Session:13:02
(finanzas — uarup receipt corpus → project-tag seed)
Investigated ../uarup/data/: 445 WhatsApp chats (JSONL
+ transcripts), 3,888 media files with classification.jsonl
(2,408 classified, 337 receipt OCR samples),
media-dupes.jsonl, chat-name resolution in
chats.json
Key chats mapped: Por mis Pagos (recurring bills:
Edenor/Metrogas/AYSA/expensas, card payments, Adriana rent, Gleria
psico), Notas CHCM (Suipacha obra receipt dump —
vidrios/ferretería/tranquera/lavadero purposes in OCR)
Verified the tagging bridge: Galicia statement descriptions carry
counterpart CUIT verbatim (receipt CUITs ↔︎ ledger rows match; e.g.
20374635663 Barrionuevo tranquera receipt ↔︎ ledger row)
Drafted ~21 CUIT-keyed rules.edn rows in
.wai/projects/finanzas/research/2026-10-04-project-tag-seed-from-uarup-receipts.md;
user go on obra/sucesion
Applied + parsed: 65 project: obra, 4
project: sucesion tags, ~120 reclassifications;
bonus fix — Consorcio/PyCA rows previously misrouted to
Transfers:Currency-Exchange by the kielbowicz FX token
Fixed own mistakes: malformed EDN from sed (missing }),
typos gate (Routin surname → _typos.toml)
just test all suites 0 failures; ah check
ok; committed ed3ee05, pushed, finanzas-2s8
closed
Next: HITL leftovers from the research doc: Gaete
mixed-purpose (llave vs Londres2026), Lucia Dibner,
comida perro needs a Pets account decision,
finanzas report project obra review
(reports/project-obra.org: 21.1M ARS)