Weights concern tradeoffs, not importance alone

A question about which objective is important is not yet a question about the tradeoff represented by a weight. An organisation may regard safe operation as indispensable while every admissible alternative satisfies the same required condition. The importance of safety then belongs in eligibility and governance, rather than necessarily distinguishing the scored options. Conversely, a modest-looking attribute can carry a consequential difference across the candidate set. A useful weight concerns the value of the performance swing being compared, not the rhetorical prominence of the criterion name.

Swing weighting makes that distinction explicit. The facilitator describes a relevant movement from the least favourable to the most favourable performance on a criterion, while holding the other represented attributes at a specified reference. The decision maker compares the value of that movement with the swings on other criteria. Relative weights express those comparisons. The question is not how much the organisation cares about the word readiness. It is how much it values the particular improvement in readiness represented by the chosen scale, relative to the other improvements the decision can buy.

The performance ranges therefore belong in the weighting discussion. If the cost swing represents a small difference between already affordable proposals, its preference value may be limited. If it represents a difference that consumes much of the discretionary capital envelope, the same criterion name can describe a more consequential swing. A weight elicited for one range should not automatically carry over to another. Observed-range normalisation makes this especially important because the candidate universe determines the endpoints. The set and scale should be stable enough to support the elicitation being attempted.

Holding other attributes fixed is a thought experiment with a purpose, not a claim that every reference profile is physically feasible. It isolates the preference comparison. If the decision maker cannot value an improvement without knowing how another attribute performs, the interaction may matter. For example, improved monitoring information may be useful only when an operating team can act on it. A simple additive model can misrepresent that dependence. The facilitator should record the difficulty instead of forcing a clean weight from a question whose premise the participant rejects.

The relevant condition is about preferences, rather than statistical independence. Criteria can move together in reality while the organisation can still assess the value of their changes separately. Conversely, technically independent features can have joint preference value because one is useful only with the other. Clear definitions and an overlap review help, but they do not establish additivity by themselves. The committee should understand what kind of separability it is accepting when it uses a weighted sum. The app performs the sum; it does not test the committee's underlying preference structure.

A facilitation session should begin with the decision mandate and the scoring definitions. Participants should know whether they are advising an authorised committee, expressing different stakeholder perspectives or agreeing a collective preference record. Those activities can produce different legitimate outputs. A facilitator should not turn disagreement into a single weight vector merely because the software expects one vector at a time. The model can calculate each declared scenario separately, while the governance process decides which scenario, compromise or further discussion has authority.

The facilitator then presents the swings in concrete terms. Raw units and descriptions matter more than a generic common-scale endpoint. Participants should be able to explain what improves, what remains unchanged and why the improvement affects the objective. If they cannot interpret a swing, the criterion rubric or evidence perimeter needs work. Assigning an arbitrary importance rating at that point creates a number without resolving the ambiguity. The later precision of the score will conceal, rather than correct, the unresolved interpretation.

A reference swing can organise the discussion. Participants identify a particularly valuable movement and express the remaining movements relative to it, checking the comparisons for consistency. The resulting raw weights are normalised by the engine so that a common change of units in the weight vector does not alter the aggregation. Normalisation removes an arbitrary common scale; it does not validate the relative judgements. A weight vector can be internally well formed and still reflect an inappropriate mandate or misunderstood performance ranges.

The record should capture reasons, not only values. A preference for readiness may reflect a near-term service obligation, an unwillingness to commit before operating ownership is established or a need to preserve scarce delivery capacity. Those reasons have different implications when the context changes. The committee should be able to identify which reason justified the weight and what event would require reconsideration. A vector copied into later meetings without its rationale can become an accidental policy that nobody remembers authorising.

The facilitator should also check the meaning of a zero weight. It can express indifference within the relevant admissible range, or it can reveal that a concern has been handled as a gate elsewhere. It should not silently mean that the concern is unimportant in every context. In the dominance study, an impact-only vector leaves candidates tied on weighted preference even though their raw cost, readiness and risk differ. That result is an opportunity to test the intended indifference. It is not a reason for the engine to invent a preference the user did not supply.

The illustrative mandates are deliberately different. They allow the paper to test how an authorised change in priorities can alter preference without pretending that a proposal's performance has improved. An actual workshop would need an accountable process for adopting its vector. The committee might choose the balanced mandate, give priority to a particular strategic outcome or impose an initial-capital preference. The calculation can reveal the consequence of each choice. It cannot determine which mandate the organisation is morally, legally or strategically entitled to impose.

Sensitivity belongs after the initial elicitation as a challenge to interpretation. A participant may discover that a proposed weight produces a recommendation inconsistent with a stated tradeoff. That discovery can justify revisiting the weight or criterion definition, provided the reason is documented. It should not become an invitation to tune the model until a favoured proposal wins. The difference lies in the rationale. Correcting a misunderstood swing improves the preference record. Changing a weight solely to secure a desired outcome conceals the real decision.

The team should examine joint changes as well as the app's individual shifts when the decision requires it. The canonical sensitivity routine moves a single normalised weight and rescales the others. It provides a transparent local challenge set, not an exhaustive exploration of every legitimate preference vector. A coherent stakeholder mandate may change several weights together. The frozen mandate studies provide such comparisons through complete alternative vectors. Neither the individual tests nor the selected vectors establish an empirical distribution of preferences.

A facilitator should preserve dissent when it remains substantive. The record can state that a delivery owner places greater value on an executable near-term intervention while a strategy sponsor places greater value on a capacity outcome. Both may understand the criteria correctly. Their disagreement concerns priorities or authority, rather than an error in arithmetic. The committee can decide whether to adopt one mandate, seek a negotiated compromise or stage investments under different decision envelopes. An unexplained average hides the conflict while making the result appear consensual.

The session should finish with a weight-use statement. It identifies the decision stage, authorised perspective, criterion definitions, ranges and rationale governing the vector. It also names a change that would reopen the elicitation, such as a different candidate set, a revised cost perimeter or a material change in objectives. That statement gives the weights a controlled life. The organisation can then reuse the record when appropriate without treating a context-specific tradeoff as a universal preference policy.