Extraction was previously manual: an agent read each PDF, mapped the columns
by hand and wrote data/source/invoice-*.json. That was the only way anything
got in, and the upload button just filed the document without reading it.
That supplier's report is one fixed template — the columns sit at identical
x-positions on every invoice, and only the line count and page count vary —
so the figures are now read by coordinate instead. Deliberately not an AI
extractor: this input does not need one, and a deterministic parse that fails
loudly is worth more here than a flexible one that can hallucinate a total.
lib/invoice-parser.ts reads the header, the machine lines across all pages,
the "Plant Hire Total" row and the SUB TOTAL / Escalation / PLUS VAT /
INVOICE TOTAL block, then applies the same arithmetic gates the seed does:
lines must sum to the subtotal, each line must satisfy qty x rate = amount
and amount + VAT = total, and VAT must match the stated rate. Anything that
does not reconcile is refused with the reason — never partially imported.
Validated by parsing all five existing invoices and comparing every field
against the hand-extracted JSON: identical, including the R24,205.13-VAT
column that a plain text read gets wrong. PHI 169 was then deleted and
re-ingested through the HTTP endpoint alone and came back field-for-field
identical, with no agent involved.
- "Add Invoice" drops a PDF in and imports it; lines are matched to plant
items by the code the invoice prints, and an unmatched line (the
container codes MPLANT04/MPLANT05) is left unassigned rather than guessed.
- A "Problem invoices" inbox lists every PDF that failed, with the reason,
and Retry re-reads the stored document and re-runs the parse. Retry
re-reports rather than claiming success, and only files imported data.
- invoice_intake collection records each attempt.
pdfjs-dist is in serverExternalPackages: bundled, it rewrites its own lazy
worker import to a .next path absent from the standalone image and every
parse dies with "Setting up fake worker failed". The parser also copies the
buffer before handing it over, since pdfjs transfers and detaches it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
WIP carried over from the working tree, committed separately so it does not
sit inside the invoice work.
- inline Edit cells for code / name / type and an active toggle
- GET /api/assets/:id/impact returns the cascade counts so the client can
confirm before a delete
- DELETE /api/assets/:id hard delete; PocketBase cascade clears the dependent
income / running / cost-line rows
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds an Invoices sub-menu under /income with a summary table (one row per
invoice, click to expand the full detail) and the data model behind it.
Data:
- invoices + invoice_lines PocketBase collections, created idempotently by
scripts/setup-invoice-collections.mjs
- line items transcribed from PHI 166/167/171 with per-invoice arithmetic
validation in the seed; a bad invoice is refused without blocking the rest
- invoice-*.json in data/source are auto-imported, so adding an invoice is
one file
Booking:
- manual book-in per line (Booked to), grouped by category
- POST /api/invoice-lines/:id/book moves the income row rather than adding a
second one, so a booking can be changed freely
- scripts/book-invoice.mjs is idempotent: rows are stamped with
monthly_income.invoice_line and replaced on re-run
- booked amount follows lib/invoice-booking-policy.json (bookInclusiveVat),
read by both the app and the CLI so they cannot disagree
Money:
- the R24,205.13 column on these invoices is PLUS VAT, not escalation. The
seed now rejects VAT that disagrees with the stated rate.
- booked amounts are VAT-inclusive per the client ruling of 2026-10-01; the
VAT stays visible rather than hidden.
Uploads:
- drag-and-drop PDF attached to invoices.source_pdf
- GET /api/invoices/:id/document proxies it — the collection is rule-locked,
so a direct PocketBase /api/files link 404s in a browser
Fixes a month-aggregation bug this introduced: buildIncomeRows assigned into
months[i], so a machine with two income rows in one month (one per invoice,
which is normal multi-site use) showed only the last. SS01 September read
R20,700 instead of R35,305 while its own YTD included both. Months now
accumulate, per-month profit is computed once after accumulation, and
lib/rows.test.ts pins it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Part A — Round 3 regressions:
- A.1: verify-cost-reconciliation.mjs now loads .env.local, wraps in try/catch, exits 1 on any failure
- A.2: Vehicle Costs restored Santam — back to R 14,996.31 (matches row-by-row sum)
- A.3: incl<excl / VAT math / VAT>20% now FATAL — seed exits 2 with full report (4 anomalies on Charl-CCS + Buildsmart)
- A.4: Reconciliation target R 694,962.93 with both deltas itemised (Personnel +R19,800, Santam -R5,322.44, unexplained R 0.00)
- A.5: 'Lines' view uses actual lines (recharts LineChart), no fills; fills only in 'Cost stack'
Part B — Mobile standard:
- B.1: header wraps to 2 rows on mobile; YTD badge cards stack; page-wide overflow eliminated on /income, /charts, /overheads, /running, /expenses
- B.2: IncomeView card row layout stacked — name gets its own line on mobile, never truncated to single char
- B.3: Plant type filter is a <select> dropdown below lg (was 18 pills wrapping to 13 rows); same for sort on /income and /running
- B.4: Plant table first column sticky; -webkit-overflow-scrolling: touch; tables scroll inside their own container
- B.5: All buttons/links use min-h-[44px] tap targets; text below 12px is bumped where it matters
- B.6: Charts legend wraps below title; chart min-height h-44 → h-52 on mobile
- B.7: paddingTop env(safe-area-inset-top) on sticky header for notch/home indicator clearance
AGENTS.md created with the standing mobile checklist rule.
