No. Statistical pattern recognition provides probabilistic estimation, not legal or academic certainty. Elevated scores indicate a heavy density of common machine patterns, but human review is always recommended before making disciplinary decisions.

Our image detection engine uses a Swin Vision Transformer running through ONNX Runtime. Rather than checking metadata (which can easily be stripped), it analyzes microscopic pixel-level denoising patterns, spatial coherence, and border degradation typical of modern diffusion models.

No. Submissions are processed entirely in volatile memory (RAM) strictly for the duration of inference and are discarded immediately after results are generated. We do not store or train models on user data.

For reliable statistical significance in perplexity and burstiness analysis, we recommend submitting at least 50 to 100 words. Short inputs under 20 words lack sufficient token distribution to reliably distinguish human writing from AI phrasing.