> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vaero.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Metrics

> Quality metrics to evaluate the results

## Basics

Responses from the style transform endpoint include quality metrics when the `include_quality` parameter is set to `true`. The quality metrics show differences between the `unpersonalized` text, which is the AI-generated text without Vaero styling, the `personalized` text, which is the text after Vaero styling, and `ground_truth` text, which is the dataset used for fine-tuning the model. Many metrics also include a `personalization_improvement` value that indicates the relative improvement from styling.

* **ROUGE scores** — measures n-gram overlap between the output text and the training data
  * ROUGE-1 — unigram overlap (precision, recall, F-measure)
  * ROUGE-2 — bigram overlap (precision, recall, F-measure)
  * ROUGE-L — longest common subsequence (precision, recall, F-measure)

* **Sentence length statistics** — sentence length measured in words
  * Mean — the mean sentence length in the text
  * Median — the median sentence length in the text
  * Std dev — the standard deviation in sentence length in the text

* **Word length statistics** — word length measured in characters
  * Mean — the mean word length in the text
  * Median — the median word length in the text
  * Std dev — the standard deviation in word length in the text

* **Readability**
  * Flesch Reading Ease — higher values indicate easier-to-read text
  * Flesch-Kincaid Grade Level — the U.S. school grade level needed to understand the text

* **Function word frequency** — measures similarity of function word distributions to the training data using Jensen-Shannon distance

* **Part-of-speech frequency** — measures similarity of part-of-speech distributions to the training data using Jensen-Shannon distance

* **Lexical density** — the ratio of content words to total words, indicating how informationally dense the text is

* **Sentence structure frequency** — distribution of sentence structure types (simple, compound, complex, compound-complex, fragment/unknown) with Jensen-Shannon distance

* **Sentence voice frequency** — distribution of active vs. passive voice with Jensen-Shannon distance

* **Punctuation** — density of punctuation types (comma, semicolon, colon, period, question mark, exclamation mark, parenthetical) with Jensen-Shannon distance