Merchant Feed Quality Auditor

What it does

Check a customer-supplied product CSV or JSON before a Merchant Center upload. The auditor identifies missing base fields, duplicate offer IDs, malformed prices, currency conflicts, sale prices that are not lower, invalid GTIN lengths or checksums, invalid URL syntax and unsupported availability or condition values. It returns a row-level correction list and a readable summary. This is a limited offline validation pilot for merchants and agencies.

Input

Supply exactly one of rows or csv. Use Merchant Center field names: id, title, description, link, image_link, availability and price. Optional fields include sale_price, gtin, brand, mpn, identifier_exists, condition, item_group_id, color and size. CSV accepts a BOM, quoted multiline values and an explicitly selected comma, semicolon or tab delimiter. The g: prefix is accepted. Duplicate normalized column names fail rather than silently replacing data.

IDs, GTINs and MPNs must be strings, preserving leading zeros. Prices use a decimal point, a space and an uppercase currency code, for example 129.90 PLN. Up to six decimal places are compared with exact integer arithmetic. expectedCurrency optionally checks price currency consistency. No automatic currency conversion occurs.

Output

One dataset report contains counts, every finding, product row statuses, corrective actions and explicit limitations. AUDIT is the complete JSON report, FINDINGS_CSV is a downloadable row-level checklist and REPORT is Markdown. CSV cells with spreadsheet formula prefixes are neutralized. No product values are changed automatically.

Pricing and cost control

Proposed PAY_PER_EVENT price: USD 0.03 per completed feed-report event, with up to 5000 products and 3 MB total input. maxRows can lower the limit. Overflow is rejected without truncation. Exactly one report is emitted after exports are saved. No AI, paid APIs, proxy, ad-account access or network fetches are required. Cloud pricing and actual cost remain unverified until the guarded deployment and run.

Limitations and fair use

These checks do not guarantee Google approval. Category-specific fields, country policies, tax, shipping, image content, URL availability, landing page prices and GS1 identifier assignment are not verified. Some products have valid identifier exemptions. Missing identifiers, shared GTINs and incomplete variant attributes therefore produce review warnings rather than universal rejection claims. Use Google Merchant Center diagnostics for the final decision.

Input is saved in the run storage according to the platform retention settings; do not upload credentials or private customer records. No third-party AI processor receives the feed. Technical errors fail the run, and storage failures prevent the report billing emission. A valid feed with zero findings still produces one report.

Example

examples/input.json includes a valid offer, an offer with deliberately broken fields and a custom product exemption. examples/input-csv.json demonstrates the CSV path. Run npm ci --ignore-scripts, npm test, then use the existing factory release.mjs --local-run --write-example. The canonical production source stays in this directory; the Apifi boty folder is only a view.