The dashboard, the deadline engine, the three sweeps and the notification
channels. Confirm a comprobante and the IVA and IRP figures move; every
headline number opens the documents behind it.
The dashboard does no arithmetic of its own: it picks inputs and calls
packages/rules. The acceptance test asserts the API's numbers equal what
computeF120 and computeF515 produce over the same rows, before and after the
bandeja is confirmed, so a drift in either direction fails.
Deadlines come from the calendario perpetuo, skipping any period earlier than
the date the taxpayer took the obligation on. Without that guard a brand new
account opens on a red overdue card for a period that predates it.
Sweeps are idempotent by dedupe key rather than bookkeeping: a restarted
poller, a second replica and a crash mid-sweep all converge on one run, and one
reminder per user per period per milestone. They queue notifications rather
than sending them, so a channel being down retries on the job schedule. The
T-10 test derives the date from dueDateFor rather than restating the calendar.
Auto-confirm only touches what the rules were confident about and never a
decision the user already made. An unconfigured channel is absent rather than
broken: the fan-out skips it, the UI hides it, and a frozen account receives
nothing.
Web: the dashboard's three zones with swipeable position cards and traceable
numbers, the deadlines timeline, and the app shell with the tab bar and the
persistent scan button FLOWS.md asks for.
Two additions to the specs, both marked: hasDocuments on DashboardDto, without
which a position of all zeros is indistinguishable from a real one and the
first-run state never shows; and an insight_dismissals table, which FLOWS.md
requires and SPEC.md has nowhere to put.
Also: pnpm db:reset, because the e2e suite changes the seed it runs against and
a suite that is not repeatable is not a suite.
280 vitest tests, 51 Playwright tests, rules coverage still 100%, typecheck and
lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The whole pipeline: storage behind one driver interface (local disk and S3),
a portable job queue with a poller, QR and CDC parsing, OCR through the
Anthropic API, dedupe, manual entry, and the bandeja that turns all of it into
one decision per card.
Scanning tries the trustworthy door first: a QR is parsed and prefilled from
its CDC; a photo without one is queued for OCR when a key is configured and
otherwise opens the manual form against the stored file. A QR that will not
parse records an ingest error and falls through rather than losing the photo.
Job claiming is the only dialect divergence, as SPEC allows: FOR UPDATE SKIP
LOCKED on Postgres, a conditional UPDATE against SQLite's single writer. Retry
backoff follows SPEC exactly and a job abandoned by a killed process returns to
the queue once its lock goes stale, which is the phase 3 acceptance case.
OCR uses structured outputs rather than parsing prose, so the model cannot
return anything but the RULES.md schema, and every field is nullable because
unreadable is a real answer.
Web: scan with live QR decoding (BarcodeDetector, ZXing fallback, wasm served
from our own origin), manual entry with the IVA split worked out from the
total, the bandeja with swipe, buttons and keyboard all doing the same thing,
and the documents list and detail with an editable classification.
The seed now carries Maria's 34 purchases and 8 sales and Carlos's 6, all
classified through the real rules, plus the two open ingest errors.
Two defects found and fixed with tests: seeded documents could be dated in the
future, which would corrupt any projection computed from them, and the category
buttons announced their keyboard shortcut as part of their name.
255 vitest tests, 37 Playwright tests, rules coverage still 100%, typecheck and
lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
