The melt-up's concentrated core finally broke. South Korea's KOSPI fell 9.99% in a single session — twin circuit-breakers, a 20-minute halt, SK Hynix and Samsung each down about 12%, the two names that are roughly half the index and drove most of its 2026 gains. It is a direct hit to the book's largest position, the Korea AI sleeve. And the system's answer was: hold. Not because holding is comfortable — the book gave back nearly $12,000 and the drawdown deepened to −6.37% — but because every condition the protocol checks said hold. No hard trigger fired. The −8% mandatory de-risk line sits 1.63 points below, untouched. Credit spreads stayed flat through the entire crash, the unmistakable signature of a leveraged-positioning unwind rather than credit contagion or a fundamental break in AI demand. And by the close of the same session, the rebound had already begun — KOSPI +3.26%, Asia stabilising. This is the week the defensive posture, held through a fortnight of "why aren't we participating," earned its keep. The discipline is the edge: you do not liquidate a +37% winner into a forced-seller flush because the headline is frightening. You check the lines, and the lines said hold.
The single most important distinction of the session: credit spreads did not move. A −10% crash in the world's most AI-concentrated index, with IG OAS flat at 0.74, is the textbook signature of a forced unwind of a leveraged, over-appreciated position — not contagion, not a credit seizure, not a verdict on AI demand. SK Hynix had run roughly 190–350% on the year; foreign investors dumped ~$3.8B; retail bought the dip. The cause is concentration mechanics plus a misread HBM4 capacity-shift headline and an MSCI developed-market snub — all of which compress the price without touching the businesses. Regime stays CHOP, but the composition shifted: the bullish breadth leg that was building last week flipped bearish. CHOP deteriorating, where it had been CHOP improving.
EWY (Korea) is the book's largest position and largest P&L. The crash took it to 197.10 — still +37.2% from cost. The Korea AI sleeve has no −X% auto-exit, by deliberate design: you do not liquidate a concentrated winner into a single-session forced-seller flush. No hard trigger fired (the [v5-25] cap blocks adds, never forces trims). Credit-flat confirmed it was not systemic. And the same-session rebound (KOSPI +3.26%) validated the hold in real time. Selling into the −10% would have locked a forced-seller loss the day before the bounce.
Real yield ~2.26% via FRED (corrected from a stale 2.14% carried in prior sessions — verify via the function). The 10-year nominal yield dipped today, but that was a flight-to-safety bid from the tech crash, not easing — September hike odds jumped to ~68–75% (from 29% a week ago) and the 2-year is at its highest since February 2025. Higher-for-longer real yields compress AI valuations regardless of demand, and they are the through-line connecting the FOMC, the crash, and the GLD clock. At ~2.26% and rising, the [v5-1] full-exit at 2.50% is ~0.24 away — closer than prior sessions represented.
No trades. Marks from the June 24 15:15 UTC snapshot — $224,370, +12.19%, drawdown −6.37% vs the $239,644 HWM. The crash cost ~$11,900 in a session, concentrated in the AI sleeve. AI primary is 49.6% (still over the [v5-25] cap — the sleeve fell together so the ratio held), BIL 26.2% (above the 15% floor). The book is in the Amber drawdown band, worse than June 18's −1.40% but well off the June 10 trough of −9.7%.
| Ticker | Sleeve | Live wt | P&L | Note |
|---|---|---|---|---|
| EWY | AI / Tech | 19.2% · $43.0K | +$11,652 (+37.2%) | The crash epicentre — KOSPI −10%. Still +37% from cost. No auto-exit on concentrated winners; rebound +3.26% same session. |
| SMH | AI / Tech | 17.2% · $38.5K | +$11,487 (+42.5%) | Fell with the global semi complex. HBM sold out at 72% margins — demand intact, position unwound. Micron is the read-through. |
| TQQQ | AI / Tech | 11.6% · $26.1K | +$7,814 (+42.7%) | 3× leverage amplified the down day. Trimmed June 6; no rebuild — Bull sequence + JPY Zone 1/2 + real yield <1.80% all fail. |
| GLD | Debasement | 4.4% · $9.9K | −$2,296 (−18.8%) | Fell on a risk-off day — safe-haven failure. Clock: real yield ~2.26% → 2.50% exit, ~0.24 away. Hold, watch the line. |
| GDX | Debasement | 3.4% · $7.5K | −$2,254 (−23.0%) | The book's weakest position. High-beta miners taking the worst of the real-yield headwind. No add. |
| FXI | BRICS / China | 2.4% · $5.4K | −$635 (−10.6%) | China softer in the risk-off. Held. |
| VGK | Deglobal. | 7.2% · $16.1K | +$118 (+0.7%) | Europe held up better than Asia (Stoxx −1%). Amplify at EUR 1.20, ~0.07 away. |
| ITA | Defense | 4.6% · $10.2K | +$560 (+5.8%) | Oil sub-$70 puts the up-trigger out of play. Defense premium compresses on Iran resolution. Hold. |
| IBIT | Alt Monetary | 2.3% · $5.1K | −$867 (−14.5%) | BTC fell with the risk complex. Zone 3 blocks adds. |
| BIL | Liquidity | 26.2% · $58.7K | +$98 (+0.2%) | 26.2% live vs 33% target vs 15% floor. The dry powder that limited the drawdown. Redeploy only on a confirmed Bull reclaim — not into a crash. |
| CEG | AI Power | 1.7% · $3.8K | −$384 (−9.2%) | Rate-gated shut. Add gate (real yield <1.80%) fails at ~2.26%, further out post-FOMC. |
| Total · equity only | $224,370 | +$24,370 · +12.19% since inception · AI primary 49.6% (over cap) · BIL 26.2% · drawdown −6.37% vs HWM · −8% line 1.63 pts away | ||
For two weeks the book sat defensive through a melt-up, carrying 26% BIL and vetoing every AI add while the indices ran. The recurring question was "why aren't we participating." June 24 is the answer. The melt-up's concentrated core — Korean AI — cracked −10%, and the book was already de-risked: cash-heavy, adds blocked, no leverage chased. The drawdown stopped at −6.37% instead of the double digits a full-AI-exposure book would have printed. The cost of defence is underperformance in the melt-up; the payoff is survival when the melt-up's core breaks. This session is the payoff. The system traded certainty of survival for some forgone upside, and on June 24 it collected.
Micron beat and confirmed: the stock rose ~9% after the close, erasing the day's losses, with management citing an HBM4 ramp "tracking twice as fast" as the prior generation and over $1B of HBM4 revenue already booked. This is the strong-guidance branch. The crash thesis — an HBM/AI-demand doubt read off the SK Hynix capacity-shift headline — was directly refuted by the one company that is the cleanest read on that demand. The [v5-24] −8% line ($220,472) did not fire and is now much less likely to be tested: the catalyst that could have driven a second AI down-leg broke bullish instead. EWY and SMH should recover ground on the read-through, lifting the book off the −6.37% mark.
Crucially, this does NOT flip the regime to Bull. Micron resolves the AI-demand tail; it does nothing to the real-yield clock (2.27%, ≫1.80% gate) or the VIX/MOU gates. The downside tail closed — the structural rate headwind did not. Posture stays DEFENSIVE; this is "the bear catalyst failed," not "go risk-on." The 48-hour sequence — KOSPI crash, then Micron — was a stress test the protocol passed twice: it held through the positioning unwind, and the fundamental catalyst then vindicated the hold.
1.63 points away. The nearest hard line. A weak-Micron second leg below $220,472 fires it. The one thing that converts hold → forced action.
Corrected from a stale 2.14% carry. ~0.24 away and rising. Hawkish Fed, Sept hike odds ~68-75%. The structural headwind under gold and AI multiples.
Still over — sleeve fell together so the ratio held. Adds vetoed. Not a trim signal.
The decisive tell: −10% KOSPI with credit unmoved = positioning unwind, not contagion. The reason the answer is hold, not cut.
The crash began reversing the same session — Samsung/SK Hynix-led, retail bought the dip. Consistent with a one-day unwind, not a sustained Korea bear. Validates the hold.
US green (DJI +0.9%, NQ +0.5%), Nikkei +0.6%, only Brazil/DAX red. Day-after stabilisation, not a cascade.
The cash that capped the drawdown. Redeploy only on a confirmed Bull reclaim (MOU + real yield <1.80% + VIX <18) — none met. Not into a crash.
BRL 5.20, depreciated away from the 4.90 line; z-score near norm, not weak-and-turning. Absolute and relative agree: out.
One email when a new protocol session is published. Nothing else.
[v5-23] Session close confirmed: NO TRADES — all 11 positions held · book $224,370 (+12.19%) from June 24 15:15 UTC snapshot · EVENT: KOSPI −9.99% (8,203.84), twin circuit-breakers, SK Hynix/Samsung −12% — leveraged-positioning unwind, NOT credit contagion (OAS 0.74 flat) · rebound +3.26% same session · [v5-24] drawdown −6.37% vs $239,643.73 HWM — INSIDE the −8% mandatory de-risk line by 1.63 pts ($3,900 / −1.74% to the line at $220,472) · [v5-25] AI primary 49.6% over the 40% cap — adds vetoed (no forced trim) · [v5-2] JPY 161.80 Zone 3 · [v5-1] real yield ~2.26% (CORRECTED from stale 2.14 — verify via function), ~0.24 from 2.50% exit, rising · [v5-8] BRL 5.20 — both absolute and z-score lenses say stay out of EWZ · [v5-27] posture DEFENSIVE · MICRON WATCH: HBM guidance the swing factor; the only trigger in range is [v5-24] −8% if a weak-Micron second leg breaks $220,472 · credit via LQD/IEF (ERROR-010), oil WTI spot not USO (ERROR-009) · Bagheera ES101/TY60/CL14/GC6 · YTD [MISSING ×8]