AI portfolio arbitration
A ranking is sound only when its possible reversals are known, outstanding disagreements remain visible and the final choice has an accountable owner.
The pivot scenario summarises measurements and preferences. It does not establish that the first-ranked option will remain first when evidence, weights or constraints change. Sensitivity analysis searches for those reversals.
Measurement range does not mean the same thing as preference disagreement. Both are tested, but they retain distinct owners and treatments.
| Source | What varies | Treatment |
|---|---|---|
| Measurement | Data, population, error or extrapolation | Justified low, central and high values |
| Appraisal | Different readings of the same evidence | Separate scenarios and reason for disagreement |
| Preference | Relative value of the swings | Weight sets defended by the authorities |
| Context | Budget, horizon, baseline or regulation | Named scenario and new version where required |
Each test changes one identified assumption and retains the rest of the model. Plausible variations are documented before their effect is interpreted.
A switching value expresses the evidential effort or preference change required to reverse the ranking. If it is impossible, the tested assumption is not decisive; if it is close, it becomes the next question to examine.
The verdict states what was tested and what was not. It never claims to cover scenarios absent from the record.
The committee receives the ranking, its conditions and limitations. The authority may select another option, but it must state the reason, consequences and follow-up control.
The record retains inputs, value functions, weights, scripts or calculation files, scenario results, divergences, signed decision and versions. A decimal without this trail is not evidence.
Primary sources
Last documentary review: 7 September 2026.
Value functions, swing weighting, decision conferencing, sensitivity and ranking robustness.
Swing weighting, sensitivity analysis, switching values and the warning against simple weighted scoring.
Govern, map, measure and manage the risks of an AI system.
Assessing the impacts of an AI system on individuals, groups and society.
Before you write to us
It is the smallest change in a datum, value function or weight that reverses a ranking or changes the decision.
When it remains unchanged across the plausible variations actually tested, all gates pass and its limitations are documented.
Yes. It must then record the selected option, accepted trade-off, determining assumptions, conditions and the authority taking responsibility for the decision.
Robustness
Present the ranking, assumptions and disagreements. We identify the switches to test before commitment.
Stress-test the ranking