A score can be gamed before the meeting

A preference process creates incentives around what it measures. Sponsors who understand the criteria may present their proposals in terms that improve the ratings, emphasise convenient evidence or define scope to avoid an unfavourable comparison. This is a plausible design concern, not an empirical claim about how often organisations behave. The review process should anticipate it because the calculation cannot distinguish a genuinely improved proposal from a strategically revised description. Evidence acceptance, scope control and independent challenge therefore belong beside the score.

The most direct failure is unsupported rating improvement. A sponsor may assert that a delivery plan is ready without demonstrating operating responsibility or may describe an uncertain outcome as high impact. The analyst should ask which evidence changed and whether the same rubric was applied to every candidate. A revised number should travel with a revised proposition and source. If the underlying evidence is unchanged, the record should explain why the interpretation changed. Otherwise the preference result can improve while the actual decision basis does not.

Scope splitting can create a subtler problem. A large intervention may be divided into a modest apparently ready component and an omitted enabling commitment. The scored component receives a low cost and favourable readiness while retaining a narrative about the whole programme's benefit. The repair is a consistent perimeter. The candidate should include the dependencies required for the claimed outcome or explicitly represent the present commitment as a stage with a narrower benefit. A small-looking option is not automatically comparable with a complete implementation package.

Metric substitution occurs when participants optimise the rating rather than the objective. A readiness rubric based only on document completion may reward paperwork without operating capability. An impact rubric based on activity volume may reward more activity without the desired outcome. The committee should inspect the mechanism linking the proxy to the objective and the conditions under which that link fails. The problem cannot be corrected solely by assigning a different weight. The underlying criterion definition may need revision.

Observed-range normalisation makes candidate admission especially consequential. Adding a reference proposal can change the scale even if it never becomes a serious choice. The range-artifact fixture intentionally adds a proposal that is worse on every represented raw attribute than relevant comparators. Original candidate values and weights remain fixed. The exercise demonstrates how the scores of retained candidates can move when the comparison universe changes. It does not claim that the leader must reverse or that any actual sponsor has used this manoeuvre.

In the illustrative baseline, Reliability architecture leads at 72.13 preference points. After the additional reference proposal changes the ranges, the same candidate leads at 89.98. The new reference proposal has a score of 0.00 and is dominated on the represented raw comparison. The original leader remains the leader. The higher score is a consequence of the new comparison scale, not evidence that its raw performance improved.

An inclusion rule limits this discretion. The committee should define what qualifies a candidate for the present universe before reviewing its effect on the score. A reference proposal can be legitimate when it is a credible available option within the envelope. It should not be admitted solely to improve the relative position of a preferred candidate. The review should document the source and rationale for the option, its scope and its disposition. If it is removed, recompute the entire comparison rather than retaining scores from the altered scale.

Independent rating review can help without pretending that judgement disappears. A second reviewer should examine a selected rating's evidence, rubric and perimeter before seeing the preferred outcome where practical. The review asks whether the interpretation is supportable and consistently applied. It does not need to invent an ostensibly objective number. Differences between reviewers can reveal ambiguity in the rubric or a genuinely uncertain proposition. Those differences should feed an evidence scenario or a definition repair rather than an unexplained average.

Version control is another useful restraint. A revised input should be stored as a new scenario with its reason and authorisation, while the original remains available. This makes outcome-driven changes easier to identify and legitimate corrections easier to defend. The record should connect the numerical difference to the changed evidence or mandate. A sequence of scenario files without reasons can still obscure the decision history. The platform's input-only vault supports reproducibility, but the committee record supplies the justification for each revision.

The team should test for favourable omissions as well as unfavourable values. A proposal may omit a necessary operating obligation or describe risk only within a narrow component. Required numerical fields cannot detect that omission. A perimeter checklist should ask which commitments, interfaces and dependencies are included and which are excluded. The same checklist should apply to every candidate. Consistency matters because a complete proposal may appear more costly or less ready than an incomplete one precisely because it acknowledges what implementation requires.

The analyst should also inspect how the story changes between proposal and briefing. A cautious estimate in a source document can become a confident rating in the model and a definitive statement in the recommendation. Each translation can lose qualifications. The evidence record should retain the original proposition, the interpretation supporting the rating and the scope of the resulting recommendation. A reviewer can then identify whether the briefing has overstated what the evidence supports. A polished output should not erase the uncertainty that matters to the commitment.

Governance can overcorrect. A rigid anti-gaming process may reject legitimate redesign or improved evidence because it treats every revision as suspicious. The purpose is to distinguish justified change from unexplained outcome tuning. A sponsor should be able to improve a proposal, narrow its scope or supply better evidence. The process should reward that work by evaluating the revised candidate transparently. It should preserve the previous version so the committee can understand what improved and what new obligations accompany the revision.

An efficient gaming review compares the pattern of revisions across candidates. A new readiness definition applied only to the sponsor's favoured proposal suggests asymmetric treatment; the same revised definition applied to every comparable proposal may be a legitimate correction. A cost exclusion that appears in every scope is different from an exclusion that appears only after one candidate loses. Inspect the timing and distribution of the changes before inferring intent. The decisive evidence is the comparison rule and its application, rather than the length of the edit log. A suspected bias can therefore be challenged precisely without treating every late improvement as misconduct.

The committee should be willing to reject the model use when its incentives overwhelm its evidential discipline. If sponsors cannot supply comparable scope, if ratings are negotiated to achieve an outcome or if the universe is repeatedly tuned without reason, a numerical ranking may conceal more than it reveals. The remedy can be a qualitative options review, a staged evidence process or a different appraisal method. Retaining a convenient score is not the overriding objective. The process should preserve a decision that can be challenged on its actual basis.