Amazon Supplier Offer Variance Audit

Paste two dated supplier offer spreadsheets and get one audit row per exact ASIN and seller. The Actor compares landed offer totals (unit price + shipping + fee), and flags cost outliers, expiring or expired quotes, and incomplete data.

It works only on the data you supply. It does not fetch Amazon pages, call any API, use proxies, or collect personal data.

What it checks

• Total-cost change: baseline vs. comparison total for the same ASIN and seller; flagged at or above your threshold (default 10%), with price, shipping and fee deltas.
• Peer outliers: an offer far above or below the median of the other sellers for the same ASIN and currency (needs at least 3 offers by default).
• Quote validity: quote_expired and quote_expiring_soon relative to a reference date; quote_dates_not_ordered when the newer sheet is not actually newer.
• Completeness: missing or invalid price, shipping, fee, currency, quote date or valid-until date, listed per side.
• Coverage: offers present only in one sheet, duplicate rows (newest quote kept), currency mismatches, and invalid ASIN or seller keys.

Input

Provide baselineCsv and comparisonCsv (or JSON arrays in baselineRows and comparisonRows). Columns (aliases such as supplier, vendor, price, expires are accepted):

asin, seller, unit_price, shipping, fee, currency, quote_date, valid_until

Dates use YYYY-MM-DD. Enter 0 for a genuinely zero shipping or fee; an empty cell is treated as missing. The default sample input can be run as is.

Output

Dataset rows have asin, seller, status, severity (high, medium, low, ok), flags, totals, deltas, peer median, expiry days, and the source row numbers. A SUMMARY record in the key-value store holds counts per severity and flag plus caveats.

Pricing and errors

The code emits one offer-key-audited event per audited ASIN/seller key after results are saved. Invalid input (missing sheet, bad reference date, too many rows) fails the run without any charge, and invalid-key rows are not counted. Pricing is not configured in this repository.

Limitations

• No currency conversion: offers in different currencies are flagged, not compared.
• Totals are per unit as entered; taxes, MOQ tiers, and volume breaks are not modeled.
• Outlier thresholds are simple percentages, not statistical tests. Peer medians from 3 or 4 sellers are weak evidence; treat flags as prompts to check, not conclusions.
• Sheets are limited to 5,000 rows each to keep memory low.
• Seller names match after trimming spaces and ignoring case; different spellings of the same seller are treated as different sellers.
• Results are only as accurate as the spreadsheets supplied.