Import ING and Kontist statements behind a reviewed column mapping

CSV import is now mapping-driven: N26, ING (metadata preamble, Windows-1252,
German decimals) and Kontist exports are recognized locally, and any other
layout can have its columns proposed by the configured model from a sample in
which letters are replaced by x and digits by 0. Proposals are untrusted: every
column must name a supplied header, money must come from one signed column or
one debit/credit pair, and formats must be from a closed list.

Uploading no longer imports. /api/import is replaced by prepare/confirm/cancel:
prepare parses, deduplicates and previews the exact facts, and only confirming
at the reviewed revision writes them. ING and AI-mapped facts carry no
transaction reference, because repeating SEPA mandate references must never
become a transaction identity.
This commit is contained in:
Lars Nolden
2026-09-11 17:49:03 +02:00
parent 6f791b1277
commit dc767799bc
16 changed files with 2182 additions and 301 deletions
+2 -1
View File
@@ -61,6 +61,7 @@ type App struct {
bank banking.Provider
classifier classification.Client
previews map[string]Preview
csvImports map[string]CSVImport
authStates map[string]authorization
callbackURL string
bankingSettings bankingSettings
@@ -72,7 +73,7 @@ func Open(dir string) (*App, error) {
if err != nil {
return nil, err
}
a := &App{dir: dir, journal: j, previews: make(map[string]Preview), authStates: make(map[string]authorization), syncRequested: make(chan struct{}, 1)}
a := &App{dir: dir, journal: j, previews: make(map[string]Preview), csvImports: make(map[string]CSVImport), authStates: make(map[string]authorization), syncRequested: make(chan struct{}, 1)}
fail := func(e error) (*App, error) { j.Close(); return nil, e }
if err = os.MkdirAll(filepath.Join(dir, "state"), 0700); err != nil {
return fail(err)
+302
View File
@@ -0,0 +1,302 @@
package app
import (
"context"
"encoding/json"
"net/http"
"net/http/httptest"
"reflect"
"strings"
"testing"
"finance-duck/internal/classification"
"finance-duck/internal/domain"
)
const n26Statement = "Booking Date,Value Date,Partner Name,Partner IBAN,Type,Payment Reference,Account Name,Amount (EUR)\n" +
"2026-09-01,2026-08-31,Cafe,DE02120300000000202051,Card,Lunch,Main,-12.30\n" +
"2026-09-02,2026-09-02,Employer,DE89370400440532013000,Transfer,Salary,Main,2400.00\n"
// unmappedStatement uses columns no preset recognizes, so it needs a proposed
// mapping: German booking dates, split Soll/Haben money and a free-text note.
const unmappedStatement = "Datum;Empfänger;Soll;Haben;Notiz\n" +
"01.09.2026;Cafe Sonne;12,30;;Mittagessen Berlin\n" +
"02.09.2026;Arbeitgeber GmbH;;1.200,00;Gehalt September\n"
func mockCSVMapper(t *testing.T, a *App, proposal map[string]any) *[]string {
t.Helper()
prompts := &[]string{}
mock := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
var request struct {
Messages []struct {
Role string `json:"role"`
Content string `json:"content"`
} `json:"messages"`
}
if err := json.NewDecoder(r.Body).Decode(&request); err != nil {
t.Error(err)
w.WriteHeader(http.StatusBadRequest)
return
}
for _, message := range request.Messages {
if message.Role == "user" {
*prompts = append(*prompts, message.Content)
}
}
content, err := json.Marshal(proposal)
if err != nil {
t.Error(err)
w.WriteHeader(http.StatusBadRequest)
return
}
json.NewEncoder(w).Encode(map[string]any{"choices": []any{map[string]any{
"finish_reason": "stop", "message": map[string]any{"content": string(content)},
}}})
}))
t.Cleanup(mock.Close)
a.classifier = classification.Client{APIKey: "test-private-key", Model: "test/model", BaseURL: mock.URL}
a.settings.Model = "test/model"
return prompts
}
func prepare(t *testing.T, a *App, s State, statement string) CSVImport {
t.Helper()
prepared, err := a.PrepareCSVImport(context.Background(), s.Revision, "n26", strings.NewReader(statement))
if err != nil {
t.Fatal(err)
}
return prepared
}
func TestPreparedCSVImportWritesNothingUntilConfirmed(t *testing.T) {
a, s := testApp(t)
before := domain.Clone(s.Data)
prepared := prepare(t, a, s, n26Statement)
if prepared.Source != "n26_csv" || prepared.SourceLabel != "N26" || prepared.MappedBy != "preset" || prepared.Model != "" {
t.Fatalf("recognized export was not mapped locally: %+v", prepared)
}
if prepared.Records != 2 || prepared.New != 2 || prepared.Duplicates != 0 || len(prepared.Samples) != 2 {
t.Fatalf("unexpected preview counts: %+v", prepared)
}
// The preview must show what would be written, including direction and dates.
sample := prepared.Samples[0]
if sample.BookingDate != "2026-09-01" || sample.ValueDate != "2026-08-31" || sample.Amount.String() != "-12.30" ||
sample.RawDescription != "Lunch" || sample.Counterparty != "Cafe" || sample.Currency != "EUR" {
t.Fatalf("unexpected preview sample: %+v", sample)
}
if prepared.Samples[1].Amount.String() != "2400.00" {
t.Fatalf("preview hid the credit record: %+v", prepared.Samples[1])
}
mapped := map[string]string{}
for _, column := range prepared.Columns {
mapped[column.Field] = column.Column
}
if mapped["Booking date"] != "Booking Date" || mapped["Amount"] != "Amount (EUR)" || mapped["Description"] != "Payment Reference" ||
mapped["Currency"] != "EUR (from the amount column header)" || mapped["Dates read as"] == "" || mapped["Decimal separator"] == "" {
t.Fatalf("mapping was not reviewable: %+v", prepared.Columns)
}
current, err := a.Snapshot(context.Background())
if err != nil {
t.Fatal(err)
}
if !reflect.DeepEqual(before, current.Data) {
t.Fatal("preparing an import changed the journal")
}
// Cancelling must discard the statement without importing anything.
a.CancelCSVImport(prepared.ID)
if _, err := a.ConfirmCSVImport(context.Background(), prepared.ID, s.Revision); err == nil {
t.Fatal("confirmed a cancelled import")
}
current, err = a.Snapshot(context.Background())
if err != nil {
t.Fatal(err)
}
if !reflect.DeepEqual(before, current.Data) {
t.Fatal("cancelling an import changed the journal")
}
prepared = prepare(t, a, s, n26Statement)
result, err := a.ConfirmCSVImport(context.Background(), prepared.ID, prepared.Revision)
if err != nil {
t.Fatal(err)
}
if result.Imported != 2 || len(result.State.Data.Transactions) != 2 {
t.Fatalf("confirmation did not import the previewed records: %+v", result)
}
for _, tx := range result.State.Data.Transactions {
if tx.Facts.Source != "n26_csv" || tx.Facts.Currency != "EUR" || tx.Facts.AccountID != "n26" {
t.Fatalf("imported facts differ from the preview: %+v", tx.Facts)
}
}
// A confirmed statement is consumed: confirming again must not double count.
if _, err := a.ConfirmCSVImport(context.Background(), prepared.ID, prepared.Revision); err == nil {
t.Fatal("confirmed the same prepared import twice")
}
repeat := prepare(t, a, result.State, n26Statement)
if repeat.Records != 2 || repeat.New != 0 || repeat.Duplicates != 2 {
t.Fatalf("reimport preview did not report duplicates: %+v", repeat)
}
again, err := a.ConfirmCSVImport(context.Background(), repeat.ID, repeat.Revision)
if err != nil {
t.Fatal(err)
}
if again.Imported != 0 || len(again.State.Data.Transactions) != 2 {
t.Fatalf("reimport duplicated transactions: %+v", again)
}
}
func TestUnrecognizedStatementNeedsAConfiguredModel(t *testing.T) {
a, s := testApp(t)
_, err := a.PrepareCSVImport(context.Background(), s.Revision, "n26", strings.NewReader(unmappedStatement))
if err == nil || !strings.Contains(err.Error(), "unrecognized CSV layout") {
t.Fatalf("unmapped statement without AI: %v", err)
}
current, err := a.Snapshot(context.Background())
if err != nil {
t.Fatal(err)
}
if !reflect.DeepEqual(s.Data, current.Data) {
t.Fatal("a failed mapping changed the journal")
}
// Recognized exports must keep working with no AI configured at all.
if prepared := prepare(t, a, s, n26Statement); prepared.MappedBy != "preset" {
t.Fatalf("preset import required AI: %+v", prepared)
}
}
func TestProposedMappingIsPreviewedFromRedactedSample(t *testing.T) {
a, s := testApp(t)
prompts := mockCSVMapper(t, a, map[string]any{
"booking_date_column": "Datum",
"value_date_column": "",
"amount_column": "",
"debit_column": "Soll",
"credit_column": "Haben",
"currency_column": "",
"description_column": "Notiz",
"counterparty_column": "Empfänger",
"counterparty_iban_column": "",
"date_format": "dd.mm.yyyy",
"decimal_format": "comma",
})
prepared := prepare(t, a, s, unmappedStatement)
if prepared.MappedBy != "openrouter" || prepared.Model != "test/model" || prepared.Source != "csv" {
t.Fatalf("proposed mapping was not attributed: %+v", prepared)
}
if prepared.Records != 2 || prepared.New != 2 || len(prepared.Samples) != 2 {
t.Fatalf("unexpected preview counts: %+v", prepared)
}
if prepared.Samples[0].Amount.String() != "-12.30" || prepared.Samples[1].Amount.String() != "1200.00" {
t.Fatalf("split debit and credit lost direction: %+v", prepared.Samples)
}
if prepared.Samples[0].RawDescription != "Mittagessen Berlin" || prepared.Samples[0].Counterparty != "Cafe Sonne" || prepared.Samples[0].BookingDate != "2026-09-01" {
t.Fatalf("unexpected preview sample: %+v", prepared.Samples[0])
}
if len(*prompts) != 1 {
t.Fatalf("expected exactly one mapping request, got %d", len(*prompts))
}
// Only column names and value shapes may leave this machine.
for _, secret := range []string{"Cafe Sonne", "Mittagessen", "Arbeitgeber", "12,30", "1.200,00", "01.09.2026"} {
if strings.Contains((*prompts)[0], secret) {
t.Fatalf("mapping request leaked %q: %s", secret, (*prompts)[0])
}
}
for _, header := range []string{"Datum", "Soll", "Haben", "Notiz", "Empfänger"} {
if !strings.Contains((*prompts)[0], header) {
t.Fatalf("mapping request omitted column %q: %s", header, (*prompts)[0])
}
}
result, err := a.ConfirmCSVImport(context.Background(), prepared.ID, prepared.Revision)
if err != nil {
t.Fatal(err)
}
if result.Imported != 2 {
t.Fatalf("confirmed mapping imported %d records", result.Imported)
}
for _, tx := range result.State.Data.Transactions {
if tx.Facts.Source != "csv" || tx.Facts.ExternalID != "" {
t.Fatalf("mapped facts carry an unexpected identity: %+v", tx.Facts)
}
}
}
func TestUnsafeProposedMappingsAreRejected(t *testing.T) {
valid := map[string]any{
"booking_date_column": "Datum", "value_date_column": "", "amount_column": "",
"debit_column": "Soll", "credit_column": "Haben", "currency_column": "",
"description_column": "Notiz", "counterparty_column": "Empfänger",
"counterparty_iban_column": "", "date_format": "dd.mm.yyyy", "decimal_format": "comma",
}
with := func(changes map[string]any) map[string]any {
proposal := map[string]any{}
for key, value := range valid {
proposal[key] = value
}
for key, value := range changes {
proposal[key] = value
}
return proposal
}
cases := map[string]map[string]any{
"invented column": with(map[string]any{"description_column": "Verwendungszweck"}),
"missing date": with(map[string]any{"booking_date_column": ""}),
"missing description": with(map[string]any{"description_column": ""}),
"no money": with(map[string]any{"debit_column": "", "credit_column": ""}),
"both strategies": with(map[string]any{"amount_column": "Soll"}),
"half split": with(map[string]any{"credit_column": ""}),
"unsupported format": with(map[string]any{"date_format": "%d.%m.%Y"}),
"column reused": with(map[string]any{"counterparty_column": "Notiz"}),
"unknown field": with(map[string]any{"balance_column": "Soll"}),
}
for name, proposal := range cases {
t.Run(name, func(t *testing.T) {
a, s := testApp(t)
mockCSVMapper(t, a, proposal)
prepared, err := a.PrepareCSVImport(context.Background(), s.Revision, "n26", strings.NewReader(unmappedStatement))
if err == nil {
t.Fatalf("accepted an unsafe mapping: %+v", prepared)
}
current, e := a.Snapshot(context.Background())
if e != nil {
t.Fatal(e)
}
if !reflect.DeepEqual(s.Data, current.Data) {
t.Fatal("a rejected mapping changed the journal")
}
})
}
}
func TestConfirmationRequiresTheReviewedJournalRevision(t *testing.T) {
a, s := testApp(t)
prepared := prepare(t, a, s, n26Statement)
// An unrelated edit invalidates the preview: its duplicate analysis and
// sample were computed against the revision the operator reviewed.
changed, err := a.Mutate(context.Background(), s.Revision, func(d *domain.Dataset) error {
d.Tags = append(d.Tags, domain.Tag{ID: "tag_new", Name: "new"})
return nil
})
if err != nil {
t.Fatal(err)
}
if _, err := a.ConfirmCSVImport(context.Background(), prepared.ID, prepared.Revision); err == nil {
t.Fatal("confirmed a preview of a superseded journal")
}
if _, err := a.ConfirmCSVImport(context.Background(), prepared.ID, changed.Revision); err == nil {
t.Fatal("confirmed a preview with a substituted revision")
}
if _, err := a.PrepareCSVImport(context.Background(), s.Revision, "n26", strings.NewReader(n26Statement)); err == nil {
t.Fatal("prepared an import against a superseded revision")
}
if _, err := a.PrepareCSVImport(context.Background(), changed.Revision, "missing", strings.NewReader(n26Statement)); err == nil {
t.Fatal("prepared an import for an unknown account")
}
current, err := a.Snapshot(context.Background())
if err != nil {
t.Fatal(err)
}
if len(current.Data.Transactions) != 0 {
t.Fatalf("rejected confirmations imported %d transactions", len(current.Data.Transactions))
}
}
+258 -11
View File
@@ -77,26 +77,273 @@ func (a *App) importFacts(ctx context.Context, s State, facts []domain.Facts) (I
}
return ImportResult{Imported: len(added), State: state}, nil
}
func (a *App) ImportCSV(ctx context.Context, rev, accountID string, r io.Reader) (ImportResult, error) {
// CSVColumn is one reviewable source-column assignment.
type CSVColumn struct {
Field string `json:"field"`
Column string `json:"column"`
}
// CSVImport is a parsed statement awaiting confirmation. Nothing is written to
// the journal until ConfirmCSVImport applies the exact facts previewed here.
type CSVImport struct {
ID string `json:"id"`
Revision string `json:"revision"`
AccountID string `json:"account_id"`
Source string `json:"source"`
SourceLabel string `json:"source_label"`
MappedBy string `json:"mapped_by"`
Model string `json:"model,omitempty"`
Mapping banking.CSVMapping `json:"mapping"`
Columns []CSVColumn `json:"columns"`
Records int `json:"records"`
New int `json:"new"`
Duplicates int `json:"duplicates"`
Samples []domain.Facts `json:"samples"`
facts []domain.Facts
created time.Time
}
const csvImportLifetime = time.Hour
const maxPreparedCSVImports = 5
const maxCSVSamples = 10
// PrepareCSVImport maps and parses an uploaded statement without importing it.
// Known N26, ING and Kontist exports are recognized locally; any other layout
// needs a configured model to propose a column mapping from the statement's
// redacted shape. The result must be reviewed and confirmed.
func (a *App) PrepareCSVImport(ctx context.Context, rev, accountID string, r io.Reader) (CSVImport, error) {
a.mu.Lock()
s, err := a.snapshot(ctx)
model := strings.TrimSpace(a.settings.Model)
client := a.classifier.WithModel(model)
configured := strings.TrimSpace(a.classifier.APIKey) != "" && model != ""
a.mu.Unlock()
if err != nil {
return CSVImport{}, err
}
if rev != s.Revision {
return CSVImport{}, errors.New("revision conflict: reload before importing")
}
index := slices.IndexFunc(s.Data.Accounts, func(account domain.Account) bool { return account.ID == accountID })
if index < 0 {
return CSVImport{}, errors.New("unknown account")
}
account := s.Data.Accounts[index]
file, err := banking.ReadCSV(r)
if err != nil {
return CSVImport{}, err
}
prepared := CSVImport{ID: domain.NewID("csvimport"), Revision: s.Revision, AccountID: account.ID, MappedBy: "preset", created: time.Now()}
mapping, source, label, recognized := banking.DetectCSVMapping(file)
if !recognized {
sample, e := file.Sample()
if e != nil {
return CSVImport{}, e
}
if !configured {
return CSVImport{}, errors.New("unrecognized CSV layout: import an N26, ING or Kontist export, or configure an OpenRouter key and model in Settings to map these columns")
}
proposal, e := client.ProposeCSVMapping(ctx, classification.CSVMappingRequest{
Delimiter: sample.Delimiter, Headers: sample.Headers, ShapedRows: sample.ShapedRows,
DateFormats: banking.CSVDateFormats(), DecimalFormats: banking.CSVDecimalFormats(),
})
if e != nil {
return CSVImport{}, e
}
mapping = banking.CSVMapping{
HeaderRow: sample.HeaderRow,
BookingDateColumn: proposal.BookingDateColumn,
ValueDateColumn: proposal.ValueDateColumn,
AmountColumn: proposal.AmountColumn,
DebitColumn: proposal.DebitColumn,
CreditColumn: proposal.CreditColumn,
CurrencyColumn: proposal.CurrencyColumn,
DescriptionColumn: proposal.DescriptionColumn,
CounterpartyColumn: proposal.CounterpartyColumn,
CounterpartyIBANColumn: proposal.CounterpartyIBANColumn,
DateFormat: proposal.DateFormat,
DecimalFormat: proposal.DecimalFormat,
}
source, label = "csv", "AI-mapped CSV"
prepared.MappedBy, prepared.Model = "openrouter", proposal.Model
}
facts, err := banking.ParseMappedCSV(file, account, mapping, source)
if err != nil {
return CSVImport{}, err
}
// Dedupe now so the preview reports what confirming would actually add, and
// so cross-source conflicts are reported before anything is written.
added, err := banking.NormalizeAndDedupe(s.Data, facts)
if err != nil {
return CSVImport{}, err
}
prepared.Source, prepared.SourceLabel, prepared.Mapping = source, label, mapping
prepared.Columns = csvColumns(mapping, account)
prepared.Records, prepared.New, prepared.Duplicates = len(facts), len(added), len(facts)-len(added)
prepared.Samples, prepared.facts = csvSamples(facts), facts
a.mu.Lock()
defer a.mu.Unlock()
for id, old := range a.csvImports {
if time.Since(old.created) > csvImportLifetime {
delete(a.csvImports, id)
}
}
if len(a.csvImports) >= maxPreparedCSVImports {
return CSVImport{}, errors.New("too many statements awaiting confirmation; confirm or cancel one first")
}
a.csvImports[prepared.ID] = prepared
return prepared, nil
}
// ConfirmCSVImport imports exactly the facts that were previewed, provided the
// journal has not changed since.
func (a *App) ConfirmCSVImport(ctx context.Context, id, rev string) (ImportResult, error) {
a.mu.Lock()
defer a.mu.Unlock()
prepared, ok := a.csvImports[id]
if !ok || time.Since(prepared.created) > csvImportLifetime {
return ImportResult{}, errors.New("prepared import expired or unknown; upload the statement again")
}
if rev != prepared.Revision {
return ImportResult{}, errors.New("revision conflict: reload before importing")
}
s, err := a.snapshot(ctx)
if err != nil {
return ImportResult{}, err
}
if rev != s.Revision {
return ImportResult{}, errors.New("revision conflict: reload before importing")
if s.Revision != prepared.Revision {
return ImportResult{}, errors.New("revision conflict: data changed after the preview; upload the statement again")
}
for _, account := range s.Data.Accounts {
if account.ID == accountID {
facts, e := banking.ParseCSV(r, account)
if e != nil {
return ImportResult{}, e
}
return a.importFacts(ctx, s, facts)
result, err := a.importFacts(ctx, s, prepared.facts)
if err != nil {
return ImportResult{}, err
}
delete(a.csvImports, id)
return result, nil
}
// CancelCSVImport discards a prepared statement without importing anything.
func (a *App) CancelCSVImport(id string) {
a.mu.Lock()
defer a.mu.Unlock()
delete(a.csvImports, id)
}
// csvColumns lists the mapping as reviewable field/value pairs, including where
// the currency comes from and how dates and decimals are read: an inferred
// convention is the easiest part of a mapping to get wrong.
func csvColumns(mapping banking.CSVMapping, account domain.Account) []CSVColumn {
columns := make([]CSVColumn, 0, 13)
for _, field := range []CSVColumn{
{"Booking date", mapping.BookingDateColumn}, {"Value date", mapping.ValueDateColumn},
{"Amount", mapping.AmountColumn}, {"Debit", mapping.DebitColumn}, {"Credit", mapping.CreditColumn},
{"Currency", mapping.CurrencyColumn}, {"Description", mapping.DescriptionColumn},
{"Secondary description", mapping.FallbackDescriptionColumn}, {"Counterparty", mapping.CounterpartyColumn},
{"Counterparty IBAN", mapping.CounterpartyIBANColumn}, {"Transaction reference", mapping.ExternalIDColumn},
} {
if field.Column != "" {
columns = append(columns, field)
}
}
return ImportResult{}, errors.New("unknown account")
if mapping.CurrencyColumn == "" {
currency, origin := mapping.FixedCurrency, "from the amount column header"
if currency == "" {
currency, origin = account.Currency, "from the selected account"
}
columns = append(columns, CSVColumn{"Currency", currency + " (" + origin + ")"})
}
return append(columns,
CSVColumn{"Dates read as", csvDateFormatLabel(mapping.DateFormat)},
CSVColumn{"Decimal separator", csvDecimalFormatLabel(mapping.DecimalFormat)},
)
}
func csvDateFormatLabel(format string) string {
switch format {
case "yyyy-mm-dd":
return "2026-09-01"
case "dd.mm.yyyy":
return "01.09.2026 (day first)"
case "mm/dd/yyyy":
return "09/01/2026 (month first)"
case "dd/mm/yyyy":
return "01/09/2026 (day first)"
case "iso-date-time":
return "2026-09-01T14:30:00 (date and time)"
case "iso-or-german":
return "2026-09-01 or 01.09.2026"
default:
return format
}
}
func csvDecimalFormatLabel(format string) string {
switch format {
case "dot":
return "point (1234.56)"
case "comma":
return "comma (1.234,56)"
case "dot-or-comma":
return "point or comma"
default:
return format
}
}
// csvSamples keeps a bounded, ordered excerpt that always shows the extremes and
// both directions of money when the statement contains them: an inverted sign or
// a misread date convention has to be visible before confirming.
func csvSamples(facts []domain.Facts) []domain.Facts {
if len(facts) == 0 {
return []domain.Facts{}
}
chosen := map[int]bool{0: true, len(facts) - 1: true}
if len(facts) > 1 {
chosen[1] = true
}
if len(facts) > 2 {
chosen[len(facts)-2] = true
}
credit, debit, largest := -1, -1, 0
for i, f := range facts {
minor, err := f.Amount.Minor()
if err != nil {
continue
}
if minor >= 0 && credit < 0 {
credit = i
}
if minor < 0 && debit < 0 {
debit = i
}
if previous, e := facts[largest].Amount.Minor(); e != nil || abs64(minor) > abs64(previous) {
largest = i
}
}
for _, index := range []int{credit, debit, largest} {
if index >= 0 && len(chosen) < maxCSVSamples {
chosen[index] = true
}
}
indexes := make([]int, 0, len(chosen))
for index := range chosen {
indexes = append(indexes, index)
}
slices.Sort(indexes)
samples := make([]domain.Facts, 0, len(indexes))
for _, index := range indexes {
samples = append(samples, facts[index])
}
return samples
}
func abs64(v int64) int64 {
if v < 0 {
return -v
}
return v
}
func (a *App) Backfill(ctx context.Context, rev, accountID string, historyMonths int) (ImportResult, error) {