Em dashes (—) and en dashes (–) had spread across comments, docs, tests, and a few UI strings, reading as AI-generated boilerplate rather than house style. Replaced each with punctuation matching its context: colon for explanatory clauses, comma for asides, plain hyphen for numeric/legal ranges (e.g. "21-23§"), "to"/"till" for date ranges, parentheses for paired-dash asides. messages/en.json and messages/sv.json were fixed by hand together to keep sv/en in sync. Left untouched where the dash is the functional subject rather than decorative punctuation: date-range-parser.ts's separator regex, charset-repair.ts's CP1252 byte-mapping table (and its test), the SIE encoding mojibake docs, generic-csv.ts's minus-sign normalizer, the agent system-prompt files that already instruct against em dashes, and a golden iXBRL test fixture compared byte-for-byte. Also fixes two bugs surfaced along the way: an off-by-one in ApiKeysPanel's scope-label split (a leftover from an earlier partial pass), and a charset-repair test that had lost the literal en-dash it exists to verify. Regenerated the agent atom seed migration (skills:generate) since 27 SKILL.md files changed. Added a CLAUDE.md rule against em/en dashes, with an explicit carve-out for the functional-dash cases above. Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
136 lines
4.2 KiB
TypeScript
136 lines
4.2 KiB
TypeScript
/**
|
|
* Fuzzy supplier-candidate matching for gnubok_create_supplier_invoice_from_inbox.
|
|
*
|
|
* Exact resolution (matched id → org_number → full-name ilike) misses the
|
|
* common OCR variants: punctuation ("Polarn O. Pyret" vs "Polarn o Pyret"),
|
|
* legal-form suffixes ("… AB"), and formatted org numbers ("556677-8899" vs
|
|
* "5566778899"). When resolution fails, the tool surfaces near-miss candidates
|
|
* from these matchers so the agent can retry with supplier_id_override instead
|
|
* of dead-ending: the agent (or the approving human) makes the final call,
|
|
* fuzzy scores never auto-resolve.
|
|
*/
|
|
|
|
export type SupplierRow = {
|
|
id: string
|
|
name: string
|
|
org_number: string | null
|
|
}
|
|
|
|
export type SupplierCandidate = {
|
|
supplier_id: string
|
|
name: string
|
|
org_number: string | null
|
|
score: number
|
|
matched_on: 'org_number' | 'name'
|
|
}
|
|
|
|
// Legal-form suffixes carry no identity signal and OCR/extraction includes
|
|
// them inconsistently. Longest-first so 'aktiebolag' is stripped before 'ab'.
|
|
const LEGAL_SUFFIXES = [
|
|
'ekonomisk förening',
|
|
'kommanditbolag',
|
|
'handelsbolag',
|
|
'aktiebolag',
|
|
'ek för',
|
|
'filial',
|
|
'ekf',
|
|
'ab',
|
|
'hb',
|
|
'kb',
|
|
]
|
|
|
|
export function normalizeSupplierName(raw: string): string {
|
|
let s = raw
|
|
.toLowerCase()
|
|
// Punctuation → space; keep letters (incl. åäöé), digits, ampersand.
|
|
.replace(/[^a-z0-9åäöéü&]+/gi, ' ')
|
|
.replace(/\s+/g, ' ')
|
|
.trim()
|
|
for (const suffix of LEGAL_SUFFIXES) {
|
|
if (s.endsWith(` ${suffix}`)) {
|
|
s = s.slice(0, -suffix.length - 1).trim()
|
|
break
|
|
}
|
|
}
|
|
return s
|
|
}
|
|
|
|
function digitsOnly(s: string): string {
|
|
return s.replace(/\D/g, '')
|
|
}
|
|
|
|
/**
|
|
* Canonical 10-digit key for a Swedish org number. Orgnr is exactly 10
|
|
* significant digits; enskild firma uses the owner's personnummer, which
|
|
* appears in both 10-digit (YYMMDDXXXX) and 12-digit (YYYYMMDDXXXX) forms:
|
|
* the last 10 digits are the same identifier. Anything else is not a Swedish
|
|
* org number and must not fuzzy-match.
|
|
*/
|
|
export function orgNumberKey(raw: string): string | null {
|
|
const d = digitsOnly(raw)
|
|
if (d.length === 10) return d
|
|
if (d.length === 12) return d.slice(-10)
|
|
return null
|
|
}
|
|
|
|
/**
|
|
* Similarity in [0, 1]. Exact normalized match = 1; containment
|
|
* ("polarn o pyret" ⊂ "polarn o pyret sverige") = 0.9; otherwise token
|
|
* Jaccard scaled to max 0.8 so partial overlaps never outrank containment.
|
|
*/
|
|
export function scoreSupplierName(a: string, b: string): number {
|
|
const na = normalizeSupplierName(a)
|
|
const nb = normalizeSupplierName(b)
|
|
if (!na || !nb) return 0
|
|
if (na === nb) return 1
|
|
if (na.includes(nb) || nb.includes(na)) return 0.9
|
|
const ta = new Set(na.split(' '))
|
|
const tb = new Set(nb.split(' '))
|
|
let intersection = 0
|
|
for (const t of ta) if (tb.has(t)) intersection++
|
|
const union = new Set([...ta, ...tb]).size
|
|
return union === 0 ? 0 : Math.round((intersection / union) * 0.8 * 100) / 100
|
|
}
|
|
|
|
export function findSupplierCandidates(
|
|
suppliers: SupplierRow[],
|
|
extractedName: string | null,
|
|
extractedOrgNumber: string | null,
|
|
options: { limit?: number; minScore?: number } = {},
|
|
): SupplierCandidate[] {
|
|
const limit = options.limit ?? 5
|
|
const minScore = options.minScore ?? 0.4
|
|
const extractedOrgKey = extractedOrgNumber ? orgNumberKey(extractedOrgNumber) : null
|
|
|
|
const scored: SupplierCandidate[] = []
|
|
for (const s of suppliers) {
|
|
// Canonical-key equality catches formatting variants ('556677-8899' vs
|
|
// '5566778899') and the 10- vs 12-digit personnummer forms of enskild
|
|
// firma org numbers: none of which the exact .eq() lookup upstream can.
|
|
if (extractedOrgKey && s.org_number && orgNumberKey(s.org_number) === extractedOrgKey) {
|
|
scored.push({
|
|
supplier_id: s.id,
|
|
name: s.name,
|
|
org_number: s.org_number,
|
|
score: 1,
|
|
matched_on: 'org_number',
|
|
})
|
|
continue
|
|
}
|
|
if (extractedName) {
|
|
const score = scoreSupplierName(extractedName, s.name)
|
|
if (score >= minScore) {
|
|
scored.push({
|
|
supplier_id: s.id,
|
|
name: s.name,
|
|
org_number: s.org_number,
|
|
score,
|
|
matched_on: 'name',
|
|
})
|
|
}
|
|
}
|
|
}
|
|
|
|
return scored.sort((a, b) => b.score - a.score).slice(0, limit)
|
|
}
|