Connected calculations still apply
If COGS is modeled as a percentage of revenue, sampling the owning revenue assumption changes the linked COGS amount. Working capital, tax, financing and downstream valuation respond through their implemented relationships. You do not need to sample every downstream result to retain those links. Sampling the cost percentage as well introduces additional uncertainty in that ratio.
External shocks are independently sampled
The current system does not provide a correlation matrix, copula or common-factor configuration for sampled inputs. For example, independently sampling selling price and purchase price does not encode an assumed market correlation between them. Latin Hypercube improves marginal sampling coverage; it does not add that economic dependence.
Some valid sources are coupled by design
A supported 13W timing-profile change keeps the total profile at 100% through an offsetting component. This is a source constraint, not a statistical correlation setting. Do not configure overlapping representations as independent uncertainty. Likewise, a derived total and its owning Business Line are not two independent sources merely because both are visible.
Treat rankings as associations
The result ranking uses Spearman rank association between sampled inputs and the valid target outcomes. Its sign indicates association direction in that run, not a causal proof or a share of explained variance. Non-monotonic relationships can have a weak rank association despite an important effect. Use deterministic follow-up tests and source inspection to understand the mechanism.
Documents the current implemented behavior. Numerical examples are synthetic, not market assumptions.