The complete ATC
candidate assessment pipeline.

Knowledge assessment alongside practical performance. Each candidate session includes a structured learning module with knowledge-check questions covering terminal procedures, separation standards, emergency handling, and coordination protocols.

Study-level terminal area radar control, built in Unity and served to the browser via WebGL. No downloads, no plugins — candidates access their simulation session from any modern browser.

The environment models realistic aircraft flight dynamics and procedural clearances. Traffic complexity scales with each successive run, allowing the platform to stress-test candidate performance under increasing workload.

Candidates interact using voice commands through any microphone.

Real speech. Real scoring. The voice layer uses speech-to-text transcription followed by an LLM-based inference model to parse candidate transmissions, correct transcription noise, and evaluate:

• Adherence to standard phraseology
• Rate of speech (words per minute)
• Readback error detection — the simulator plants deliberate readback mistakes to test vigilance
• Transmission redundancy — issuing commands to aircraft already executing that instruction

The model handles non-standard phraseology without penalising candidates for accents or minor linguistic variation — only for procedurally incorrect transmissions.

Every session is scored automatically by the Separation Integrity Engine (SIE) in real time. The SIE detects >99% of separation violations using predicted intercept vectors — the same principle underlying real STCA systems — and classifies each event by severity:

• Separation Loss — actual radar or vertical separation infringement
• Critical — clearance that would directly cause a separation loss
• Significant — procedural error with safety implications
• Minor — non-standard practice or communication error

Run scores are weighted by session difficulty — later, more complex runs contribute more to the final simulator score.

Candidate performance is decomposed into distinct skill categories across four domains:

• Planning
• Managing Multitasking
• Communication
• Perceptual Speed

Each subcategory is graded A through F, aggregated from all simulation runs. Grades are weighted by run importance.

Performance-targeted questions. After each assessment, the platform reviews a candidate's results and selects the most relevant questions from a curated bank — each written to probe a specific skill gap the simulator identified. The right question surfaces automatically.

Each question is paired with a model answer, so panel members can evaluate responses effectively without specialist ATC knowledge.