Changelog

​

Platform-derived category signals and pre-auth rate-limiting refinements

​

Added

• Known platform intermediaries such as Uber Eats, DoorDash, Deliveroo, Wolt, Instacart, Getir, Lazada, and Shopee can now contribute deterministic transaction categories such as Food Delivery, Groceries, and Online Marketplaces.
• Added platform_category_match as a category confidence reason for platform-derived category results.

​

Changed

• Improved category accuracy for delivery, grocery, and marketplace platform transactions so known platform categories can outweigh weaker merchant-derived classifications.
• Refined pre-auth rate limiting so missing or malformed API keys are handled separately from valid-format API key cache misses, reducing false positives for legitimate authenticated traffic while still protecting authentication lookups.

​

Fixed

• Prevented duplicate platform category signals from inflating category weighting when the same intermediary is encountered through both cached and newly enriched data.

​

Upstream lookup reliability improvements

​

Changed

• Improved lookup reliability during upstream rate limits and temporary outages.

​

P2P disambiguation and parent-merchant resolution reliability improvements

​

Changed

• Improved person-versus-merchant disambiguation for transfer-style transactions by adding P2P intermediary context (platform, wallet, and transfer-token signals) to merchant analysis.
• Improved P2P routing so person-detected payees keep intermediary platform context, leading to more complete peer-to-peer enrichment results.
• Improved parent-merchant resolution reliability during enrichment with a more resilient retry path.

​

Fixed

• Prevented merchant-stage P2P placeholders from overwriting valid intermediary-derived P2P recipient data.
• Ensured merchant entities[].data.id remains canonical at the parent-merchant level when a parent profile is temporarily unavailable.
• Fixed parent-merchant redirect lookups to apply consistently for merchants missing a stored parent ID.

​

Merchant category response field and enrichment quality improvements

​

Added

• GET /v1/merchants/{id} now includes the merchant’s category field with primary, secondary, and tertiary classification (same CategoryStructure format as enrichment responses).

​

Changed

• Improved merchant name normalization internally in the enrichment pipeline.
• Improved merchant confidence scoring for single-token acronyms and abbreviations by using fuzzy domain-anchor matching against search results and resolved domains.
• Location enrichment now skips purely numeric store-ID hints because of misclassification of enrichment results.

​

Fixed

• Merchant logos are now more likely to show if existing entity internally has no logo.

​

Enrichment responsiveness and merchant normalization performance improvements

​

Changed

• Faster response times for common enrichment traffic, especially when similar transaction patterns repeat.
• Better responsiveness during high-traffic periods through more efficient capacity checks.
• Faster merchant normalization for clear brand matches, with heavier reconciliation work moved to background processing.

​

Fixed

• Reduced response delays when merchant redirect lookups are slow.
• Reduced repeated work for known unresolved redirect mappings with short-lived miss caching.

​

Intermediary alias detection and location cross-checking reliability improvements

​

Changed

• Improved intermediary detection for short processor aliases in transaction titles, increasing recognition accuracy for delimiter-based formats.
• Improved location cross-checking for store/branch-specific transactions to better validate that selected places match the transaction context.

​

Fixed

• Reduced incorrect location matches for store-ID and branch-hint transactions when geographic validation signals are missing or inconsistent.

​

Enrichment throughput and parent-merchant consolidation performance improvements

​

Changed

• Improved enrichment throughput for concurrent requests, reducing end-to-end latency in high-traffic periods.
• Improved parent-merchant consolidation flow so enrichment results are returned faster while keeping merchant identity consistency.

​

Merchant/intermediary recognition and location extraction reliability improvements

​

Changed

• Improved merchant detection for noisy and reference-heavy transaction titles.
• Improved intermediary and merchant brand recognition using broader keyword and domain matching.
• Improved location extraction reliability when using selective entity filters.

​

Fixed

• Prevented location-only extraction paths from reusing merchant values as location fallbacks.
• Reduced repeated warning noise for invalid AI model overrides in long-running environments.

​

Parent merchant matching and confidence improvements

​

Changed

• Improved parent merchant matching accuracy for brands that appear on shared marketplace/portal domains.
• More reliable merchant confidence scoring for ambiguous payment-style titles and legal-name search results.
• Better consensus handling when search evidence consistently points to a single merchant brand.

​

Fixed

• Reduced incorrect parent-merchant merges for unrelated merchants sharing the same host domain.

​

Transaction lifecycle endpoints and classification improvements

​

Added

• GET /v1/transactions/count endpoint to retrieve the total number of stored transactions, with optional afterDate filter.
• DELETE /v1/transactions/batch endpoint for bulk deletion by date range or list of IDs (max 1000). Associated cache and all internal organization related entries are purged automatically.
• Per-organization transaction retention setting (transaction_retention_days, 1–90 days) to control how long enriched results are cached.

​

Changed

• Improved person vs. merchant classification: legal entity suffixes (LTD, LLC, GmbH, B.V., etc.) now strongly favor merchant classification in both title dissection and merchant analysis.
• Improved P2P detection: transfer phrases followed by business names with legal suffixes are no longer misclassified as person-to-person transfers.
• Improved merchant detection signals: company registries, commercial directories, and invoice/reference patterns now contribute to merchant classification confidence.

​

Legal-name accuracy and IBAN anonymization improvements

​

Changed

• Accuracy improvements for legal names in transaction enrichment POST /v1/transactions/enrich pipeline.
• Improved anonymization: IBANs are now always hidden in internal pipeline processing.

​

Recurring-payment metadata removal from transaction responses

​

Breaking

• Removed the legacy recurring-payment object from enrichment transaction responses.
• Removed recurring-payment enum/type surfaces from the public API contract.

​

Migration Notes

• Action required if your integration reads or validates the legacy recurring-payment object.
• Update response parsing, DTOs, and schema validation to use data.transaction.category and data.transaction.confidence only.
• This change does not affect billing plan flows in the dashboard and Stripe.

​

Changed

• Internal enrichment persistence no longer stores legacy recurring-payment classification fields for transactions and merchants.

​

Coordinate-based location context and resolution improvements

​

Added

• Core API now accepts GPS coordinates in options.location.coordinates as location context for enrichment.
• When coordinates are provided without a city name, the API now resolves them to a nearby place name.

​

Changed

• Location enrichment is now more accurate for transactions where the title has little or no clear location text.
• If both coordinates and cityName are provided, cityName remains the source of truth.

​

Fixed

• Prevented country-level matches from being used as location results when resolving coordinates.
• Ensured coordinate-based location hints are not applied when options.filters.noLocation is enabled.

​

More conservative fallback categorization for IBAN-heavy transfer-like titles

​

Changed

• Fallback categorization is now more conservative for transfer-like titles that contain full IBAN-style account references.
• For IBAN-heavy titles, you may see lower confidence and reference_code in confidence reasons, even if transfer wording is present.

​

Fixed

• Improved reliability of request gating under high-load/throttling scenarios.
• Improved consistency of credit reservation handling when requests are denied by system-wide capacity protections.

​

Canonical brand-level merchant IDs

​

Breaking

• Merchant entities[].data.id now returns the canonical brand-level merchant ID (parent merchant UUID), not a store-level/sub-merchant ID.
• If your integration stores or joins on merchant IDs from previous versions, you should treat this as a new canonical ID space.

​

Migration Notes

• Action required if you persist merchant IDs in your database, analytics model, or downstream joins.
• No action required if you only display merchant name/icon/category in responses.
• Recommended approach:
  • Store the new entities[].data.id as your canonical merchant key.
  • Re-map historical references over time by re-enriching recent transactions or by backfilling your local mappings.

​

Added

• Parent merchant deduplication: location/store variants of the same brand are grouped under one canonical merchant.
• Responses now return a consistent merchant identity (id) and canonical brand profile across transactions for the same brand.

​

Changed

• Merchant names and descriptions are now curated at the brand level when multiple merchant observations exist.
• You may see cleaner brand naming (less branch/store-specific noise) while keeping relevant merchant metadata.

​

Channel removal and merchant-detection improvements

​

Breaking

• Removed data.transaction.channel from enrichment responses.
• Removed channel-related enums and shared types from the public API contract and SDK type surface.

​

Added

• Platform-domain validation for merchant enrichment. Social/content platform domains are no longer assigned as merchant domains unless the merchant is the platform itself.
• URL-based merchant recovery in title dissection when a merchant is missing but a URL is present in the transaction title.
• New venue_or_attraction confidence reason for non-traditional merchants such as venues, attractions, and transport operators.

​

Changed

• Improved merchant confidence scoring for abbreviated and concatenated merchant names.
• Broadened merchant classification so payees that provide goods, services, entry, or transport are more consistently recognized as merchants.
• Improved broad_merchant_name tagging so known brand names are less likely to be marked as broad/generic.
• Improved merchant domain extraction to avoid selecting review sites, app stores, and social media pages as the merchant domain.

​

Merchant-location cross-checking and fallback improvements

​

Added

• Post-enrichment merchant-location cross-checking on POST /v1/transactions/enrich to detect merchant/location mismatches and automatically re-select better location candidates when available.
• Location recovery path for cases where location initially returns no match but a merchant-aligned place exists in search results.
• New location confidence reasons in responses: merchant_location_crosscheck_corrected, merchant_location_crosscheck_recovered, merchant_location_mismatch, and geo_mismatch.

​

Changed

• Improved fallback routing for non-P2P transactions: fallback classification now runs when merchant extraction ends in no match, and income classification runs only when a merchant is confidently found.
• More conservative fallback confidence calibration for ambiguous or reference-heavy titles, with stronger routing to Uncategorized or Other Income in low-confidence cases.
• Improved title dissection for separator-heavy and URL-containing transaction strings.
• Improved country handling with stricter ISO 3166-1 alpha-2 validation and broader country-name/address resolution for multilingual aliases.

​

Fixed

• Reduced wrong-branch location matches for similarly named merchants by cross-validating merchant identity against selected location evidence.
• Fixed cases where weak fallback evidence could inflate confidence or produce over-specific categories.

​

Intermediary alias inference improvements

​

Added

• Alias inference in intermediary post-processing to improve processor/platform recognition from variant names.

​

Changed

• Improved intermediary detection reliability for mixed and abbreviated transaction strings.

​

Merchant disambiguation and fallback confidence updates

​

Added

• Expanded merchant analysis signals with country-aware disambiguation and legal-entity marker handling.
• Expanded fallback confidence reason handling for ambiguous and reference-like titles.

​

Changed

• Improved title normalization for merchant analysis and fallback categorization.
• Improved fallback confidence scoring to better reflect uncertainty.

​

Fixed

• Negative fallback reason tags now take precedence over positive tags during confidence calibration.

​

Enrichment cache performance and observability updates

​

Changed

• Faster enrichment retrieval path with cache transaction lookups.
• Improved entity-cache lookup/save performance for lower enrichment latency.
• Added more granular cache phase tracking for enrichment observability.

​

New options field for enrichment endpoint

​

Added

• Enrichment options on POST /v1/transactions/enrich: new optional options field in the request body.
  • Filters (options.filters): set noMerchant, noIntermediary, or noLocation to true to skip extraction of that entity type. Useful when you already have the data or want faster responses.
  • Pre-filled merchant (options.merchant): supply id, name, or domain when you know the merchant. The API will use it directly instead of extracting from the title. IDs must exist in our system (returns 422 if not found); names and domains use best-effort lookup.
  • Pre-filled location (options.location): supply cityName, streetName, storeNumber, physicalLocation, or coordinates to improve location accuracy.
  • Pre-filled intermediaries (options.intermediaries): supply up to 5 intermediaries by id, name, or domain when you know the payment processor(s).
  • You cannot combine a filter (e.g. noMerchant) with pre-filled data for the same entity type; the request will return 422.

​

Changed

• Improved extraction accuracy for transaction titles that include country codes (e.g. NETFLIX NL).
• Improved handling of reference-like patterns (payment IDs, order numbers).
• When you pre-fill merchant and skip intermediary + location, the API may bypass title dissection for lower latency.

​

Fixed

• Titles ending in a country code (e.g. NETFLIX NL) are now parsed correctly instead of being treated as merchant-only.

​

Incremental accuracy and performance improvements

​

Changed

• Incremental accuracy and performance improvements across the enrichment pipeline.
• Improved entity detection for noisy transaction strings, reducing merchant false negatives.
• Faster cold-start behavior on POST /v1/transactions/enrich.
• Tuned confidence reason-tagging to make confidence scores more reliable, especially for merchant and location entities.
• Additional latency improvements in enrichment processing.