feat(#71): migrate service layer to list[Transaction] — remove dict round-trips

[longieirl]

Summary

Closes #71. Migrates the service layer to use list[Transaction] throughout, eliminating the dict round-trips that previously occurred in every service call. Also fixes a regression introduced during migration where CSV output had empty Debit/Credit/Balance columns.

Changes

• Transaction gains document_type and transaction_type fields; from_dict()/to_dict() updated
• IDuplicateDetector, ITransactionSorting, IIBANGrouping, ITransactionFilter protocols updated to list[Transaction]
• DuplicateDetectionStrategy.detect_duplicates() and all create_key() methods accept Transaction directly
• ServiceRegistry.process_transaction_group() / group_by_iban() operate on list[Transaction]; enrichment helpers mutate tx fields in place
• processor.py passes list[Transaction] through pipeline; dicts only at output boundary via transactions_to_dicts()
• Bug fix: transactions_to_dicts() now passes currency_symbol="" so dict keys ("Debit") match neutral DEFAULT_COLUMNS names — previously caused empty CSV columns
• Unused transactions_to_dicts import removed from pdf_processing_orchestrator
• All 10 affected test files updated: fixtures use Transaction.from_dict(), assertions use attribute access
• Converter tests updated to assert neutral column names; new regression test builds a real DataFrame to prove key alignment

Type

• Bug fix
• New feature
• Breaking change
• Refactoring
• Documentation
• Performance
• Security

Testing

• Tests pass (coverage ≥ 91%)
• Manually tested
• make docker-integration passed locally (required when touching Dockerfile, entrypoint.sh, docker-compose.yml, or packages/parser-core/)

Checklist

• Code follows project style
• Self-reviewed
• Documentation updated (if needed)
• No new warnings

[longieirl]

Code review

Found 2 issues:

1. _is_header_transaction calls tx.to_dict() with default currency_symbol="€", inflating non_empty_fields and diluting the 50% header-detection threshold

to_dict() emits 5 extra metadata keys beyond the column fields (source_page, confidence_score, extraction_warnings, document_type, transaction_type). These are always non-empty after enrichment, so they inflate non_empty_fields without contributing to matches. A header row that matched 3/5 fields (60%) pre-enrichment would match 3/8 fields (37.5%) post-enrichment and slip through the filter. The call should be tx.to_dict(currency_symbol="") at minimum, but ideally _is_header_row should explicitly skip the known metadata keys (as it already skips "Filename").

bankstatementprocessor/packages/parser-core/src/bankstatements_core/services/transaction_filter.py

Lines 153 to 159 in 9f8e412

	def _is_header_transaction(self, tx: Transaction) -> bool:
	"""Check if a Transaction is a header row by inspecting its field values."""
	row = tx.to_dict()
	return self._is_header_row(row)
	def _is_header_row(self, row: dict) -> bool:
	"""Check if a row is a header row.

2. DateAmountDuplicateStrategy.create_key now produces different keys than the old dict-based code — source_page and confidence_score were previously summed incidentally as floats

The old code iterated all dict keys and summed any value parseable as a float. source_page (e.g. "1" → 1.0) and confidence_score (e.g. "0.95") were therefore included in the total. The new code explicitly iterates only ("debit", "credit", "balance") plus additional_fields, so those two values are dropped. A transaction with debit=50.00, balance=100.00, source_page=1, confidence_score=0.95 previously produced a key with total 151.95; now it produces 150.00. Transactions from overlapping statement periods that were previously detected as duplicates will no longer match.

bankstatementprocessor/packages/parser-core/src/bankstatements_core/patterns/strategies.py

Lines 117 to 133 in 9f8e412

	def create_key(self, transaction: "Transaction") -> str:
	"""Create key from date and sum of all amounts."""
	total = 0.0
	for field_name in ("debit", "credit", "balance"):
	value = getattr(transaction, field_name, None)
	if value:
	amount = to_float(str(value))
	if amount is not None:
	total += abs(amount)
	for value in transaction.additional_fields.values():
	if value:
	amount = to_float(str(value))
	if amount is not None:
	total += abs(amount)
	return f"{transaction.date}:{total:.2f}"

🤖 Generated with Claude Code

_{If this code review was useful, please react with 👍. Otherwise, react with 👎.}