Mission competence: systems that know what they can do

Ask an operator of critical systems what they fear most in an artificial-intelligence platform, and the answer will rarely be "that it fails." It will be "that it claims it can do something it cannot." A phantom capability is worse than an absent one: the absent capability is known and worked around; the phantom is believed — until the moment the world collects the difference.

Zendrix treats this as an architecture property. We call mission competence the discipline by which a system only declares — to itself and to its operators — the capabilities that were validated in the context in which it finds itself. This article describes the principles of that modeling, in continuity with our publications on autonomy with a formal envelope and calibrated uncertainty.

Capability is not binary — it is contextual

The most common error in capability management is treating capabilities as permanent attributes of equipment. "The system has an optical sensor" appears in the inventory as a fact; but an optical sensor under clear sky and the same sensor under closed overcast are, operationally, two different systems. Capability exists at the intersection of equipment, context, and validation — and any model that ignores one of the three factors produces false self-knowledge.

Zendrix competence modeling represents that intersection explicitly. Each capability is linked to the contexts in which it was validated — environmental conditions, communication regime, task class — and to each (capability, context) pair corresponds a formal autonomy level, the same declared and auditable levels the platform already uses for action governance. A capability validated in a broad context operates with broad autonomy; the same capability in a never-validated context operates at level zero: the system knows that, there, it does not know.

Choosing by competence, not by availability

The most important operational consequence is the change in the selection criterion. Conventional systems assign tasks by what is available: which resource exists, which is free, which responds. Systems with mission competence assign by what is validated: which resource has verified competence for that task in that context.

The difference appears entirely in the hard cases. Under overcast sky, an imaging resource that depends on direct line of sight may be available and incompetent; an active sensing resource, immune to cloud cover, may be less convenient and competent. Selection by availability schedules the first and discovers the error in operation. Selection by competence schedules the second — and records the reason for the choice in the audit trail, so the decision can be reconstructed and contested later.

This applies equally to internal self-assignment: when the system plans, it plans with the competences it holds in the context it is in, not with the ones it would hold under ideal conditions. A plan that requires capability not validated in the current context is downgraded, reassigned, or refused — never executed on hope.

Virgin context: the honesty of level zero

There will always be contexts in which no validation exists — a new environment, an unprecedented condition, a combination never exercised. The mature response is not to extrapolate with decorative confidence; it is to declare level zero and operate accordingly.

Level zero is not paralysis. It is the system saying, explicitly and auditably: "in this context, this capability has not been validated; I recommend, but I do not execute autonomously." Human authority receives the recommendation with declared uncertainty — and decides. If the decision is to proceed, the outcome feeds the validation base, and future competence is earned with evidence, not presumed by analogy. This is how a system expands its own envelope responsibly: every new context crossed under human supervision becomes a validated context for the next time.

This architecture transforms the relationship between operator and system. The operator does not have to guess when to trust: the system declares its own competence level per context, with the same objectivity with which it declares battery level. Trust stops being a feeling and becomes an interface — with numbers, levels, and audit trails behind it.

Competence measured: the bridge to the culture of evidence

Mission competence is only real if it is measured. Zendrix maintains, as the infrastructure of this modeling, validation records per (capability, context) pair: scenarios exercised in high-fidelity simulation, acceptance floors reached, date and version of the validation. A competence without a validation record is treated as nonexistent — not out of pessimism, but because the alternative is exactly the phantom that opens this article.

These records integrate with the assurance mechanisms we have already described publicly: formal, verifiable autonomy levels; tamper-resistant audit trails; and cryptographic verification of the declared effect of actions. The competence model is not a presentation layer over the system — it is part of the governance system, consulted at every assignment decision and recorded at every choice.

Dual use and sovereignty

The problem of contextual competence crosses defense and critical infrastructure with the same force. A power-grid operator needs to know whether the anomaly-detection system was validated for today's load regime; a civil-defense authority needs to know whether the image-analysis capability was validated for night and rain conditions; an aviation authority needs to know what an unmanned system can and cannot do under link degradation. In every case the question is the same — and the honest answer requires explicit modeling, not optimism.

There is also a sovereign dimension. A system that knows and declares its own competences is a system auditable by its own nation: the validations, the floors, and the levels remain under national jurisdiction and control, instead of depending on the word of a foreign supplier about what its black box "can do." Internal transparency and sovereignty walk together.

The state of development

Mission competence modeling is implemented and operating in Zendrix validation environments, integrated with autonomy governance and the simulation records described in our previous publications. The assignment criteria — what evidence promotes a competence level, what degradation demotes it — are explicit, versioned, and auditable, and continue to harden with every validation cycle.

The institutional position is the usual one, applied to a new object: we do not promise omniscient systems. We build systems that know what they can do, declare what they cannot, and never cross that boundary in silence. Such a system does not impress in demonstration — it impresses in operation, which is where defense and critical infrastructure need to impress.