Method · multi-criteria decision

TRARDI Impact Score™: build a defensible arbitration

A method transparent by construction: visible assumptions, reproducible calculation, retained disagreements and explicit limits. Here is the full protocol, from value functions to sensitivity.

Dr Youssef Trardi·12 min·Updated 7 September 2026

An index decides nothing

The TRARDI Impact Score™ is transparent when every input, transformation, weight and exclusion can be examined and recalculated. It replaces neither the decision authority, the evidence nor the risk review. It compares options pursuing the same objective, over the same horizon and against the same baseline.

The final number is not the decision. The record must allow a reviewer to trace the rank to every value, each value to its evidence and each assumption to the person who accepted it.

When an additive model is admissible

Weighted addition is retained only if four conditions can be defended before scoring. Otherwise, we preserve an axis-by-axis reading or use another method.

Comparability: The same decision, objective, horizon and baseline scenario.
Preferential independence: The value of a swing between two levels does not depend on the levels taken by other axes; overlaps are removed.
Measurability: Every value maps to an observation and anchors defined before the cases.
Acceptable compensation: A gain on one axis may offset a loss on another; otherwise the condition becomes an exclusion.

Build the value functions

Raw data does not enter the index directly. Each measure is converted to a common 0–100 scale through a documented value function. Anchors represent the worst and best states useful to the decision — not a universal ideal. Missing evidence blocks ranking; calculation does not turn it into certainty.

Axis	Observed variable	Anchor 0	Anchor 100
Business impact	Expected change in an attributable outcome, compared with the baseline	No defined outcome or no defensible causal link	Effect measured on a comparable scope, with attribution and interval
Data readiness	Coverage, quality, rights, lineage and stability of required data	Critical data absent, inaccessible or lacking defensible rights	Population covered, rights established, versions and drift monitored
Operational feasibility	End-to-end performance under load, incident and change	Critical dependency untested or operations undefined	Service proven, recovery tested and dependencies reversible
Adoption readiness	Process integration, owner, recourse and team capability	Target process, owner or recourse absent	Use embedded, recourse exercised and autonomy demonstrated
Risk and control maturity	Coverage of risks by testable, assigned controls	Critical risk without a control, owner or stop authority	Controls proven, monitored and reviewed after change

Business impact

Observed variable: Expected change in an attributable outcome, compared with the baseline
Anchor 0: No defined outcome or no defensible causal link
Anchor 100: Effect measured on a comparable scope, with attribution and interval

Data readiness

Observed variable: Coverage, quality, rights, lineage and stability of required data
Anchor 0: Critical data absent, inaccessible or lacking defensible rights
Anchor 100: Population covered, rights established, versions and drift monitored

Operational feasibility

Observed variable: End-to-end performance under load, incident and change
Anchor 0: Critical dependency untested or operations undefined
Anchor 100: Service proven, recovery tested and dependencies reversible

Adoption readiness

Observed variable: Process integration, owner, recourse and team capability
Anchor 0: Target process, owner or recourse absent
Anchor 100: Use embedded, recourse exercised and autonomy demonstrated

Risk and control maturity

Observed variable: Coverage of risks by testable, assigned controls
Anchor 0: Critical risk without a control, owner or stop authority
Anchor 100: Controls proven, monitored and reviewed after change

Weights measure a swing, not the importance of a word

Asking whether “impact” matters more than “data” does not produce a defensible weight. We weight the value of moving between the two anchors on each axis — the swing — for the decision being examined.

Set the range — establish the genuinely plausible worst and best state for every axis.
Choose the reference swing — identify which move from 0 to 100 would create the greatest value for the decision.
Compare the other swings — assess their relative value without looking at the candidates yet.
Normalise — scale the judgements to a total of 100% and retain the intermediate calculations.
Challenge — record disagreements, the deciding authority and the alternative weights to test.

Exclusions precede any compensation

A non-compensatory condition must not be hidden inside a low score. It becomes a decision gate examined before any ranking.

Observed condition	Effect	Evidence required to reopen
Purpose, affected population or delegated decision undefined	Exclude	Approved boundary and named decision authority
Right to use critical data not established	Defer	Legal basis, provenance, licence and retention verified
Unacceptable risk without an effective mitigating measure	Stop	Tested control and residual risk accepted by the competent authority
Critical dependency without recovery or exit route	Defer	Recovery test or exit scenario run on the real scope
Benefit supported only by a supplier claim	Defer	Independent measure compared with the baseline

Purpose, affected population or delegated decision undefined

Effect: Exclude
Evidence required to reopen: Approved boundary and named decision authority

Right to use critical data not established

Effect: Defer
Evidence required to reopen: Legal basis, provenance, licence and retention verified

Unacceptable risk without an effective mitigating measure

Effect: Stop
Evidence required to reopen: Tested control and residual risk accepted by the competent authority

Critical dependency without recovery or exit route

Effect: Defer
Evidence required to reopen: Recovery test or exit scenario run on the real scope

Benefit supported only by a supplier claim

Effect: Defer
Evidence required to reopen: Independent measure compared with the baseline

Recalculate the index, value by value

The three use cases have passed the exclusions. The same weights are applied to each; every value remains visible and the result can be reproduced.

Use case	Impact	Data	Feasibility	Adoption	Controls	Index	Reading
A — Call transcription and CRM update	70	85	80	70	65	74.25	Portfolio reference
B — Legal contract review assistant	90	35	70	55	30	60.75	High value, weak evidence
C — Generative design for a production line	55	40	35	50	45	45.75	Non-priority investigation

A — Call transcription and CRM update

Impact: 70
Data: 85
Feasibility: 80
Adoption: 70
Controls: 65
Index: 74.25
Reading: Portfolio reference

B — Legal contract review assistant

Impact: 90
Data: 35
Feasibility: 70
Adoption: 55
Controls: 30
Index: 60.75
Reading: High value, weak evidence

C — Generative design for a production line

Impact: 55
Data: 40
Feasibility: 35
Adoption: 50
Controls: 45
Index: 45.75
Reading: Non-priority investigation

Additive model

Index(j) = Σ [weight(i) × value(i,j)]

Scale: 0–100
Weight: ≥ 0
Total: 100%

Weights

What would change the conclusion

Sensitivity analysis seeks the smallest variation capable of changing the ranking or the decision. It identifies the assumption that must be tested before committing more resources.

Assumption tested	Variation	Calculated effect	Conclusion
Relative value of impact and data	Weights 30/20 replaced by 20/30	A = 75.75 · B = 55.25	The first rank does not change
Data and control evidence for case B	Data 35→70 · controls 30→65	B = 73.0	B remains behind A at 74.25
Prior evidence and adoption for case B	Additional change: adoption 55→70	B = 75.25	B moves ahead of A: the rank switches
Data rights for case B are not established	Decision gate triggered	Index excluded from the decision	B is deferred regardless of rank

Relative value of impact and data

Variation: Weights 30/20 replaced by 20/30
Calculated effect: A = 75.75 · B = 55.25
Conclusion: The first rank does not change

Data and control evidence for case B

Variation: Data 35→70 · controls 30→65
Calculated effect: B = 73.0
Conclusion: B remains behind A at 74.25

Prior evidence and adoption for case B

Variation: Additional change: adoption 55→70
Calculated effect: B = 75.25
Conclusion: B moves ahead of A: the rank switches

Data rights for case B are not established

Variation: Decision gate triggered
Calculated effect: Index excluded from the decision
Conclusion: B is deferred regardless of rank

Make the arbitration reproducible

A third party must be able to rebuild the result without asking what a cell meant. The record retains at least six versioned objects.

Decision frame — objective, options, horizon, baseline, constraints, authority and date.
Evidence register — source, population, version, owner, quality and limits of use.
Raw assessments — measure, value function, interval, disagreement and rationale by axis.
Calculation model — formula, weights, rounding, executable file and version fingerprint.
Robustness tests — scenarios, switching values, dominance and triggered gates.
Signed decision — selected rank, conditions, owner, expected evidence and review trigger.

What the score cannot claim

The index is neither a return on investment, a probability of success, causal proof nor a conformity assessment. It is comparable only across options examined within the same decision frame.

The calculation is rerun when a material change affects the objective, population, data, supplier, process, risks, controls or available resources. A cosmetic change does not justify a new score; an assumption that has become false does. Sensitivity envelopes are not confidence intervals, and no empirical result yet supports presenting the index as a predictor of project success.

The TRARDI Impact Score™ is a proprietary decision-support method. The cited references support its construction and traceability rules; they are neither an endorsement of the method nor a certification of the result.

Sources and references

UK Government — practical multi-criteria analysis guidance (new tab)
Value functions, swing weighting, decision conferencing, sensitivity and ranking robustness.
NIST AI RMF 1.0 (new tab)
Govern, map, measure and manage the risks of an AI system.
ISO/IEC 42005:2025 (new tab)
Assessing the impacts of an AI system on individuals, groups and society.
UK Government — Green Book 2026 (new tab)
Swing weighting, sensitivity analysis, switching values and the warning against simple weighted scoring.

Frequently asked questions

What makes the method transparent?

The decision frame, sources, raw measures, value functions, weights, exclusions, rounding, sensitivity tests and signed decision are versioned. The final number can therefore be recalculated and challenged.

Are the 30–20–20–15–15 weights universal?

No. They exist solely to make the example reproducible. A real decision requires context-specific anchors and weights derived from the relative value of the swings, fixed before candidates are examined.

What happens when assessors disagree?

Divergent values and their rationales remain in the record. They become sensitivity scenarios or a request for further evidence; they are not hidden by an automatic average.

Is a high score enough to fund a use case?

No. An exclusion gate may stop or defer a case regardless of its index. The ranking informs allocation; the competent authority makes and signs the decision.

Put the model under challenge

Which assumption governs your portfolio without being tested?

Bring the decision, candidates and available evidence. We rebuild the frame before calculating a single score.

Frame the arbitration