What the synthetic portfolio scores mean

This qualification accompanies a retained reference calculation. It does not replace its original model card or claim a fresh platform execution.

A source ambiguity, made explicit

The original model card describes “ordinal/value-model” outputs, but the implementation adds contributions and divides scenario scores by an optimum. Arbitrary ordinal ranks do not justify those operations. The website therefore uses the narrower interpretation below. This is a declared demonstration assumption, not a claim that the original wording was sufficient or that customer preferences were elicited.

The interpretation contract

Candidate benefits are constructed cardinal demonstration points. Zero denotes no modeled benefit contribution. Contributions are additive within each scenario, without benefit interactions beyond the stated selection rules. A common constructed unit across scenarios permits the secondary absolute-regret comparison. None of this demonstrates monetary value, ROI, probability, customer utility or a scale-independent ranking.

The primary rule maximizes the minimum ratio of portfolio value to that scenario’s feasible optimum. No tie at that primary score affects the selected result; the secondary regret and cost rules do not decide the winner here. The underlying costs, staffing, dependencies and feasibility set stay fixed in all four scenarios. Only the constructed benefits change.

The choice gives something up

The selected mix uses $0.5M and two staff-months more, relinquishes three balanced-scenario points, and improves the weakest relative result by approximately six percentage points. It is justified only under the stated objective, constraints, scores and zero—not as a universally best investment mix.

Retained results and comparator · Unchanged original model card