The Explanation Checker: Why Getting the Right Answer Isn't Enough Anymore

Technical assessments have a credibility problem. Candidates produce correct answers through methods that have nothing to do with actual competence: AI chatbots, real-time messaging with helpers, solution repositories, screen sharing. The automated scoring system awards full points, and hiring teams advance candidates who can't perform the work.

"The problem is not detection of cheating during the test. The problem is verification of understanding after the test."

Post-Test Verification: A Different Approach

Rather than trying to prevent external help during problem-solving, the Explanation Checker verifies whether the candidate understands the solution they submitted. After completion, the platform selects 2–3 correctly answered questions and asks the candidate to record a video walkthrough of their reasoning.

Candidates can't prepare. They don't know which questions will be selected. They can't consult resources during recording as it's a single continuous take. And they can't bluff, because the AI evaluation checks for logical flow, decision rationale, technical precision, and edge-case awareness.

Why It Actually Works

The system exploits a fundamental asymmetry: obtaining a correct answer takes seconds of access to external resources. Understanding why that answer is correct requires internalised knowledge that can't be quickly transferred. A language model can generate working code instantly, but it can't implant the understanding required to explain it coherently.

Candidates who cheated exhibit consistent patterns: they describe what code does without explaining why they wrote it that way. They stumble on alternatives. They rely on vague generalities instead of specific technical reasoning. The system flags these for human review. It doesn't attempt to definitively prove cheating.

The Deterrent Effect

Perhaps the most interesting outcome: the existence of explanation verification changes candidate behaviour before submission. Candidates who know they'll need to explain their solutions become more cautious about external assistance. They can't memorise explanations in advance, can't prepare generic templates, and can't get real-time help during recording. The mechanism functions as both filter and deterrent.

The result: assessment scores that reliably predict technical capability, restoring the assessment's function as a genuine filter rather than an easily gamed checkpoint.