PSD.5:4 - Solution

Build a question-led set of complementary models. For each, state its subject, intended use, assumptions, evidence, and what it omits. Compare only the overlapping claims whose meanings can be aligned, retain material differences, and return the usable claims together with their limits.

PSD.5:4.1 - Recover the questions and current boundary

Use the plural formulation set from PSD.3 where its differences identify claims or intervention logics that models must discriminate. A formulation is an interpretation of the problem, not already a validated model of the situation. Use the PSD.4 scope and reopen basis for the same subject, receiving decision, operating conditions, and horizon.

These are conditional inputs, not compulsory earlier stages. A qualified direct source may supply the needed formulation or scope. If a needed result is unavailable, stale, outside the present use, or incompatible, return its exact missing premise or narrow the inquiry; do not infer it from the presence of another model.

Turn the live differences into questions, such as “Does this arrangement retain service when road access is lost?” and “Whose loss is omitted by the aggregate service measure?” Keep question, assumed condition, and proposed intervention separate.

PSD.5:4.2 - Choose complementary contributions at the needed grain

Consider what each model can actually contribute before selecting a notation or tool.

Select the smallest set that covers the material questions. A qualitative model may reveal a question worth quantifying; a numerical model may reveal a missing qualitative distinction. Neither direction is a universal order. Drop a model that supplies no distinct contribution unless its independently grounded evidence is needed to challenge another.

A concern map can already complete the task of making participants’ concerns discussable. When another material question asks what causes an outcome or what an intervention would change, use an adequate causal model directly. If the causal relations still need to be constructed or compared, C.28.CM helps turn the relevant subject knowledge and mechanism accounts into explicit models for that question. Return their assumptions, conditional consequences and unresolved alternatives to the combined account. Keep each existing model for the question it can answer; the constructed causal model still needs the claim-specific assurance in :4.5.

When the question concerns participants’ intentions, their dependence on one another and possible alternatives, the Reference’s goals-and-dependencies walkthrough develops an actors-and-goals model from a service difficulty. It explains how to select what a view reveals, turn a question into analysis assumptions, interpret the answer and return to an omitted assignment when the result exposes a conflict. Use that connected account when these joins need explanation; a sufficient existing model can be used directly.

PSD.5:4.3 - Make each model’s use intelligible

A reader should be able to say: “This model concerns this subject, answers this question under these conditions, and supports this limited use.” Supply the assumptions, evidence source and interval, important omissions, and an observation that would invalidate or narrow that use. Identify the exact model content or edition when a changed value could alter the result.

For a mathematical formalism, simulation, or learned representation, apply C.29 when the choice of mathematical object, mapping, preserved structure, or information loss changes the claim. For example, a road graph can preserve routes and travel-time assumptions while omitting residents’ access rights and the reliability of an untested deployment procedure. C.29 does not establish those omitted facts.

Ordinary local meaning needs only the relevant meaning, units, scope, or evidence statement. Use A.1.1 when the decision needs to establish whether a model applies to a stated subject within a stated claim scope, whether an assignment holder actually uses the model during Work concerning that subject, or whether fixed model and expression contents satisfy a declared coherence criterion under a comparison scheme. Select its broader bounded model-use structure only when the organization of those relations changes the receiving decision. Neither a model collection nor the word “context” supplies that structure.

PSD.5:4.4 - Reconcile overlaps without forcing one model

For each overlap that matters, compare the subject and configuration, population, time horizon, units, definitions, assumptions, and evidence dependencies. Establish whether the models concern the same object or different related objects before treating them as views of one thing. For example, a travel-expense model and a repair-performance model can concern different work commonly called a business trip; matching that label does not equate their completion conditions. State an actual correspondence only as far as the objects and meanings support it. If “service failure” means lost pumping capacity in one model and inability to reach a refuge in another, keep the meanings distinct and ask whether a further relation can be established.

Classify a consequential mismatch before repairing it: different questions, different assumptions, incompatible meanings, contradicted evidence, or a genuine unresolved conflict. Correct a unit or input error locally. Retain alternative assumptions when their truth is not known. When the conflict can reverse the receiving result, compare an obtainable direct investigation with retaining the qualified alternatives or narrower answer through C.11.DUA. Use A.15.9 for a selected outside-practice contribution, including its whole acquisition burden.

Do not average unlike outputs, infer a shared probability from scenario counts, or count models drawing on one dataset as independent confirmation. A combined explanation may show how models inform one another without claiming that they form one unified model.

PSD.5:4.5 - Match assurance to the claim being used

Recognition can begin with a sketch and a plausible contrast. Before consequential reliance, inspect the claims that bear the result: source qualification, implementation or calculation correctness, fitness to the intended use, validation evidence, sensitivity, and extrapolation limits as applicable. The direct domain practice determines what is sufficient; this pattern supplies no universal validation threshold.

Separate confidence in the modeler, internal consistency, empirical adequacy, robustness to assumptions, and authority to decide. A stakeholder’s confidence can affect whether a model is used, but cannot substitute for evidence. A model that cannot yet support a recommendation may still support the narrower result “this conclusion depends on an unresolved assumption”. That limit does not itself select a test.

Return a qualified claim, a narrowed use, or an explicit inability to discriminate; include an evidence request when its contribution warrants acquisition. Where an unresolved safety, rights, legal, or technical premise blocks the proposed reliance, retain that stop. A stronger use needs its direct competent result; obtaining it remains a separate feasible and worthwhile continuation.

PSD.5:4.6 - Return the account and its reopen basis

Keep enough content for the recipient to recover:

• the subject, question, receiving use, configuration, and horizon;
• each selected model’s distinct contribution, current content, source basis, assumptions, and losses;
• the material correspondences, conflicts, and shared evidence dependencies;
• what the account supports, what remains hypothetical or unknown, and which alternatives or claims it can discriminate; and
• the observation that would reopen a model, its connection to another, the problem formulation, or the boundary.

This can be a short annotated comparison; it need not be a new repository or universal record form. If no model combination answers the question, say so. A missing direct result is more useful than a polished but unsupported combined answer.

The same account may inform Method selection, alternative generation, uncertainty treatment, or conflicts between simultaneous inquiries. Each recipient takes only the current, compatible contribution it needs; the account does not complete those later judgements.

PSD.5:4.7 - What changes in practice

Instead of asking which model is “the right one”, the practitioner asks which claims need a model, what each selected model contributes, and which difference would change the next move. Models can remain usefully plural while their overlaps and limits are explicit.