot Retail Sales print, I sit out — the Dow would be dragged into an event-repricing tape, not a recognizable pattern-library day.
• What I'm watching to learn: (1) Whether B/C data returns today (a tradeable day is the first honest test of the 08-11 "early window works, later don't" diagnosis since it was sharpened — the last several days have been skip days or no_signal); (2) whether the Dow's cyclicals leg leads the firm side of the two-tier tape (tests whether my industrial tilt is the right instrument for this divergence); (3) the 60%-cluster question — if A resolves, today's vote is honest (not event-contaminated), so a repeat 60% (6-of-10) resolution would be the seventh same-shaped instance.

Pooled note (Family 5)

Both stacks are long-only, LOW-VIX-regime-backed, and tradeable today — not a skip day, so any co-firing (both log a real trade same-day) counts as one clustered observation for the pooled Family 5 record, per TEMPER's correlated-session rule. I forecast MNEME the more likely to trade (0.55) and TRACE muted by data availability + low votes (0.2); both direction long (the mechanism is long-only). The session's defining risk for both is two-tier divergence — MNEME's documented failure mode (07-24) and TRACE's cyclical tilt's better-positioned shape. The briefing's calibration directive (weight the macro leg; treat the AMAT gap-down as a conditional risk leg, not the index base case) is the operative read for both. Each stack sizes independently at standard 0.75%; neither adjusts for the MEDIUM tradeable event. GHOST stays frozen (Phase 2, Decision D2) — no plan.

Plan Filed

• Filed: 2026-08-14 07:52 ET (pre-open — briefing/watchlist both pre-open, authoritative)
• Frontmatter forecasts complete for every active member: yes
• GHOST: FROZEN (Phase 2) — not planned for
• Firm-wide status: GO — Retail Sales 8:30 / Michigan 10:00 AM second-tier (tradeable), no Tier-1 in 2-day window; AMAT sell-the-news gap-down + oil/Hormuz are the conditional risk legs, benign macro is the base case per the 08-13 calibration directive

No trades taken.

House Mneme Desk Reflection — 2026-08-14

Plan reference: desks/house-mneme/plans/2026-08-14-plan.md EOD briefing: /Users/dadbot/Desktop/ClaudeBod/projects/dadbrain/Analysis/briefings/2026-08-14-eod.md

What Happened

Period since last reflection (2026-08-13 → 2026-08-14 — one session, quiet Friday)

Combined desk cumulative (Phase 3): +$418.25 across 24 trades (17 MNEME +$767.46, 7 TRACE −$349.21), 41.7% WR. No new observation this period — neither member traded, so the Family 5 pool is unchanged at 24.

MNEME — all three windows evaluated, none decisive; A grazed the vote floor by 2pp on a rangebound tape

MNEME logged no trade, and for once the near-miss row is a full three-window read rather than a partial one: blocked=knn_gate A:vote=60%/62% sim=0.847/0.5 B:vote=47%/62% sim=0.676/0.5 C:vote=47%/62% sim=0.656/0.5. Every window returned data and evaluated; none produced the decisive ≥62% vote. The similarity is healthy across all three (0.847 / 0.676 / 0.656, all comfortably above the 0.50 floor) — the candidate library matched, but on a quiet, compressed, sub-ATR ranging/choppy day no window's pattern carried a decisive continuation vote. Window A was the closest at 60% (2pp below the 62% floor); B and C resolved at 47%, well below.

Binding-gate verdict for MNEME: regime, not calibration. Today is a single 2pp miss on Window A against a genuinely absent setup — the session was a flat, slightly-red, sub-ATR range with no clean pattern for a continuation library to lock onto, and the tape held just under the record on consumer weakness rather than extending. A single 60% on A (the same 2pp-miss value A printed on 07-15) is not a "repeatedly close misses against one threshold" cluster — MNEME's A window trades regularly at 66–73% (08-10, 08-11, 08-13 all fired), and the binding constraint today was a structurally pattern-less day, not a threshold being grazed. No calibration signal, no unilateral tweak, no TEMPER conversation warranted. Consistent with SPRT (CONSISTENT-WITH-BACKTEST, sticky, LLR +3.594). The plan's forecast (0.55 long med, calibrated Brier 0.3025) was honest but wrong-exercised in the plan-calibration sense — it leaned on a LOW-VIX continuation read that a quiet consumer-weakness fade never offered.

TRACE — a full blank, but this one is a runner-starved artifact, not a genuine market read

TRACE logged no trade with blocked=knn_gate A:no_data B:no_data C:no_data — all three windows returned no data. This is a different shape from the 08-11/08-13 pattern (where Window A evaluated and B/C blanked): today even A returned nothing. The cause is operational, not market — TRACE's runner was starved by the run-trades 300s timeout; the orchestrator backfilled TRACE's runner, but the backfill produced no data for any window. So today yields no honest DIA market read at all — no sim, no vote, no first-blocking-gate value.

Binding-gate verdict for TRACE: data availability (observability) — specifically a runner-execution failure, not a market regime and not a calibration signal. Because no window produced any data, no calibration question can even be posed, and the 60%-cluster question (six distinct sessions at exactly 6-of-10) stays exactly where it was — no honest decisive vote added. This is the concrete hand-off for the data-infra / runner-logging diagnosis, and today it sharpens the picture: the B/C no_data window-dependence (08-11, 08-13) and today's all-window blank-from-starvation are two distinct failure modes — one is a same-day DIA fetch observability issue on the 24-bar windows, the other is orchestration timing that starves the runner entirely. Both keep TRACE shut out; neither is a market read. SPRT CONTINUE, n=7 (W1/L6), LLR −1.350, −1.595 from DEGRADED; the dry stretch since TRACE's last qualifying trade (07-27) is now thirteen sessions, and the DEGRADED clock still cannot start on a no-trade day.

Plan calibration (WS4b rows — 2026-08-14)

Both forecasts calibrated cleanly by Brier, but for the recurring wrong-exercised reason: MNEME's med-conviction long (0.55) rewarded a session that evaluated all windows and produced nothing (the plan over-weighted a LOW-VIX continuation day that never offered one), and TRACE's muted 0.20 — set specifically because of the B/C data path — rewards a session where the runner never even got to read the market. TRACE's probability calibration continues to look honest but under-informed; MNEME's all-long direction accuracy holds its strong recent run (long called on the last several traded days) even though today is a no-trade.

Desk-level observations

• No same-day clustered observation — neither member traded, so the Family 5 pooled pool is unchanged (24 observations); no correlated-session counting issue.
• The intra-desk asymmetry this period is the inverse of 08-13. On 08-13 MNEME fired high-conviction and lost on morning chop while TRACE got a genuine read; today MNEME got a full honest three-window read (all no-signal on a rangebound day) while TRACE got nothing at all because its runner was starved. One member exercised its mechanism and correctly declined; the other couldn't exercise its mechanism for operational reasons.
• Two standing threads, both evidence-gated, neither a gate action: (1) MNEME's all-long LOW-VIX continuation calibration — today's 0.55 med forecast was a plan-calibration miss (no extension came), but a single 2pp A-vote miss is not a threshold-grazing cluster; watch it, don't act. (2) TRACE's data path now shows two distinct failure modes (B/C window-dependence + runner starvation from the 300s timeout) — the runner-starvation blank is the operative one today and is the concrete hand-off for the data-infra/runner-logging diagnosis. Neither is calibration evidence that would license a unilateral change.

Paper P&L

Combined desk P&L this period: $0.00 MNEME cumulative Phase 3: 17 trades, +$767.46 net, 52.9% WR, SPRT CONSISTENT-WITH-BACKTEST (sticky, LLR +3.594) TRACE cumulative Phase 3: 7 trades, −$349.21 net, 14.3% WR, SPRT CONTINUE (LLR −1.350, −1.595 to DEGRADED) Combined desk cumulative: 24 trades, +$418.25 net, 41.7% WR

Event Risk vs. Expectation

• Event risk call (MEDIUM — Retail Sales 8:30 + Michigan 10:00, second-tier, tradeable): Correct as a risk call — no Tier-1, neither stack carries these in its skip set, and the desk was right not to pre-close. But the driver interpretation was wrong-side, and the EOD briefing grades this PARTIAL firm-wide: the plan's base case was benign-retail → record-high extension; the actual consumer read (Retail Sales -0.6%, weakest in over a year, plus Michigan falling to 51.0) was read as consumer weakness, capping the index rather than extending it. For this desk that meant neither stack's pattern library found the continuation it was positioned for — MNEME's A-window 60% on a day that never extended is the direct expression.
• Session character call: Missed (composition/direction). The plan wrote "trending-up bias, event-conditioned"; the actual session was a quiet, flat-to-slightly-down, sub-ATR (0.42–0.47×) ranging/choppy fade-and-hold with a flat final hour. Magnitude was small (so a mild miss, not a broad reversal the AM should have flagged), but the base-case direction (record extension) did not hold. This is the same consumer-leg blind spot the EOD flags firm-wide — on a soft hard retail print AND deteriorating soft sentiment, the consumer read can cap even a record.
• Invalidations: none triggered for either stack — no >1% SPY/DIA gap at the open, VIX stayed LOW (14.25, never >18), no hot-print repricing flip. MNEME's invalidation (broadened gap + VIX >18) never approached; TRACE's stand-down (A no_data + low-vote pattern) is not what closed it — the runner never returned data for any window.
• Forward risk (next sessions): FOMC Minutes Wed 08-19 — Tier-1, hard skip for both stacks by pre-registered rule; Jackson Hole Aug 21-22; Target/Walmart retail earnings (next consumer test); NVDA Aug 26. The consumer-weakness rotation risk is the live thread — whether it is a one-day growth-scare or the start of a consumer-led pullback.

Reflection Filed

• Filed: 2026-08-14 (EOD reflection, desk cadence — Friday)
• Next session event risk (from EOD briefing): MEDIUM-HIGH into next week — FOMC Minutes Wed 08-19 is Tier-1 (hard skip for both members); retail earnings + Jackson Hole ahead.
• GHOST: frozen (Phase 2, Decision D2) — no reflection written.
• MNEME and TRACE per-member reflections filed.
• Flags embedded for learnings.md (desk-blind, own members): MNEME — full three-window evaluation on 2026-08-14, none decisive on a quiet sub-ATR ranging day: blocked=knn_gate A:vote=60%/62% sim=0.847/0.5 B:47% sim=0.676 C:47% sim=0.656. Healthy similarity across all windows (library matched) but a pattern-less consumer-weakness fade offered no continuation vote; A grazed the 62% floor by 2pp (a single occurrence, not a cluster — A trades at 66-73% regularly). Binding-gate verdict: regime (setup absent, no fix needed); no calibration signal, no unilateral tweak. SPRT CONSISTENT (LLR +3.594). TRACE — blocked=knn_gate A:no_data B:no_data C:no_data on 2026-08-14, but this is an all-window blank from runner starvation (run-trades 300s timeout starved the runner; the orchestrator backfilled it, and the backfill returned no data for any window), NOT a genuine market read. This is a distinct failure mode from the 08-11/08-13 B/C window-dependence (where A evaluated): today even A returned nothing, so no sim/vote/first-gate value exists and no calibration question can be posed. Concrete hand-off for the data-infra/runner-logging diagnosis — two distinct data failure modes (window-dependent observability vs. full runner starvation), both shutting TRACE out. SPRT CONTINUE, LLR −1.350, −1.595 from DEGRADED, thirteen sessions dry since 07-27; DEGRADED clock still cannot start on a no-trade day.