Decision object model
Keep states distinct.
Evidence, inference, modeled consequence, recommendation, proposal, approval, execution, confirmation, and closure are different objects. A later state references the earlier one; it does not rewrite it.
Evidence & method
A decision is credible when its construction can be inspected.
CrisisFlow exposes how evidence becomes an exposure path, how bounded consequence scenarios are constructed, how sensitivities are tested, how portfolios are screened, and where validation remains open.
Matching hierarchy
Resolve before inferring.
Use stable identifiers and authoritative coordinates or polygons first. Then apply hazard-appropriate administrative, distance, corridor, entity, product, jurisdiction, route, or technology relationships. Material ambiguity remains unresolved.
Consequence construction
Publish the estimator.
Show inputs, units, evidence state, and transformation rule. Low, base, and high are explicit modeled scenarios, not percentiles. Decision-switch sensitivity shows where the preferred portfolio changes while other inputs remain fixed.
Portfolio formulation
Optimize the decision, not a ranked list.
The public case enumerates all 32 subsets of five actions. It minimizes a declared weighted shortfall objective, screens eight constraint classes, and applies deterministic tie-breakers to the eligible set.
Reproducibility
Same release, same result.
Every case identifies its source version, input manifest, formula version, and release hash. The calculation is deterministic and local: unchanged schema-checked inputs produce unchanged outputs.
Bounded formulation
One objective. Five binary actions. Thirty-two complete portfolios.
No solver, remote model, or lookup table is used in the public case. Every subset is evaluated in the browser against the same disclosed rules.
Objective / lower is preferred
3 × priority missed units + standard missed units- Decision variables
- A, B, C, D, E ∈ {0, 1}
- Hard screens
- Budget · timing · capacity · prerequisite · incompatibility · authority
- Tie-breakers
- Objective → cost → time to effect → portfolio key
- Solution status
- Exact within the displayed five-action set
Validation boundary
Open questions belong inside the method.
The demonstration shows traceability, deterministic behavior, and constraint handling. It does not establish predictive accuracy, causal action effectiveness, universal network coverage, product superiority, or improved customer outcomes.
- How stable are consequence ranges under alternative planning assumptions?
- Which action-effect assumptions require customer or operating validation?
- Which missing dependencies could change exposure or portfolio eligibility?
- When does a pattern transfer across operating contexts, and when should it not?
Technical diligence without a black box.
Inspect the calculation, then inspect its boundaries.
The Technical Methods Note expands the decision objects, matching rules, range construction, portfolio formulation, reproducibility controls, validation scope, and selected research references with complete citations.