Which retrofit packages can become preferred? Exact weight-space geometry and SMAA-2 rank acceptability for an anonymised nonresidential building

A package ranking based on one set of criterion weights can conceal how strongly the conclusion depends on preferences. This study applies a deterministic-criteria version of stochastic multicriteria acceptability analysis (SMAA-2) to five alternatives in an anonymised energy consultancy report. Total investment, annual CO2 and annual NOx are minimised after observed-range normalisation. Preference weights are explored over the three-criterion simplex. Uniform-weight first-rank acceptability is calculated exactly by polygon clipping and checked against 200,000 Dirichlet draws; all-rank acceptability and hypothetical preference-emphasis profiles are also evaluated. Exact first-rank shares are 21.998% for P2, 68.696% for P3 and 9.307% for P4, while P1 and P5 have no positive-area winning region. P5 is dominated by P3 in the selected criteria. P1 is nondominated but unsupported by the additive model, a distinction explained by a hypothetical convex combination of P3 and P4. P5 nevertheless ranks second in approximately 66.84% of uniform draws, showing why first-rank acceptability alone is incomplete. Cost-emphasised weights shift the most frequent winner to P2. These shares describe an imposed distribution over preferences, not stakeholder survey results or probabilities of successful renovation. Source alternatives are treated as separate model states, and a common reference-area normalisation avoids silently adopting inconsistent printed per-area denominators. The study provides a reproducible preference audit rather than an unconditional investment recommendation.