Back The ethical gate

Five checks. Pass all five

Every candidate control, policy, metric or model must clear five invariants at once. Fail one and it is not admitted to the Compliance Core. The invariants are minimums, not ceilings: add your own, but none of the five can be removed or weakened.

Invariant 1

Purpose primacy

Every automation, control, metric or model must demonstrate traceability to the organizational purpose and the Net Positive North Star. Targets are means. Purposes are ends. Confusing them is how organizations hit the KPI while degrading the thing it was supposed to measure.

Passes when the candidate has a documented chain from bounded context, through a strategic objective, to the purpose declaration.

Fails when a candidate exists whose only justification is that the metric improved. Traceability to efficiency alone is not traceability.

Invariant 2

No manipulation. Judgment stays yours

Decision methods that rely on hidden persuasion are forbidden: dark patterns, internal propaganda, undisclosed nudging, or decision support that quietly becomes decision capture. This applies to everything the organization produces, from user-facing interfaces to management reporting to automated decisions.

Passes when a reasonable third party, examining the mechanism, would recognize it as informing judgment rather than replacing it. The assessment is recorded against published reference norms.

Fails when the person being steered cannot see the steering.

Invariant 3

Extraction limits

For every material sustainability topic, the organization identifies the relevant ecological and social limits, sets internal boundaries that keep its impact within them, and holds repair and regeneration plans before a limit is at risk, not after the breach.

Passes when each material topic has a documented limit, a monitoring mechanism and a contingency plan, with thresholds of concern published in advance.

Fails when a topic is treated as unmeasurable and therefore exempt. An imprecise limit that is documented and monitored governs. An unmeasured one does not.

Invariant 4

Accountability for externalities

Ethical evaluation must consider unintended and second-order effects, including those carried by suppliers, contractors and downstream users. Responsibility follows power, not contract: the organization answers to parties with no seat at the governance table, and must act on consequences it discovers, not merely disclose them.

Passes when the risk register includes externalized effects with named owners, and at least one evaluation cycle per year addresses externalities explicitly.

Fails when costs displaced onto third parties or future generations stay off the register because they are off the balance sheet.

Invariant 5

Auditability of the ethical limit

The limit produces evidence. Decisions, exceptions, compensations and overrides are recorded in a form that allows continuous evaluation with verifiable results. Ignorance of consequences is not innocence when the power to investigate exists.

Passes when every ethical decision, including the decision to grant an exception to an invariant, leaves a persistent record an auditor can retrieve without asking the decision-maker.

Fails when decisions live in memory, chat threads or hallway context. Evidence that needs the decision-maker's cooperation is not evidence.

Beyond the gate

Three tests guard the core

The invariants are the first test. Two more decide what the Compliance Core admits.

Ethical acceptability

Does the candidate satisfy all five invariants, simultaneously? This is the gate itself.

Traceability

Can the candidate be traced to at least one strategic objective and one bounded context?

Proportionality

Is the candidate's operational cost proportionate to the risk it addresses and the value it protects?

A candidate that fails any one of the three is not a candidate. Inside the gate, MCDA decides what gets funded, scaled and monitored first. Outside the gate, nothing is considered. Ethical limits cannot be optimized away by aggregating across other considerations.

In practice

Limits you can operate

Two questions recur when the invariants meet reality. Both have working answers.

Who is the reasonable third party?

Not a hypothetical. The standard is inherited, published and recorded:

Legal precedent. The reasonable-person standard runs through tort, contract and professional regulation in common and civil law alike. NPM inherits centuries of case law rather than inventing its own test.

Published reference norms. The governance body publishes the norms against which reasonableness is evaluated: industry codes, regulatory guidance, the organization's own stated values.

An audit trail. Every reasonableness assessment is recorded with the norms applied and the reasoning. An auditor who disagrees gets recorded alongside. Subjective judgment, made transparent and contestable.

Can social limits be measured?

Not with ecological precision. NPM requires engagement, not perfection. The practical test runs five steps:

Map the affected stakeholders and impact dimensions.
Reference the applicable frameworks: ILO conventions, GDPR, the UN Guiding Principles, sectoral codes.
Set thresholds of concern, quantitative where possible, qualitative triggers where not.
Instrument leading signals that monitor proximity to each threshold.
Plan the contingency for a breach, in advance.

The limits will be imprecise, contested and evolving. NPM's commitment is that they exist, are documented, are monitored and are revised. A documented, monitored, imprecise limit governs better than a precise one nobody measured.

"A control that does not pass the logical gate is not a control. It is a liability with documentation."

From the conclusions of the NPM whitepaper