Oct 23, 2025
FormatBench: Structural Fidelity
By TypeOS Research
Read PaperThe first benchmark for formatting in AI writing: APA, MLA, Chicago, legal contracts, business documents, tables, headings, citations, indentation, spacing, and structural fidelity.
No other major model lab currently measures formatting accuracy. TypeOS recognizes that formatting is not just decoration—it is meaning.
Formatting as Meaning
In professional contexts, an indentation error in Python breaks code. In legal contracts, a misplaced clause structure can invalidate an agreement. In academic writing, citation errors are plagiarism.
FormatBench evaluates the "Structural Fidelity" of generated documents, ensuring that models respect the rigid rulesets of professional domains.
Fidelity Metrics
SCS
Structural Correctness Score
Adherence to DOM-level structure rules.
CSA
Citation Style Accuracy
Correctness of inline citations and bibliographies.
FFI
Formatting Fidelity Index
Visual match to gold-standard templates.
CCI
Contract Clause Integrity
Preservation of legal numbering and hierarchy.
