Appendix: source scope and a dominated-option reversal

The sources in this edition are public primary guidance and official publisher records. They serve as narrow methodological or institutional context. The manuscript's decision architecture, fictional records and fixture interpretations are original analysis. No source is treated as empirical evidence of the performance of the fictional proposals. The distinction matters because a public manual can explain a method without validating its application to a private capital decision. Each source card retains a locator and scope so the reader can identify exactly what support is being claimed.

Institutional guidance retains its original audience and authority. A public-sector appraisal manual does not automatically impose a private committee's decision rule. Corporate-governance principles do not replace the duties applicable under a particular jurisdiction. Environmental and social standards do not become a universal legal conclusion merely because the paper cites them. The manuscript uses these materials to identify questions and boundaries, while actual applicability requires the relevant institution, agreement and process. A preference score cannot perform that applicability assessment.

The methodological manuals support a similarly bounded role. They provide context for scales, tradeoffs, uncertainty and analytical selection. They do not establish that the app's fixed criteria are sufficient for every investment or that a particular vector is legitimate. The author identifies where the implemented additive comparison follows a clear operation and where it lacks representation. The conclusion is therefore conditional on decision fit. A source citation should strengthen a defined proposition, rather than create a general aura of endorsement around the whole workflow.

The publisher record for the economic-appraisal book has an especially narrow receipt. This edition retains the readable official description and uses it only for the stated distinction between economic value and investor financial profitability. It does not claim to have reviewed the full book or import unseen detailed procedures. The card's scope makes that boundary explicit. A title, summary or search result should not be expanded into a claim about contents that the author has not actually read.

The claims ledger gives numerical statements a defined epistemic status. Supplied fixture values are illustrative assumptions, while computed values are illustrative results from the canonical engine. Narrow primary guidance notes are sourced qualitative propositions. Original process reasoning is not labelled as measured evidence. These categories prevent the document from sliding between mathematical demonstration, institutional guidance and project findings. A reader should be able to identify which kind of statement is being made before deciding how it can inform a real action.

A narrow citation is not a substitute for honest intellectual attribution. If a paragraph derives substantially from a source, its use should be classified and counted accordingly rather than hidden under an original-analysis label. Conversely, a long original decision framework should not be attached wholesale to several sources as though each supports every word. The author should separate the primary proposition from the reasoning developed around it. This makes the source audit meaningful and the reader's understanding more precise.

Source receipts, the numerical ledger and actual edition status accompany the publication. They establish different properties: access, reproducibility and recorded review. For a new capital decision, establish current institutional applicability and replace fictional assumptions with verified proposal evidence. The following controlled experiment concerns the behaviour of a preference scale; it does not claim that any capital commitment is feasible or approved.

An additional proposal can change a recommendation even when it has no realistic claim to approval. This happens when the comparison set also defines the value scale. To isolate the mechanism, retain the original proposals, their raw criteria and the committee weights, then add a clearly dominated option. If the original ordering changes, the committee has learned something about the representation of preferences rather than the merits of a new investable proposal. The test complements the earlier range-artifact study, which moved scores without changing its leader. A non-reversal in one experiment does not establish invariance across all admissible sets.

The original fictional pair has Lean retrofit at USD10.00m with impact 60.00, and Deep renewal at USD20.00m with impact 90.00. Both retain readiness 50.00 and execution risk 50.00 on the app scales. Cost receives 60.00 of the declared weight total and impact 40.00; readiness and risk each receive 0. These are supplied preferences and illustrative ratings. They are not measured benefits, estimated project cash flows or a willingness-to-pay valuation.

Illustrative. On the original set, Lean retrofit leads at 60.00 preference points, while Deep renewal receives 40.00. Add Screened option at USD100.00m with impact 60.00, retaining the same readiness and risk. The original proposals and weights remain byte-for-byte unchanged. The extended set now puts Deep renewal first at 93.33, followed by Lean retrofit at 60.00. The added option receives 0.00. Its presence changes the range used to interpret the original cost difference even though it adds no non-dominated choice.

The mechanism is visible in the contribution tables. On the original pair, the cost endpoints are the costs of the eligible proposals, so the entire cost swing is their difference. The added expensive option extends only the upper cost endpoint. Lean retrofit keeps its best-cost endpoint, while the additional cost of Deep renewal now occupies a much smaller fraction of the expanded cost range. The impact range is unchanged, so the advantage of Deep renewal on impact retains its full normalised swing. Constant readiness and risk contribute nothing under the supplied zero weights. The reversal follows from this asymmetric rescaling; no improved delivery evidence or new stakeholder preference was supplied.

Raw dominance provides a useful cross-check. Both original options weakly outperform Screened option on every declared criterion and strictly outperform it on cost. The original pair remain non-dominated because their cost and impact advantages run in opposite directions. The non-dominated set is therefore unchanged, while the preference leader changes. A dominance screen and a preference score answer different questions. Keeping both visible prevents the committee from interpreting a dominated option as new strategic opportunity merely because its inclusion has changed the score of another proposal. It also exposes the consequence of including a screened-out option in endpoint construction.

The immediate response is to freeze the eligible comparison set before eliciting and approving range-dependent tradeoffs. A late proposal should trigger an explicit decision about eligibility, endpoints and preference review. Quietly adding it to the model and retaining the same verbal weight labels hides a material change in the meaning of a cost point. The committee record should distinguish a genuine change in the opportunities available from a change caused by the representation. Both may justify reconsideration, but they require different evidence and different explanations to the authority that owns the decision.

Externally anchored value functions can avoid dependence on an incidental comparison set, but choosing anchors is substantive work. They need a defended service boundary, scale shape and tradeoff interpretation. An upper cost anchor should not be selected simply to preserve a preferred leader. Anchors far outside the useful decision range can suppress a criterion just as an irrelevant alternative does. If a fixed-anchored model is appropriate, build and validate it as a separate method with its own assumptions and evidence. The current app implements endogenous min-max scaling and should continue to describe that operation accurately.

The selected weight diagnostics do not close this question. They keep the alternative set fixed while changing a weight, so they examine another uncertainty axis. A stable leader across those shifts can still be unstable to eligibility or endpoint changes. Conversely, this constructed reversal does not show how often a real committee would encounter the problem. It establishes a possibility under declared inputs. A fuller stability programme should test the plausible option universe, admissibility rules, evidence intervals and preference disputes that matter to the actual mandate, retaining coherent cases instead of inventing a probability from the number of favourable runs.

The useful committee packet contains the original pair, the expanded set, the exact criterion ranges, contributions and raw dominance results. Keep the unchanged original rows visible so a reviewer can verify the controlled comparison. State which options are eligible for commitment and which merely test the method. The recommendation should explain whether the committee accepts range-dependent tradeoffs or requires a separate anchored representation. Record the authority that can reopen the mandate and the evidence that would cause it to do so. The result is a decision about the decision method before any authority commits funds.