The 2024-era CuentaRUT cartola wraps the ABONOS O DEPOSITOS header across
three lines and splits each transaction (docno / description / amount+date)
across multiple lines, so the old same-line header+row detection found 0
transactions. Rewrote row assembly as docno-delimited blocks with tolerant
column detection (ABONOS position recovered from neighbouring lines).
Recovers 12 transactions from the backfill statement; all 2025 statements
parse identically (36/5/36/12/36/29/2/5 unchanged). Backfill now 2375 txns.
- parsers/parse_backfill.py: decrypts staging tree, parses Santander /
Banco de Chile / Tenpo / BancoEstado / CopecPay into a SEPARATE
web/ledger-backfill.json (2363 txns, 2018-01 to 2024-12, 480 statements)
- web/engine.js: dashboard now honors ?ledger=<file>.json so the backfill
can be reviewed via dashboard.html?ledger=ledger-backfill.json without
touching the live ledger.json
- .gitignore: exclude documents/ (decrypted bank PDFs must never be committed)
Kavi