HScreen · risk

Co-crash how often two names have their worst day together (lower-tail dependence λ_L)

if 005930.KS has its worst day, 38% of the book is expected to have its worst day too

pair	λ_L	pearson	used	obs
NVDA\|MSFT	0.4549	0.5659	0.5659	1275
005930.KS\|8035.T	0.4007	0.4761	0.4761	1173
NVDA\|MU	0.4	0.5368	0.5368	1275
MSFT\|IBM	0.298	0.2352	0.298	1275
005930.KS\|MU	0.2917	0.2132	0.2917	1200
VICI\|MSFT	0.2667	0.244	0.2667	1275
VICI\|IBM	0.2667	0.2429	0.2667	1275
MU\|MSFT	0.251	0.3244	0.3244	1275
8035.T\|MU	0.2176	0.1787	0.2176	1195
MU\|IBM	0.2039	0.1253	0.2039	1275
NVDA\|VICI	0.1961	0.1622	0.1961	1275
NVDA\|IBM	0.1961	0.1511	0.1961	1275
005930.KS\|NVDA	0.1917	0.1095	0.1917	1200
8035.T\|MSFT	0.1674	0.0807	0.1674	1195
MU\|VICI	0.1647	0.0953	0.1647	1275
8035.T\|NVDA	0.1423	0.1392	0.1423	1195
8035.T\|IBM	0.1423	0.02	0.1423	1195
005930.KS\|MSFT	0.1417	0.0037	0.1417	1200
8035.T\|VICI	0.1339	-0.0312	0.1339	1195
005930.KS\|VICI	0.125	-0.011	0.125	1200
005930.KS\|IBM	0.1083	-0.0267	0.1083	1200

λ_L exceeds average correlation on 17 of 21 pairs — average correlation understates how these names behave in a bad tail, which is the only time it matters. The optimiser takes the larger of the two. Hatched rows are on the conservative default for want of history.

The chain whose failure propagates furthest, and what to hedge first

DAG assertion holds

node	role	systemic	between	reach	hedge priority	we hold here
foundry-logic	component	1.0	1.0	0.1667	1	yes
dc-infra	infra	0.8571	0.8571	0.25	2	no
semicap-equipment	capex_supplier	0.3333	0.0	0.3333	3	yes
power-thermal	infra	0.3333	0.0	0.3333	4	yes
passives	upstream	0.3333	0.0	0.3333	5	no
memory	component	0.25	0.1429	0.25	6	yes
networking	component	0.1667	0.0	0.1667	7	no
packaging-substrate	component	0.0	0.0	0.0	8	no
power-semis	infra	0.0	0.0	0.0	9	no
hyperscaler	demand	0.0	0.0	0.0	10	yes
neocloud	demand	0.0	0.0	0.0	11	yes
cn-ai-semi	parallel_chain	0.0	0.0	0.0	12	no
materials	upstream	0.0	0.0	0.0	13	no

systemic_importance = max(betweenness, downstream_reach). Betweenness ALONE scored semicap-equipment 0.0 — an EUV monopoly with zero paths through it. A chokepoint can be an intermediary OR a source; one measure misses half of them.
Edge source: policies/supply_chain.yaml v2 — NOT Futu get_industrial_chain_*, which does not exist in this SDK (verified: only get_owner_plate / get_plate_list / get_plate_stock). Our own graph carries dated rationales per edge, which a vendor

Scenario board

not built — premortem scenarios, which positions share each wound, and which hedges cover it
needs: premortem_scenarios.json — skill B8, not supplied

Exposure ledger

alpha gross78.97%

alpha chain ai_capex vs cap53.58% / 70%

alpha chain unchained vs cap0.0% / 70%

alpha chain null vs cap6.0% / 70%

income gross73.0%

income chain unchained vs cap100.0% / %

income chain null vs cap14.0% / %

income chain ai_capex vs cap4.0% / %

crowding dataabsent

crowding-visibility tightening requires desk-positioning's CROWDED_LONG flag; it does not exist yet, so the trigger evaluates FALSE and is logged as absent rather than failing open (the theme_of lesson applied)

Rendered from canonical artifacts only (T1) — no terminal database exists. Every figure is read from the system's own files at render time; where an artifact is missing the panel names the dependency rather than showing an empty box. No execution path exists anywhere in this interface (T5), not even a link.