How the engine works
The prose analysis engine, explained clearly and completely.
The prose engine is the system that grades your writing. Here is exactly how it works.
It is fully deterministic
The same text always produces the same scores. There is no randomness, no model inference, and no variation between runs. Submit the same passage twice and you get the same results both times.
There is no AI in grading
The engine uses published readability formulas and rule-based analysis. The formulas are:
- Flesch-Kincaid Reading Ease and Grade Level (Kincaid et al., 1975)
- Gunning Fog Index (Gunning, 1952)
- Coleman-Liau Index (Coleman and Liau, 1975)
- Brysbaert (2019) silent reading research for reading time estimates
These are the same formulas used in academic research, style guides, and readability tools for decades. Inkbreaker did not invent them. It applies them consistently, calibrates them per writing type, and shows you the results.
What happens to your text
Your text is processed server-side, on Inkbreaker's own infrastructure. It is not sent to a language model. It is not stored after processing. It is not used for training. The only thing it is used for is computing the metrics you asked for. There is more detail in Privacy and your writing.
The one place a model touches the product
Content moderation. When community-submitted content crosses into borderline territory that the rule-based filters do not catch cleanly, Inkbreaker uses the OpenAI Moderation API. That is a classification call. It does not read your writing for quality, and it does not generate any response.
What the engine measures
- Readability: Reading Ease, FK Grade, Vocabulary Complexity, Reading Level (Coleman-Liau), Reading Time
- Structure: Average Sentence Length, Sentence Length Variation, Word Count, Sentence Count, Paragraph Count
- Vocabulary: Vocabulary Range, Lexical Density, Repetition Score
- Style: Passive Voice, Adverb Density, Dialogue Ratio, Syntactic Depth, Sentence Opening Variety
Each metric is computed from your text directly. There is no interpretation, inference, or opinion in the computation. What each metric means defines every one. The numbers are the numbers.
Still stuck? Head back to Support to report a bug or reach the team.