Runs do() for every possible speed and conduct combination and ranks them by fault band distribution, severity, and liability. Each row is a pure causal query — back-door through Visibility is severed.
The do() operator evaluates one intervention at a time. The board never asks about one. They ask which of these three options produces the best expected outcome — and how confident the model is that the ranking is stable across the range of background uncertainty. That is structurally a different question. It requires running do() multiple times against the same evidence and same prior state, then comparing the resulting distributions.
The comparison view runs each intervention against the identical baseline and lays the resulting distributions side by side. Differences in central tendency are visible immediately; differences in tail risk and in sensitivity to specific upstream variables become visible after toggling the evidence. The decision rule the comparison supports is not "which alternative has the highest expected value" — it is "which alternative is robust enough to defend, and on what assumptions does that defence depend."
| # | do(Party B Speed) | Fault 0–25% | Fault 75–100% | Severity Severe | Liability High |
|---|