This text inspection tool evaluates written passages to identify stylistic and statistical markers common in text generated by Large Language Models (LLMs). Rather than relying on closed remote neural classifiers, the tool uses verifiable client-side linguistic heuristics to measure cadence, lexical repetition, and syntactic uniformity.

### Statistical Detection Heuristics The analyzer computes four distinct text metrics in JavaScript:

- **Sentence Burstiness (Length Variance):** Evaluates the standard deviation of sentence lengths divided by average sentence length: $ ext{Burstiness} = \min\left(100, \text{round}\left(\frac{\sigma_{\text{len}}}{\mu_{\text{len}}} \times 100\right)\right)$. Human writers alternate naturally between short punchy sentences and longer compound explanations. AI models produce more uniform sentence lengths with low standard deviation. - **Vocabulary Diversity (Type-Token Ratio):** Measures lexical variation by dividing unique word counts by total word count ($TTR = (\text{Unique Words} / \text{Total Words}) \times 100$). Lower ratios indicate repetitive phrasing. - **AI Transition Marker Density:** Scans text against a list of transition words over-represented in machine generation (including *furthermore*, *moreover*, *consequently*, *pivotal*, *vital*, *multifaceted*, and *foster*). - **Sentence Uniformity Ratio:** Calculates the percentage of sentences falling tightly within the 13-to-22 word range, a common structural plateau in unprompted LLM output.

### Concrete Analysis Walkthrough Analyzing two sample paragraphs of 65 words each: 1. **Sample A (Uniform Draft):** Three consecutive sentences measuring 21, 20, and 22 words with transitions *furthermore* and *consequently*. The burstiness score calculates at **14/100**, flagging high uniformity. 2. **Sample B (Human Draft):** Three sentences measuring 7, 34, and 12 words with zero formal transition markers. The burstiness score calculates at **68/100**, indicating varied rhythmic cadence.

### Privacy and Client-Side Processing Note All text parsing, tokenization, regular expression checks, and variance calculations execute locally in your web browser. No essays, drafts, or sensitive documents are transmitted across any network or retained in external databases.