- Rewrite detectAllHeaderSets() with repeat-field detection instead of gap-based splitting
- Require ≥2 header fields per group (filters out false matches like 'Left Over | Paid')
- Fix column leakage: right-side bills no longer pick up left-side amounts
- Add header_set_index to analyzeRow return object for frontend use
- Add isLikelySummaryRow() filter (Paycheck, Left Over, Enter how much, etc.)
- Expand isLikelyTotalRow() to catch 'Auto Total ------>' patterns
- Filter leftover calc rows (null name + negative amount, dash separators)
- Remove 'paid' from HEADER_PATTERNS.amount (was false-matching 'Paid' cells)
- Skip empty string cells in detectAllHeaderSets
- detectAllHeaderSets() finds multiple header groups per row (left 1st / right 15th)
- isBlankRowForHeaderSet() checks blanks per column range for dual layouts
- parseSheetRows() scans rows 0-4 for header row, processes each set independently
- analyzeRow() computes due_day from date/label/pattern with fallback to defaultDueDay
- Cell type validation allows 's' (shared formula) type
- Non-numeric amounts (auto, double pay, past due) become detected labels
- Day patterns (1st, 15th, 24th) parsed as due_day values
- Security: bounds validation in isBlankRowForHeaderSet, anchored regex, label sanitization
null