feat: parse DEGIRO exports in any interface language

.gitignore

@@ -9,3 +9,4 @@ coverage/
 *.log
 *.csv
 !test/fixtures/Account.csv
+!test/fixtures/Account-en.csv

AGENTS.md

@@ -33,9 +33,9 @@ Run `pnpm lint && pnpm typecheck && pnpm test` before proposing a change; that i
 ```
 src/
   csv/         tokenizer (papaparse) + Node row stream + shared options
-  dialects/    Dialect interface, French dialect, DialectRegistry
+  dialects/    Dialect interface, French/English/layout-fallback dialects, DialectRegistry
   records/     RawRecord — a normalised, dialect-agnostic row
-  classify/    Matcher registry + matchers (trade, fx, dividend, fees, interest, cash)
+  classify/    Matcher registry + matchers (trade, fx, dividend, fees, interest, cash, structural)
   group/       GroupingStrategy pipeline + Transaction union
   validate/    per-currency balance reconciliation
   portfolio/   positions, FIFO realized P/L, cash and fee roll-ups
@@ -54,6 +54,8 @@ examples/dashboard/
 
 The pipeline is `tokenizeCsv` → `mapRow` (dialect) → `ClassifierRegistry.classify` → `groupMovements` (strategy list) → optional `reconcileBalances` / `summarizePortfolio`. Every stage is swappable through `ParseOptions`; adding a locale, a description or a grouping rule should mean registering something, never editing a `switch`.
 
+Header language, number format and description language are three independent axes of a DEGIRO export — changing the interface language rewrites line 1 and nothing else. Keep them independent in code: dialects own the header and the number/date formats, matchers own the wording, and neither may assume the other. `genericDialect` closes the set by matching the column layout alone; anything heuristic sets `heuristic: true` so `ParseResult.warnings` can say so.
+
 ## Rules that are load-bearing
 
 - **The root entry imports no Node builtin.** `src/index.ts` and everything it reaches must stay free of `node:*`; anything needing a filesystem or streams goes behind `src/node.ts`. `test/entrypoints.test.ts` guards this and has caught a regression before.
@@ -69,7 +71,7 @@ The pipeline is `tokenizeCsv` → `mapRow` (dialect) → `ClassifierRegistry.cla
 - TypeScript is strict with `noUncheckedIndexedAccess` and `verbatimModuleSyntax`. Type-only imports must use `import type` — `@typescript-eslint/consistent-type-imports` is an error, not a warning.
 - Prettier: single quotes, semicolons, trailing commas, 100 columns. `.csv` fixtures and `CHANGELOG.md` are excluded and must not be reformatted.
 - `examples/**/components/ui/**` is generated by the shadcn CLI and kept verbatim so it stays upgradable; two lint rules are disabled there. Do not hand-edit those files.
-- `test/fixtures/Account.csv` is synthetic but byte-sensitive: French decimals, `U+202F` thousands separators, double-spaced product names, and running balances that reconcile exactly. Editing it usually means fixing the balances too. A real `Account.csv` at the repo root is git-ignored.
+- `test/fixtures/Account.csv` is synthetic but byte-sensitive: French decimals, `U+202F` thousands separators, double-spaced product names, and running balances that reconcile exactly. Editing it usually means fixing the balances too. `test/fixtures/Account-en.csv` is the same body behind the English header; keep the two bodies identical, since a test asserts they classify the same. Both are unignored by name in `.gitignore`. A real `Account.csv` at the repo root is git-ignored.
 - New behaviour needs a test in `test/` (library) or `examples/dashboard/test/` (analytics). The analytics layer imports nothing from React precisely so it stays testable headlessly.
 
 ## Git and PRs

README.md

@@ -29,6 +29,7 @@ A modern, ESM-only TypeScript library that turns a DEGIRO account statement into
 
 - 🧮 **Exact money** with [`big.js`](https://github.com/MikeMcl/big.js) — no float drift
 - 🧩 **Extensible** dialects, classifiers and grouping strategies
+- 🌐 **French and English exports** out of the box, and a layout-detecting fallback for every other language
 - 🧠 **Typed domain model** — discriminated unions for movements & transactions
 - 🪶 **Lenient parsing** — per-row problems are collected, never thrown
 - 🌊 **Streaming** parser for very large files
@@ -108,6 +109,40 @@ const result = await parseDegiroStream(createReadStream('./Account.csv'));
 
 Parsing is **lenient**: a row with an unparseable date is dropped and reported as an `error`; a partially-parseable amount becomes a `warning`. The only thrown conditions are an empty input and a header that matches no dialect (`UnknownDialectError`).
 
+## Supported exports
+
+DEGIRO localizes the **header row** of `Account.csv` to the interface language. Two dialects recognise a header by its words, and a third recognises the file by its shape when neither does:
+
+| Dialect          | `id`      | Header                                                                            |
+| ---------------- | --------- | --------------------------------------------------------------------------------- |
+| `frenchDialect`  | `fr`      | `Date,Heure,Date de,Produit,Code ISIN,Description,FX,Mouvements,,Solde,,ID Ordre` |
+| `englishDialect` | `en`      | `Date,Time,Value date,Product,ISIN,Description,FX,Change,,Balance,,Order Id`      |
+| `genericDialect` | `generic` | any header in DEGIRO's column layout, whatever the labels say                     |
+
+All three share the positional column layout, and the two language-aware ones share `DD-MM-YYYY` dates. They differ in how they read numbers: `frenchDialect` expects space thousands separators and a comma decimal mark, while `englishDialect` accepts either that or the US `1,060.20` form — DEGIRO does **not** switch the body of the file to English number formatting when you switch the interface language, so an English export can carry either. When a value contains both `,` and `.`, the last one is the decimal mark; a lone separator is a decimal mark unless the value is a run of exact 3-digit groups (`1,060,200`).
+
+Descriptions are a separate axis: switching the interface language rewrites the header but leaves the description text alone, so an English-header export from a French account still says `Frais DEGIRO de courtage`. The built-in matchers therefore recognise both languages regardless of which dialect read the header. Anything they do not recognise becomes an `unknown` movement — never a dropped row.
+
+### The fallback, for every other language
+
+DEGIRO's column layout is the same in every language: twelve columns, where the mutation and the balance each pair a currency cell with an amount cell under a single header label — so the 9th and 11th header cells (0-based indices 8 and 10) are always empty. `genericDialect` recognises that shape and nothing else, reading no header text at all, and interprets dates (`DD-MM-YYYY`, `DD/MM/YYYY`, `YYYY-MM-DD`) and numbers by the same per-value rules as the English dialect. It is registered last, so a French or English export never reaches it.
+
+Because it is a guess, it says so: `dialect.heuristic` is `true` and a warning lands on `ParseResult.warnings` naming the line it was decided on. Nothing silently depends on it.
+
+```ts
+const result = parseDegiroCsv(dutchCsv);
+result.dialect.id; // 'generic'
+result.dialect.heuristic; // true
+result.warnings[0].message; // 'No dialect recognised this header, so the "generic" fallback ...'
+```
+
+Two smaller guesses ride along:
+
+- If the header tokenizes to a single cell, `parseDegiroCsv` retries with `;` and then a tab before giving up. Passing `delimiter` explicitly turns this off.
+- `structuralTradeMatcher` and `structuralFxTradeMatcher` recover trades from the _shape_ of a description — `<qty> <product>@<price> <CCY> (<ISIN>)` — and take the side from the sign of the mutation, money out being a buy. Both run at a negative priority, after every language-aware matcher has declined, so `Koop 1.060 SMI ETF@106,02 CHF (CH0019852802)` classifies as a buy without anyone teaching the library Dutch.
+
+Descriptions with no such shape — a dividend, a fee, a deposit in a language the matchers do not know — stay `unknown` with their amounts intact, so balances still reconcile and cash still adds up. Register a matcher for the wording to name them.
+
 ## Domain model
 
 ### Movements
@@ -237,27 +272,31 @@ Every stage is pluggable. You rarely need to fork the library to support a new e
 
 ### Custom dialect (new locale / layout)
 
+French and English are built in; here is a third locale — Dutch headers, dot thousands, comma decimals:
+
 ```ts
-import { parseDegiroCsv, frenchDialect, parseFrenchDateTime, type Dialect } from 'libdegiro';
+import { parseDegiroCsv, DEGIRO_COLUMNS, parseDegiroDateTime, type Dialect } from 'libdegiro';
 
-const englishDialect: Dialect = {
-  id: 'en',
-  label: 'DEGIRO English',
-  columns: frenchDialect.columns, // same positional layout
-  matches: (header) => header.includes('Change') && header.includes('Balance'),
+const dutchDialect: Dialect = {
+  id: 'nl',
+  label: 'DEGIRO Dutch',
+  columns: DEGIRO_COLUMNS, // the shared positional layout
+  matches: (header) => header.includes('Omschrijving') && header.includes('Mutatie'),
   parseDecimal: (raw) => {
-    const n = raw.trim().replace(/,/g, ''); // US thousands
+    const n = raw.trim().replace(/\./g, '').replace(',', '.'); // 1.060,20
     return /^-?\d+(\.\d+)?$/.test(n) ? n : null;
   },
-  parseDateTime: parseFrenchDateTime,
-  parseDate: (d) => parseFrenchDateTime(d),
+  parseDateTime: parseDegiroDateTime,
+  parseDate: (d) => parseDegiroDateTime(d),
 };
 
-parseDegiroCsv(csv, { dialects: [englishDialect] });
+parseDegiroCsv(csv, { dialects: [dutchDialect] });
 // or force it, skipping detection:
-parseDegiroCsv(csv, { dialect: englishDialect });
+parseDegiroCsv(csv, { dialect: dutchDialect });
 ```
 
+A dialect registered through `dialects` replaces the built-ins, including the fallback; pass `createDefaultDialectRegistry().register(dutchDialect, { prepend: true })` to keep them.
+
 ### Custom classifier (new movement description)
 
 ```ts
@@ -318,4 +357,4 @@ pnpm build        # tsdown -> dist/ (ESM + .d.ts + sourcemaps)
 
 ### Test fixture
 
-`test/fixtures/Account.csv` is a **synthetic** statement, not a real export. It mirrors the shape of a genuine French DEGIRO file — column layout, movement types, order-id grouping, French decimals with `U+202F` thousands separators, double-spaced product names — and its running balances reconcile exactly, but every figure, date, ISIN and order id is fabricated. Drop your own `Account.csv` at the repo root to try the library against real data; it is git-ignored.
+`test/fixtures/Account.csv` is a **synthetic** statement, not a real export. It mirrors the shape of a genuine French DEGIRO file — column layout, movement types, order-id grouping, French decimals with `U+202F` thousands separators, double-spaced product names — and its running balances reconcile exactly, but every figure, date, ISIN and order id is fabricated. `test/fixtures/Account-en.csv` is the same statement behind an English header, which is exactly what DEGIRO produces when you switch the interface language. Drop your own `Account.csv` at the repo root to try the library against real data; it is git-ignored.

examples/dashboard/src/components/dropzone.tsx

@@ -1,10 +1,14 @@
-import { useCallback, useRef, useState, type DragEvent } from 'react';
-import { FileUp, ShieldCheck } from 'lucide-react';
+import { useCallback, useRef, useState, type DragEvent, type ReactNode } from 'react';
+import { Download, FileUp, ShieldCheck } from 'lucide-react';
 import { Button } from '@/components/ui/button';
 import { cn } from '@/lib/utils';
 import { SAMPLE_FILE_NAME, sampleCsv } from '@/lib/sample';
 import { useStatement } from '@/state/statement-context';
 
+function Ui({ children }: { children: ReactNode }) {
+  return <span className="text-foreground font-medium">{children}</span>;
+}
+
 export function Dropzone() {
   const { load, state } = useStatement();
   const [dragging, setDragging] = useState(false);
@@ -53,7 +57,7 @@ export function Dropzone() {
         <div className="space-y-1">
           <p className="font-medium">Drop your DEGIRO Account.csv here</p>
           <p className="text-muted-foreground text-sm">
-            Export it from DEGIRO under Inbox → Account statement.
+            English and French statements are both recognised.
           </p>
         </div>
 
@@ -83,6 +87,39 @@ export function Dropzone() {
         ) : null}
       </div>
 
+      <section className="text-muted-foreground w-full space-y-3 text-sm">
+        <h2 className="text-foreground flex items-center gap-2 font-medium">
+          <Download className="size-4 shrink-0" aria-hidden />
+          Exporting Account.csv from DEGIRO
+        </h2>
+        <ol className="list-decimal space-y-2 pl-5">
+          <li>
+            Sign in to <Ui>degiro.com</Ui> in a browser. The statement export lives in the web
+            client.
+          </li>
+          <li>
+            Open <Ui>Inbox</Ui>, then the <Ui>Account statement</Ui> tab.
+          </li>
+          <li>
+            Set <Ui>Start date</Ui> to the day you opened the account, or anything earlier, and{' '}
+            <Ui>End date</Ui> to today. Every number here is computed from the rows in the file, so
+            a narrower range silently gives you partial positions, fees and realized P/L.
+          </li>
+          <li>
+            Leave <Ui>Curr.</Ui> on <Ui>All</Ui> and the product search empty, so no currency or
+            instrument is filtered out.
+          </li>
+          <li>
+            <Ui>Hide cash movements</Ui> only changes the table on screen. The export contains every
+            row either way, and this dashboard needs those rows to reconcile your balances.
+          </li>
+          <li>
+            Click the download button at the top right of the table and choose <Ui>CSV</Ui>. Drop
+            the file it saves — <Ui>Account.csv</Ui> — above.
+          </li>
+        </ol>
+      </section>
+
       <p className="text-muted-foreground flex items-center gap-2 text-xs">
         <ShieldCheck className="size-4 shrink-0" aria-hidden />
         Your statement is parsed in this tab and never uploaded. The page blocks all network access,

examples/dashboard/src/components/sections/health.tsx

@@ -3,6 +3,7 @@ import type { BalanceDiscrepancy } from 'libdegiro';
 import { CheckCircle2, Copy, Info, TriangleAlert } from 'lucide-react';
 import {
   describeHealthNotes,
+  plural,
   describeHealthProblems,
   diagnosticsText,
   explainDiscrepancy,
@@ -109,9 +110,9 @@ export function HealthSection() {
         </AlertTitle>
         <AlertDescription className="flex flex-col gap-2">
           <span>
-            Parsed {health.rows} rows using the {result.dialect.id} dialect. {health.errors.length}{' '}
-            errors, {health.warnings.length} warnings, {health.unknown.length} unrecognised
-            descriptions.
+            Parsed {plural(health.rows, 'row')} using the {result.dialect.id} dialect.{' '}
+            {plural(health.errors.length, 'error')}, {plural(health.warnings.length, 'warning')},{' '}
+            {plural(health.unknown.length, 'unrecognised description')}.
           </span>
           {problems.length > 0 ? (
             <ul className="list-disc space-y-1 pl-4">

examples/dashboard/src/lib/analytics/health.ts

@@ -26,6 +26,8 @@ export interface HealthReport {
   readonly unparseableExchanges: number;
   readonly reconciliation: ReconciliationReport;
   readonly range: DateRange | null;
+  /** `true` when the header matched no language and the layout fallback read the file. */
+  readonly heuristicDialect: boolean;
   readonly ok: boolean;
 }
 
@@ -54,6 +56,7 @@ export function buildHealthReport(result: ParseResult, unparseableExchanges: num
     unparseableExchanges,
     reconciliation,
     range: statementRange(result.movements),
+    heuristicDialect: result.dialect.heuristic === true,
     ok:
       result.errors.length === 0 &&
       unknown.length === 0 &&
@@ -62,7 +65,8 @@ export function buildHealthReport(result: ParseResult, unparseableExchanges: num
   };
 }
 
-const plural = (count: number, noun: string): string => `${count} ${noun}${count === 1 ? '' : 's'}`;
+export const plural = (count: number, noun: string): string =>
+  `${count} ${noun}${count === 1 ? '' : 's'}`;
 
 /**
  * Why `ok` is false, in the user's terms.
@@ -107,11 +111,22 @@ export function describeHealthProblems(report: HealthReport): string[] {
  * ignore the panel; hiding it entirely leaves a real number unexplained.
  */
 export function describeHealthNotes(report: HealthReport): string[] {
+  const notes: string[] = [];
+
+  if (report.heuristicDialect) {
+    notes.push(
+      'No dialect recognised this header, so the file was read by its column layout alone and its dates and amounts were interpreted by guesswork. Dates, amounts and balances above are worth a spot-check against the statement.',
+    );
+  }
+
   const rounding = report.reconciliation.rounding.length;
-  if (rounding === 0) return [];
-  return [
-    `${plural(rounding, 'balance transition')} ${rounding === 1 ? 'is' : 'are'} off by less than one centime. DEGIRO rounds a half-unit price down in its amount column and up in its balance column, so the statement disagrees with itself. Nothing here is a parsing error.`,
-  ];
+  if (rounding > 0) {
+    notes.push(
+      `${plural(rounding, 'balance transition')} ${rounding === 1 ? 'is' : 'are'} off by less than one centime. DEGIRO rounds a half-unit price down in its amount column and up in its balance column, so the statement disagrees with itself. Nothing here is a parsing error.`,
+    );
+  }
+
+  return notes;
 }
 
 /** The one-sentence reading of a discrepancy: what it means and what to do. */

examples/dashboard/test/positions.test.ts

@@ -146,6 +146,28 @@ describe('describeHealthProblems', () => {
     expect(describeHealthNotes(report)[0]).toContain('1 balance transition is off');
   });
 
+  it('warns when the layout fallback read the file, ahead of any rounding note', () => {
+    const dutchCsv = [
+      'Datum,Tijd,Valutadatum,Product,ISIN,Omschrijving,FX,Mutatie,,Saldo,,Order Id',
+      '18-11-2024,00:00,18-11-2024,,,Storting,,CHF,"9.000,00",CHF,"9.000,00",',
+      '',
+    ].join('\n');
+    const parsed = parseDegiroCsv(dutchCsv);
+    const report = withDiscrepancies(buildHealthReport(parsed, 0), [
+      discrepancy('rounding', '-0.01'),
+    ]);
+
+    expect(report.heuristicDialect).toBe(true);
+    const notes = describeHealthNotes(report);
+    expect(notes).toHaveLength(2);
+    expect(notes[0]).toContain('read by its column layout alone');
+    expect(notes[1]).toContain('1 balance transition is off');
+  });
+
+  it('says nothing about the dialect when the header was recognised', () => {
+    expect(buildHealthReport(result, 0).heuristicDialect).toBe(false);
+  });
+
   it('explains a half-centime rounding from the row that caused it', () => {
     const entry = discrepancy('rounding', '-0.01', {
       statedMutation: new Money('-1528.28', 'CHF'),