Jakob WennbergandClaude Opus 4.7
07a7964e8d
* fix(supplier-invoices): guard against duplicate payment when bank tx already booked
Two-pronged fix for a UX trap where a supplier invoice could be marked paid
even though the bank payment was already booked on 2440, creating a duplicate
verifikation.
Prong A — mark-paid duplicate guard: before booking, scan for an unlinked
outgoing bank transaction matching this supplier (merchant_name ILIKE) within
±2% / ±60 days. If found, return 409 SI_PAID_LIKELY_DUPLICATE with candidates
so the UI can offer "link existing" instead. Override via { force: true }.
Prong B — categorize match suggestion: when the user assigns 2440 directly on
a negative business transaction and an open supplier invoice from the same
supplier covers the same amount, return 409 TX_CATEGORIZE_SUGGEST_SI_MATCH
with candidates and route the user to match-supplier-invoice. Override via
{ confirm_no_match: true }.
Frontend dialogs added on the supplier-invoice detail page and the
transactions inbox. Partial payments skip the mark-paid guard (deliberate
action). Tests cover the 409 path, the override path, and the no-candidates
happy path on both routes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(supplier-invoices): apply PR review fixes to duplicate-payment guards
- Add `is_business = true` filter to mark-paid candidate query so private
bank withdrawals don't surface as false-positive duplicates
- Escape LIKE wildcards (`%`, `_`, `\`) in both ILIKE patterns to avoid
silent over-matching when a supplier/merchant name contains those chars
- Round paymentAmount and remaining_amount to 2 decimals before the
partial-payment guard comparison to avoid float-equality fragility
- Require credit account to be in the 1xxx (bank/cash) series for the
Prong B 2440 intercept so 2440 against clearing/equity accounts isn't
misinterpreted as a supplier payment
- Extract DUPLICATE_AMOUNT_TOLERANCE_PCT (0.02) and
DUPLICATE_DATE_WINDOW_DAYS (60) into a shared helper module with the
LIKE-escape utility
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(supplier-invoices): broaden 244x match, audit log overrides, drop JE id from response
Second-round PR review fixes:
- Widen Prong B regex from /^2440$/ to /^244\d$/ so payments mapped to BAS
sub-accounts (e.g. 2441 leverantörsskulder i utländsk valuta) also trigger
the suggestion (swedish-invoice-compliance bot)
- Log a structured warning when force=true or confirm_no_match=true is honored,
with the relevant context (amount, date, accounts) so the override is
traceable per BFNAR 2013:2 kap 8 (behandlingshistorik)
- Drop journal_entry_id from the SI_PAID_LIKELY_DUPLICATE candidate response
payload (data minimization, GDPR Art.5(1)(c)); the UI never rendered it
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(supplier-invoices): third-round PR review — date window on Prong B, length cap, VAT message
- Add the missing date window to the Prong B categorize candidate query
(swedish-compliance bot): without it, an open invoice from years back can
surface as a "match" for an unrelated bank transaction. Uses the shared
DUPLICATE_DATE_WINDOW_DAYS against invoice_date.
- Cap supplier/merchant names to 200 chars before they enter escapeLikePattern
(OWASP V1.2.5 / ISO A.8.28). Bounds DB work on pathological inputs.
- Log a structured warning when the mark-paid guard is skipped because the
invoice has no resolved supplier name (BFL 5 kap 7 § — motpart should be
identifiable; the absence is itself worth surfacing).
- Update the SI-match suggestion error and the matching UI copy to call out
the actual compliance risk: a duplicate 244x posting double-deducts ingående
moms (ML 8 kap 3 §), not just bookkeeping symmetry.
- Reword "Bokför på 2440 ändå" to "Bokför på leverantörsskulder ändå" now that
the regex covers BAS sub-accounts 244x.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(supplier-invoices): correct Prong B framing — duplicate verifikation, not VAT double-deduction
Latest swedish-compliance review correctly walked back the earlier
finding that asked for ML 8 kap 3 § VAT framing. Plain 244x
categorization via account_override does not include VAT lines (account
class 2), so the risk is a duplicate verifikation (BFL 5 kap 5 §), not
a double VAT deduction. Update both the structured error message and
the dialog body to reflect the actual mechanism.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
31 lines
1.3 KiB
TypeScript
31 lines
1.3 KiB
TypeScript
/**
|
|
* Shared constants and helpers for the duplicate-payment / SI-match guards
|
|
* used by `/api/supplier-invoices/[id]/mark-paid` and
|
|
* `/api/transactions/[id]/categorize`. Both guards look for a likely-matching
|
|
* counterparty within a fuzzy amount + date window; keeping the thresholds in
|
|
* one place makes them tunable as we learn from real false-positive rates.
|
|
*/
|
|
|
|
/** Acceptable amount drift (±) when matching a bank tx to an invoice amount. */
|
|
export const DUPLICATE_AMOUNT_TOLERANCE_PCT = 0.02
|
|
|
|
/** Date window (±days) around the payment / invoice date. */
|
|
export const DUPLICATE_DATE_WINDOW_DAYS = 60
|
|
|
|
/** Cap on supplier / merchant names before they enter an ILIKE pattern, to
|
|
* bound query work and avoid pathological inputs degrading the index scan. */
|
|
const MAX_LIKE_NEEDLE_LENGTH = 200
|
|
|
|
/**
|
|
* Escape LIKE/ILIKE wildcards (`%`, `_`, `\`) and truncate to a safe length
|
|
* before embedding the value in an ILIKE pattern. SQL-injection is already
|
|
* handled by Supabase's parameterization; this purely prevents silent
|
|
* over-matching on names like "50% Off AB" and bounds DB work on long inputs.
|
|
*/
|
|
export function escapeLikePattern(value: string): string {
|
|
const truncated = value.length > MAX_LIKE_NEEDLE_LENGTH
|
|
? value.slice(0, MAX_LIKE_NEEDLE_LENGTH)
|
|
: value
|
|
return truncated.replace(/\\/g, '\\\\').replace(/%/g, '\\%').replace(/_/g, '\\_')
|
|
}
|