Lookahead simulation

Naive lookahead at depth H across N iterations grows combinatorially impossible. If at each of H days the system branches into N futures, you're looking at N^H trajectories — a million-fold blowup at H=10, N=10. The N^H trap is what kills most attempts at decision-level stress testing.

VYAN's approach is N×H, not N^H. Each of N iterations runs as a single trajectory advancing through H days, with each iteration making its own decisions against its own sampled uncertainty path. A trajectory at day 5 doesn't branch into N new trajectories; it continues, samples its day-6 variability, makes its day-6 decisions, advances. Linear cost, not combinatorial. Tractable on commodity compute.

As each trajectory advances, it records which Day-1 decisions held versus which broke. The result: every Day-1 decision carries a resilience score computed from Day-1 consensus (how many iterations agreed on it at commit time) times future stability (how many of those still hold through H). A decision that 91% of iterations agreed on at commit but only 38% still hold at Day 12 carries a 91% consensus, 38% Day-12 stability — and a named break day at Day 12.

Break-day attribution traces back to the specific iteration cluster that caused the break — "Supplier S2 at P85 lead time AND demand at P75 simultaneously, firing at Day 11." The planner sees the cliff before commit. The mitigation — shift the sourcing split from 70/30 to 60/40 — is a defensible decision rather than a hunch.

EVW (Expected Value of Waiting) is the related economic primitive. For every supply event near its release window, lookahead computes the expected value of waiting one more day before committing — what's the dollarized cost of premature commit versus the dollarized cost of delay? The EVW informs the auto-commit logic; high-EVW decisions surface for human review even when value at risk is below the autonomy threshold. Chapter 5.4 develops EVW in dollar terms.