A fictional steel mill can change vendor, automate, suffer industrial action or face a market shock. Draw the possible moves, then ask: which situations can it leave, and where might a chosen walk spend its time?
Try this · model · limits · sources
Try. Run 250 steps. The two exits from Baseline Ops have weights 0.35 and 0.15, so the model chooses them with chances 70% and 30%. Raise the cutoff to 0.30: Automation loses its 0.10 route to Market Shock. Raise it to 0.31: Automation has no active exit and the walk stays there. Undo an edit or perturbation to compare.
Model. Nodes are situations, not tasks. Viability is your assumed score from 0 to 1. Positive outgoing weights that meet the cutoff are normalised for each move. Closed classes are strongly connected groups without an active exit; single states with no exit are included. These are properties of the supplied graph.
Result. Dashed outlines show closed classes; thicker outlines and links show visit counts. The table provides the same information and editing controls. A closed class can contain high or low scores; being visited often does not prove resilience. A cutoff changes the model used by both analysis and walks.
Limits. Invented weights and scores are not calibrated probabilities, safety assessments or forecasts. A model perturbation is reversible here; it does not establish that an intervention would be safe in a project. There is no viability kernel, control policy, causal validation, resource or schedule model. With impacts enabled, visits change the destination's score, but do not change transition weights.
Basis. Grinstead & Snell, Introduction to Probability, chapter 11 defines transition probabilities and absorbing states. NetworkX's attracting-components reference describes closed strongly connected components. These sources support the mathematics, not the fictional assumptions. Adapted from an earlier example. Retention and validation notes.