| How should models contribute jointly without hiding their different meanings? | Select models by the questions they answer and inspect the joins between problem structuring and analysis. | One comprehensive model is the serious default; an unconstrained collection is the rival expansion. | Adapt: PSD.5:4.1–4.4 choose the smallest complementary set and expose incompatible meanings. Compared with expanding one model, this deliberately accepts some correspondence work to preserve a material concern that the single model omits; it makes no universal claim of lower cost. | Marttunen, Lienert, and Belton’s 2017 review of PSM–MCDA combinations is a critical synthesis candidate for combination benefits and interface difficulties, including value-tree and weighting issues. Its reviewed applications are not controlled evidence that more models are better or that one combination fits every inquiry. The question-led selection and non-use rule are PSD adaptations. | Reopen if a simpler model answers the same material questions with equivalent limits, or a new mismatch shows that the chosen combination loses a consequential meaning. |
| What justifies relying on a model-supported claim? | Qualify the intended use, evidence, and losses separately from confidence in the model or its producer. | Technical sophistication, modeler reputation, or participant confidence is treated as sufficient credibility. | Adapt: PSD.5:4.3–4.6 and the flood-pump case retain applicability, evidence dependencies, and unresolved assumptions. The extra assurance effort is accepted only where it can change reliance; sketches remain usable for recognition. | Schwarzburg, Trauer, and Rebentisch’s 2024 confidence study supplies bounded empirical evidence that model-, modeler-, and stakeholder-related factors are associated with confidence and reliance. Its exploratory survey and proposed application model do not validate a particular decision model or establish causal decision quality. Current C.29 supplies mathematical mapping and loss discipline, not validation, causality, or authority. A.1.1 supplies the three model relations used above; it applies only when their claims matter to the decision. | Reopen when direct validation contradicts a relied-on claim, the use leaves its qualification window, or a current modeling practice offers the same assurance at materially lower effort. |