๐Ÿ 1,000-Document Empirical Validation

Empirical Accuracy & Benchmark Report

Rigorous evaluation of AIDetector.buzz's mathematical detection engine against a dataset of 1,000 full documents (500 historical human writings created before 2010 and 500 multi-model AI texts).

Overall Accuracy
100.0%
1,000 / 1,000 tests passed
False Positive Rate
0.0%
0 / 500 Pre-2010 Human texts misclassified
True Positive Rate
100.0%
500 / 500 AI documents correctly flagged
Total Failed Tests
0
0 Human errors ยท 0 AI errors

Mathematical Separation & Diagnostic Averages

MetricHuman Corpus (Pre-2010)AI Corpus (LLM Generated)Diagnostic Implication
Sample Size500 documents500 documentsBalanced 1:1 test matrix
Average Perplexity (PPL)96.315.2High surprisal in human text vs machine predictability
Average Burstiness (ฯƒ_PPL)15.67.9Dynamic rhythmic variety vs uniform machine pacing
Classification Rate100.0% Human100.0% AI / FlaggedZero false accusations on historical prose

๐Ÿ“œ Human Corpus (Pre-2010 Verified)

To ensure 100% genuine human authenticity without contamination from generative models, the 500 human writings were sourced strictly from historical literature, classic philosophy, scientific treatises, and memoirs published prior to 2010:

  • Classical Literature: George Orwell (1984), Jane Austen, Charles Dickens, Herman Melville, Mark Twain, F. Scott Fitzgerald, Ernest Hemingway.
  • Scientific Papers: Albert Einstein (Annalen der Physik, 1905), Charles Darwin (Origin of Species, 1859).
  • Philosophy & Speeches: Marcus Aurelius, Abraham Lincoln, Winston Churchill, Plato, Aristotle, Steve Jobs (2005).

๐Ÿค– AI Corpus (Multi-Model & Multi-Domain)

The 500 AI passages span modern generative language models (GPT-4o, Claude 3.5, Gemini 1.5, DeepSeek, and Llama 3) across 25 distinct technical and professional domains:

  • Technical Architectures: Distributed consensus, quantum cryptography, neuromorphic hardware, vector search indexing.
  • Enterprise & Economics: Dynamic pricing elasticity, supply chain optimization, CI/CD automation, zero-trust security.
  • Academic Essays: Machine learning in education, biomedical genomics, geospatial satellite analytics.

Test Your Own Documents Now

Paste any essay, article, or research paper into the live detector to inspect sentence-level yellow highlights and stylometric proofs with zero data retention.

Launch AI Scanner โ†’