P→MPDFMDSystem documentation
Format referenceOpen converter ↗

Level 4 · Recognition method

Numbered paragraphs and embedded lists

A printed legal paragraph number identifies the paragraph. A list may then appear inside that same paragraph. The recognizer must preserve both structures without replacing the source label.

Printed labels are source text

Figure 4 · Illustrative containing paragraph
Paragraph #7printed marker “7.” is retained
↓
Opening prose“The supplier must provide:”
↓
Embedded list“(a) records; (b) invoices;”
↓
Following prose“within thirty days.”

The paragraph remains one addressable unit. Internal items have their own identities and source locations.

Automatic Markdown numbering must not replace a visible legal marker, including restarts, skipped values, compound markers, or unusual spelling.

Markers, indentation, and continuation

Propose markers

Bullets, decimal and compound numbers, alphabetic or Roman labels, parenthesized forms, and task symbols.

Measure local geometry

Indentation, hanging indent, text start, active column, line spacing, and paragraph gaps.

Infer levels

Marker sequence, left edge, parent compatibility, and neighboring items support a hierarchy.

Attach continuation prose

An unmarked paragraph remains inside the item only when its text start, spacing, style, and discourse agree. A body-margin reset, section break, larger gap, or higher-level marker ends the list.

Repeated item grammar inside one paragraph

The recognizer requires at least two compatible marker–item sequences and clause boundaries such as punctuation, conjunctions, or repeated syntax. It rejects likely years, measurements, citations, equation parentheses, and isolated cross-references.

Status. Page-local list detection exists, but complete preservation of containing-paragraph identity, original numeric labels, multiline embedded items, and cross-page cases is staged work.

Indentation alone is insufficient

Quote marks, a vertical quote bar, relative indentation, narrower measure, attribution or citation cues, and consistent multi-line styling contribute evidence. Sidebars, addresses, list continuations, code, and centered pull quotes are confounders. A resolved quotation renders with Markdown >; a meaningful source quote-bar can remain separate evidence.