I will run invoice automation to validate and reconcile
Ecommerce Catalog and Digital Operations Specialist
About this Gig
I'll run your invoice batch through a bounded, offline pipeline: classify against one of two documented text layouts, validate every field, flag duplicates, and apply five reconciliation checks.
Send plain-text invoices matching one of the two documented layouts (see FAQ) - no PDFs, scans, or images, never credential or live-system access. Every run is deterministic: same input, same output and hash, every time.
Includes: invoices.json/csv, line_items.csv, an exception report with exact reason codes, a hash-verifiable bundle, and a full check receipt.
Reference build (reality-reviewed twice, PASS): 104 tests, two layouts, five reconciliation rules exact-value tested, byte-identical reruns.
Not included: OCR, scanned/PDF/image parsing, handwriting, ERP/ledger/payment posting, tax/legal/compliance - message first if needed.
Message one sample file before ordering to confirm layout fit.
Technology:
Python
FAQ
What exactly do I send you?
A folder of plain UTF-8 text invoice files, each matching Layout A or Layout B exactly as documented (message me for the one-page layout spec before ordering). No PDFs, scans, spreadsheets, or images - the pipeline only reads the two documented text layouts.
Do you ever need my API keys, database, or live-system access?
No. This Gig has no configuration surface for credentials. Everything runs offline against only the file(s) you send.
Can this read scanned invoices, photographed receipts, or handwriting?
No. This is not OCR and does not parse PDFs or images. A document in any other format is rejected as UNSUPPORTED_LAYOUT, never guessed at on a best-effort basis.
What if my invoices don't match either documented layout?
Message me a sample first. If your export is a close variant, I can confirm fit before you order. A genuinely new layout is a separate scoped change order, not a silent best-effort parse.
What happens to a document with a math error or a missing field?
It's rejected with one specific, closed-vocabulary reason code (for example RECONCILE_FAIL_R2 or MISSING_REQUIRED_FIELD) in the exception report - never silently dropped, never a generic "failed."
Does this claim to be defect-free or independently certified?
No. The package passed two independent reality reviews with no unresolved high- or medium-severity defect. One disclosed low-severity item remains: an interrupted run can print a raw rather than clean error message. It never corrupts or loses data and is documented in delivery notes.
Will you post results into my accounting system or process payments?
No. You receive structured files (JSON/CSV) and a report. Posting to an ERP, ledger, or payment system is out of scope and never happens automatically.
Can I verify nothing changed after delivery?
Yes. Every delivered file's sha256 is listed in the evidence map and can be checked with a standard checksum tool. I also re-run your batch in a second clean environment before delivery and confirm both runs produce the same hash. That seller-side reproducibility check is included with every order.
What if my batch has zero errors?
Then the exception report says zero. Findings are derived only from your submitted documents - never manufactured to justify the order.
Is this tax, legal, or compliance certification?
No. The reconciliation checks are arithmetic - do the numbers your own documents state actually add up - not a determination of tax correctness, legal enforceability, or regulatory compliance.

