Apply decision gates before the score

A prohibited, intolerable or unknown condition does not become acceptable because an average is high. It must be resolved before any ranking.

Dr Youssef Trardi·Updated 7 September 2026

Principle

Separate admissibility from preference

A gate asks whether an option may enter the arbitration. A criterion asks which admissible option is preferred. Mixing the two allows a secondary benefit to offset a limit that was never meant to be compensated.

Specification

Write every gate as an executable control

A general phrase such as “acceptable risk” cannot be tested. The gate names the object, condition, evidence, person qualifying it and authority deciding the effect.

Minimum fields for a decision gate

Field	Question	Requirement
Condition	Which fact opens or closes the gate?	Observable state and scope
Evidence	Which item supports the conclusion?	Source, date, version and limitation
Authority	Who may accept or refuse?	Name, role and delegation
Effect	What happens if it fails?	Stop, defer or reframe
Reopening	What permits reassessment?	Expected evidence and owner

Condition

Question: Which fact opens or closes the gate?
Requirement: Observable state and scope

Evidence

Question: Which item supports the conclusion?
Requirement: Source, date, version and limitation

Authority

Question: Who may accept or refuse?
Requirement: Name, role and delegation

Effect

Question: What happens if it fails?
Requirement: Stop, defer or reframe

Reopening

Question: What permits reassessment?
Requirement: Expected evidence and owner

States

Three states, never pass by default

Missing evidence is not a pass. It preserves uncertainty and prevents the option from receiving a ranking presented as decision-ready.

Passed: The condition is met for the examined version and scope
Unknown: Evidence is inconclusive; the option remains outside the ranking
Failed: The condition is breached; the specified effect is applied
Expired: Old evidence or a change requires fresh qualification

Families

Adapt gates to the legal and operating context

The following families structure the examination without creating universal thresholds. Their wording depends on the use, sector, affected people and approved tolerance.

Family	Example condition	Possible effect
Admissibility	Prohibited use or undefined purpose	Stop or reframe
Impacts	Residual risk outside tolerance	Stop or reduce scope
Data	Right to use critical data not established	Defer
Operation	No practicable fallback capability	Defer or reframe
Resources	Mandatory cost or time constraint not met	Remove from the current frame

Admissibility

Example condition: Prohibited use or undefined purpose
Possible effect: Stop or reframe

Impacts

Example condition: Residual risk outside tolerance
Possible effect: Stop or reduce scope

Data

Example condition: Right to use critical data not established
Possible effect: Defer

Operation

Example condition: No practicable fallback capability
Possible effect: Defer or reframe

Resources

Example condition: Mandatory cost or time constraint not met
Possible effect: Remove from the current frame

Execution

Put every option through the same protocol

Order and evidence level remain the same for every candidate. The log retains initial states, examined evidence and every exception.

Freeze — version the gates before reading scores
Examine — connect every status to evidence and an owner
Challenge — review qualifications that remove an option
Decide — apply the effect through the specified authority
Lock — exclude unknown or failed states from ranking
Retain — date the verdict and reassessment trigger

Reopening

Reassess on new evidence, never on new pressure

An option returns when the expected evidence exists, the condition is reassessed and the authority confirms the new status. Renaming the option, artificially narrowing the risk or increasing another score does not reopen anything.

Output

Produce a register that explains every exclusion

The register connects option, gate, status, evidence, authority, effect, rationale and reopening condition. A committee can distinguish a preference choice from a legal, ethical, technical or operating impossibility.

Sources and references

UK Government — practical multi-criteria analysis guidance (new tab)
Value functions, swing weighting, decision conferencing, sensitivity and ranking robustness.
UK Government — Green Book 2026 (new tab)
Swing weighting, sensitivity analysis, switching values and the warning against simple weighted scoring.
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.

Frequently asked questions

How does a gate differ from a criterion?

A gate establishes whether an option may enter the ranking. A criterion compares options that are already admissible. A failed gate cannot be offset by a better value elsewhere.

What does unknown evidence mean?

The condition is neither satisfied nor disproved. The option remains outside the ranking until the expected evidence is produced and qualified.

Can a gate be reopened?

Yes, on new evidence defined in advance, followed by a fresh qualification and a decision by the competent authority — never through sponsor pressure alone.

Decision gates

What are you currently trying to offset with a high score?

Present the mandatory constraints and available evidence. We separate what can be compared from what must be resolved first.

Review the gates