Catalog SKU Matcher India

Purpose

Build reliable cross-store product matching for Indian catalogs so price comparison is accurate.

Disclaimer

This skill provides matching and normalization guidance only. It does not guarantee perfect match accuracy for all catalogs or seller data quality.

Use at your own risk. The skill author/publisher/developer is not liable for direct or indirect loss, incorrect match decisions, trading losses, or other damages arising from use or misuse of this guidance.

Matching strategy

Use a layered approach:

1. Hard identifiers

   • model number / GTIN / MPN / ISBN where available

2. Variant normalization

   • brand
   • model family
   • storage/RAM
   • color
   • size/pack quantity
   • condition (new/refurbished/used)

3. Soft similarity

   • token similarity on cleaned title
   • key-attribute overlap
   • seller metadata sanity checks

4. Confidence score

   • high: auto-match
   • medium: human review queue
   • low: reject

False-match guardrails

• Never match different storage/RAM variants as same SKU.
• Never match bundles/accessories to standalone products.
• Never ignore refurbished/used condition differences.
• Require manual review when two or more variant fields are missing.

Output format

When matching listings, return:

1. canonical SKU candidate
2. matched listings with confidence level
3. rejected candidates with reason codes
4. manual review queue entries

Setup

Read setup.md and define normalization dictionaries first.

Validation

Run validation-checklist.md on labeled test sets before production.

References

• Rules and reason codes: matching-rules.md
• Confidence scoring: scoring-guide.md
• Edge-case examples: examples.md