Fractal-ish Complexity for Regulations: A Practitioner-Ready, Agentic Benchmark

Agents4Science 2025 Conference Submission45 Authors

Keywords: regulatory complexity; plain language; cross-references; hierarchy; readability; legal informatics; AI agents; reproducibility.

TL;DR: We present the Regulatory Fractal-ish Index (RFI), a transparent, scope-aware signal of textual complexity for regulations and SOP-style documents.

Abstract: We present the Regulatory Fractal-ish Index (RFI), a transparent, scope-aware signal of textual complexity for regulations and SOP-style documents. RFI blends (i) size (section count and heading density), (ii) hierarchical spread (entropy of heading levels), and (iii) lookup pressure (cross-reference density), adapting automatically to full documents and short excerpts. A lightweight agentic pipeline parses text, computes RFI, and emits a one-page policy brief with actionable edits (e.g., reduce lookup hops, flatten nesting). We also report a minimal hierarchical scaling check ( ˆ Dhier with R2) across sentences → paragraphs → sections, to reconnect with fractal intuitions without overclaiming. The goal is a tool regulators can actually use, backed by transparent, reproducible computations.

Submission Number: 45