Skip to content

CHANGELOG.md

Latest commit

 

History

History
87 lines (67 loc) · 8.4 KB

CHANGELOG.md

File metadata and controls

87 lines (67 loc) · 8.4 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.1.3] — 2026-04-10

Added

  • CC statement support (#28, #29, #31–#34, PR #122) — full credit card pipeline for paid tier: ExtractionResult.card_number, BankTemplate.column_aliases, CCGroupingService grouping by last-4 card suffix, wired into ServiceRegistry.group_by_card(). processor.run() splits on card_number is None to route bank vs CC results.
  • aib_credit_card.json template (#129, PR #138) — correct CC column boundaries (Transaction Date 29–80, Posting Date 80–118, Transaction Details 118–370, Amount 370–430) so RefContinuationClassifier and RowMergerService handle CC two-line transaction splits correctly without falling back to bank columns.
  • PDFExtractorOptions dataclass (#109, PR #139) — groups 8 optional PDFTableExtractor constructor params into a single options object, reducing the constructor from 11 params to 3.
  • Pylint design gate in CI (#85, PR #104) — Xenon complexity gate and Pylint design checks added to CI pipeline.

Fixed

  • #129 — Non-transaction (empty/phantom) rows in CC CSV/JSON output eliminated. Root cause: missing CC template caused Ref: lines to be misclassified as transactions. Fixed by adding aib_credit_card.json (PR #138) and an earlier classifier fix (PR #133).
  • #131 — CC amounts ending in CR now populate the Credit column instead of Debit. reroute_cr_suffix() added to currency.py, wired via RowPostProcessor._reroute_cr_amounts() (PR #135).
  • #132 — CC transactions sorted incorrectly due to yearless dates. Year inferred from Payment Due date in statement; ordinal date suffixes added to _PAYMENT_DUE_PATTERNS (PR #133).
  • #134 — CC output dates now include the statement year (e.g. 4 Feb 2025). Transaction._enrich_date() appends year via to_dict() (PR #137).
  • #125 — Unknown-IBAN group was producing output files instead of routing to excluded_files.json (PR #126).
  • #123 — Free-tier pipeline was producing CC grouped output files instead of routing to excluded_files.json. CC grouping now gated behind paid-tier entitlement check (PR #124).
  • #106 — Credit card PDFs were unconditionally skipped on the paid tier; now correctly processed (PR #108).
  • #110 — data_retention_days was not forwarded to DataRetentionService (PR #115).
  • #78 — date_propagated extraction warnings suppressed from JSON/CSV output (PR #93).
  • #90 — 214 logging f-string violations (G004) replaced with %-formatting (PR #100).
  • #98 — _detect_text_based_table decomposed to pass Xenon C complexity gate (PR #101).
  • #80 — Pre-existing unused imports (F401) removed from test files (PR #94).

Changed

  • Service layer migrated to list[Transaction] (#71, PR #79) — all services accept/return list[Transaction]; dict round-trips removed. Output boundary conversion via transactions_to_dicts(currency_symbol="").
  • Currency-agnostic field names (#62–#64, #66, PR #67) — TransactionRow fields renamed _EUR → _AMT; strip_currency_symbols() unified in domain/currency.py; currency_symbol defaults to "" throughout.
  • ruff replaces flake8 (#84, #89, PR #91) — ruff lint config in both pyproject.toml files; pre-commit hook updated to astral-sh/ruff-pre-commit v0.8.0.
  • pip-audit replaces safety (#86, PR #97) — dependency vulnerability scanning updated.
  • Hadolint + pinned trivy-action added to CI (#87, PR #96).
  • [skip downstream] support added to dispatch-downstream CI job (PR #82).
  • #111, #112, #113 — Dead fields removed from ExtractionConfig and ExtractionScoringConfig; dead scoring_config param removed from PDFTableExtractor (PRs #116, #117).
  • CONTRIBUTING.md — Coverage threshold corrected to 91% to match pyproject.toml (PR #140).

[0.1.2] — 2026-03-25

Fixed

  • #47 — filter_service.apply_all_filters() result was computed and logged but silently discarded. Filtered rows are now written back to result.transactions in PDFProcessingOrchestrator.process_all_pdfs(), so filter_empty_rows, filter_header_rows, and filter_invalid_dates are applied to every successfully extracted PDF.
  • #52 — BankStatementProcessorBuilder.with_duplicate_strategy() and .with_date_sorting() were inert: build() called ServiceRegistry.from_config() with no services, causing the registry to create its own defaults and silently ignore the configured strategy. The builder now constructs DuplicateDetectionService and TransactionSortingService from its configured values and passes them explicitly into ServiceRegistry.from_config().
  • #55 — Credit card / no-IBAN PDFs excluded from the pdfs_extracted count in processing output. process_all_pdfs() now returns a 3-tuple (results, pdf_count, pages_read).

Changed (architecture cleanup — PRs #56, #57)

  • #49 — ChronologicalSortingStrategy sorts dicts directly via DateParserService, removing a redundant Transaction round-trip.
  • #48 — Deferred circular imports in processor.py removed; service_registry, monthly_summary, and expense_analysis import ColumnAnalysisService/DateParserService directly at module level.
  • #50 — TransactionClassifier._looks_like_date delegates to RowAnalysisService.looks_like_date, removing a duplicate regex and fixing a subtle 1-or-2-digit day matching bug.
  • #51 — ProcessorFactory.create_from_config() builds ProcessorConfig in one block via BankStatementProcessorBuilder.with_processor_config(); new config knobs now touch ≤2 files.

[0.1.1] — 2026-03-25

Added (v1.1 — Transaction Pipeline & Word Utils)

  • Transaction enrichment (source_page: int | None, confidence_score: float, extraction_warnings: list[str]) — all three fields default correctly and survive to_dict / from_dict round-trips (#16 / Phase 21).
  • ExtractionResult dataclass (domain/models/extraction_result.py) — typed extraction boundary with transactions, page_count, iban, source_file, and warnings fields. Architecture guard test enforces placement in domain/models/ (#16 / Phase 22).
  • End-to-end ExtractionResult pipeline — PDFTableExtractor.extract(), ExtractionOrchestrator, PDFProcessingOrchestrator, and processor all produce and consume ExtractionResult; zero tuple-index unpacking remains (#16 / Phase 23).
  • extraction/word_utils.py — canonical module for group_words_by_y, assign_words_to_columns (with strict_rightmost flag), and calculate_column_coverage. Five callers migrated; four private duplicate methods deleted (#21 / Phase 24).

Changed

  • ServiceRegistry introduced (feat/28, PR #44) — ServiceRegistry.from_config(ProcessorConfig, Entitlements) wires all transaction-processing services. TransactionProcessingOrchestrator deleted (PR #46 / issue #45).
  • ClassifierRegistry with explicit integer priorities added to row_classifiers.py (fix/29, PR #39).
  • recursive_scan default changed False → True in ProcessingConfig, AppConfig, ProcessorBuilder, and PDFDiscoveryService; RECURSIVE_SCAN env var added to docker-compose.yml (fix/40, PR #41).
  • ScoringConfig injectable via BankStatementProcessorBuilder.with_scoring_config() (feat/32, PR #36).

[0.1.0] — 2026-03-24

Added (v1.0 — Architecture RFC)

  • extraction/word_utils.py foundation work — RowClassifier chain injected as shared dependency (issue #17, PR #22).
  • PDFTableExtractor decomposed into PageHeaderAnalyser, RowBuilder, and RowPostProcessor (issue #18, PR #23).
  • Facade passthroughs deleted — content_analysis_facade.py, validation_facade.py, row_classification_facade.py removed; service→shim circular import chain broken (issue #20, Phase 20).
  • pdf_table_extractor.py shim rewired to module-level singletons; pdf_extractor.py cleaned of four lazy facade imports.
  • Architecture guard test test_facade_modules_deleted added.

Changed

  • Credit card templates (aib_credit_card.json, credit_card_default.json) removed from open-source repo; credit card support is PAID tier only via require_iban=False in Entitlements.paid_tier().