Project NOMAD is an offline-first knowledge and education server. Wikipedia, thousands of books, courses, maps, and optional local AI, all running on hardware you own with no internet required.
TypeScript
/**
* Parse a flattened FDA label section (drug interactions, dosage, warnings,
* contraindications, indications, ...) into ordered blocks for readable
* rendering. The split is purely structural: every word of the original text is
* preserved in order — only the numbered-subsection and bullet markers move into
* block metadata. No FDA wording is altered, summarized, or dropped. The
* content-fidelity invariant is enforced by the unit tests.
*/
/** One renderable block of parsed FDA label text. */
export interface LabelBlock {
/** Subsection label like "7.1", or null for intro/header text. */
label: string | null
/** Verbatim text (subsection label stripped from the front), or null for a bullet block. */
text: string | null
/** Verbatim bullet items, or null for a text block. */
bullets: string[] | null
}
/** Backward-compatible alias — the interaction comparison view's original name. */
export type InteractionBlock = LabelBlock
// A section's own leading header: a number then an ALL-CAPS section name, e.g.
// "7 DRUG INTERACTIONS", "2 DOSAGE AND ADMINISTRATION". Anchoring subsection
// detection to THIS section's number means inline cross-refs like "(12.3)" or
// "( 7.1 )" are never mistaken for a subsection heading.
const HEADER_RE = /^\s*(\d{1,2})\s+[A-Z][A-Z][A-Z &,/()'.’-]*/
// "<major>.<n> <Capital><lowercase>" subsection number, for the no-header fallback.
const SUBSECTION_RE = /\b(\d{1,2})\.\d{1,2}\s+[A-Z][a-z]/g
/** This section's major number — from its leading header, else the dominant subsection major. */
function detectMajor(text: string): string | null {
const header = text.match(HEADER_RE)
if (header) return header[1]
const counts = new Map<string, number>()
for (const m of text.matchAll(SUBSECTION_RE)) {
counts.set(m[1], (counts.get(m[1]) ?? 0) + 1)
}
let best: string | null = null
let bestN = 0
for (const [major, n] of counts) {
if (n > bestN) {
best = major
bestN = n
}
}
return best
}
export function parseLabelSection(raw: string | null | undefined): LabelBlock[] {
if (!raw) return []
const text = raw.trim()
if (!text) return []
const major = detectMajor(text)
let pieces: string[]
let labelRe: RegExp | null = null
if (major) {
// Split before "<major>.<n> <Capital><lowercase>". The \b prevents matching
// inside a larger number (e.g. "17.1"); the [A-Z][a-z] requirement excludes
// parenthetical refs like "( 7.1 )" (those are followed by ")"). Lookahead
// only — no lookbehind — for broad browser support.
const splitRe = new RegExp(`(?=\\b${major}\\.\\d{1,2}\\s+[A-Z][a-z])`)
pieces = text.split(splitRe)
labelRe = new RegExp(`^(${major}\\.\\d{1,2})\\s+`)
} else {
pieces = [text]
}
const blocks: LabelBlock[] = []
for (const rawPiece of pieces) {
const piece = rawPiece.trim()
if (!piece) continue
let label: string | null = null
let body = piece
if (labelRe) {
const m = piece.match(labelRe)
if (m) {
label = m[1]
body = piece.slice(m[0].length).trim()
}
}
if (body.includes('•')) {
const parts = body.split('•')
const lead = parts[0].trim()
const items = parts
.slice(1)
.map((s) => s.trim())
.filter(Boolean)
if (lead) {
blocks.push({ label, text: lead, bullets: null })
if (items.length) blocks.push({ label: null, text: null, bullets: items })
} else if (items.length) {
blocks.push({ label, text: null, bullets: items })
}
} else if (body) {
blocks.push({ label, text: body, bullets: null })
}
}
return blocks
}
/** Backward-compatible alias for the interaction comparison view. */
export const parseInteractions = parseLabelSection
/** True when a text block is just a "N SECTION NAME" FDA section header. */
export function isLabelSectionHeader(text: string | null): boolean {
return !!text && /^\s*\d{1,2}\s+[A-Z][A-Z][A-Z &,/()'.’-]*$/.test(text)
}
/** Backward-compatible alias. */
export const isSectionHeader = isLabelSectionHeader