Help Defence, sovereign industry and research teams evaluate AI models through authorised scenarios, calibrated expert review and traceable evidence without exposing protected operational knowledge.
A useful evaluation does not ask a defence and national security experts specialist for a vague opinion. It gives them a defined user, task, evidence set and rubric so their judgement can be compared, reviewed and reused. Payment, expected time, permitted materials and the review process should also be clear before a contributor accepts the work.
The required reviewer depends on what could go wrong. Routine clarity checks may need broad domain familiarity, while safety, professional standards or high-impact decisions need demonstrable defence and national security experts experience and an explicit escalation path.
Reviewers identify the source, rule or observable outcome behind important scores. Written rationales make disagreement inspectable and help the project owner distinguish a genuine model failure from an underspecified instruction.
Accepted corrections should not disappear into a spreadsheet. The strongest examples become versioned cases that teams can rerun after prompt, model, tool or policy changes to see whether quality improved without introducing a new failure.
Specialists first review shared defence and national security experts examples and compare material differences. Calibration exposes ambiguous instructions, missing evidence and inconsistent severity rules before those problems affect a larger evaluation or training dataset.
Projects should use approved, minimised evidence and clearly exclude confidential or unrelated material. Contributors need a visible way to pause, redact or escalate when a task would require information they are not authorised to disclose or assess.
SPECIALIST SKILLS
Explore all 8 capabilities. Each has its own qualification profile, example work, evidence contract and safety boundaries.
Evaluate whether AI role players maintain realistic intent, communication, tempo and decision patterns across authorised synthetic training scenarios without claiming operational equivalence from conversational realism alone.
Explore skillTest how AI systems prioritise information, preserve commander intent, communicate uncertainty and recommend escalation in synthetic command-and-control scenarios designed with authorised practitioners.
Explore skillEvaluate whether AI-supported analysis distinguishes observations from inference, weighs competing explanations and communicates confidence without fabricating sources or collapsing uncertainty into an unjustified conclusion.
Explore skillTest whether AI communication is clear, disciplined, appropriately concise and faithful to authorised scenario information while preserving uncertainty, handoffs and escalation in high-tempo training environments.
Explore skillDesign synthetic exercises that vary ambiguity, tempo, sensor reliability and coordination demands so model weaknesses can be observed, reproduced and corrected before any higher-trust trial.
Explore skillEvaluate workload, trust calibration, intervention timing and decision outcomes when people work with AI assistance, including whether operators can understand, challenge and safely override model recommendations.
Explore skillRerun controlled scenario suites after model, prompt, retrieval or tool changes to identify new failures in reasoning, communication, evidence use, latency and safe escalation before a release decision.
Explore skillEvaluate whether AI-supported logistics and sustainment analysis respects dependencies, uncertainty, maintenance constraints and evidence provenance in synthetic planning cases without exposing real readiness information.
Explore skillEXAMPLE PROJECTS
Every campaign defines its own scope, eligibility, quality checks and payment before work begins.
Every campaign defines its own scope, eligibility, quality checks and payment before work begins.
Every campaign defines its own scope, eligibility, quality checks and payment before work begins.