Formulario 120 and 515 assembly, the review screen, approval, the PDF, and the
guided checklist that walks the user through presenting it in Marangatu.
Approval is a promise about what the user actually saw: the numbers are
recomputed on the way in, and if the documents moved since the declaration was
generated the answer is a 409 with the fresh figures stored as a draft, not a
silent approval of numbers nobody read. An approved declaration is never
regenerated or invalidated.
Editing, confirming, rejecting or reclassifying a document flips any ready
declaration covering its period back to draft, which is what the dashboard
reads to stop offering it for review.
The PDF is rendered once and cached against the declaration, so two downloads
are byte for byte the same document; every write that changes the numbers
clears the cache. Approval pre-renders it, and the route renders on demand, so
a failed job costs a wait rather than a missing file. It stays Spanish in both
locales, like the form it mirrors, and carries a BORRADOR watermark until it is
approved.
The seeded declarations are computed through the same code path the product
uses, so their figures agree with the documents behind them.
Three defects the screenshots caught: the floating scan button swallowed the
tap meant for Aprobar on a phone, the summary breakdown reused labels meant for
other screens, and five tab labels collided at 390px.
Tests that can avoid depending on a fresh seed now do; the ones that cannot say
so in the assertion rather than timing out. db:reset warns that the API has to
be stopped first, having learned that the hard way.
293 vitest tests, 59 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>
