Build a compatibility system customers can trust

Represent compatibility as relationships

Treat equipment and replacement products as separate entities. Record an explicit relationship between a particular model or variant and a particular compatible part or variant. Support aliases, legacy model numbers, manufacturer part numbers, regional versions and effective dates when they matter.

Compatibility is frequently many-to-many: one filter fits several machines; one machine accepts several filters. Flat tags can help discovery, but they become fragile when model families change, substitutes are introduced or one part has an important exclusion.

Choose evidence standards

• Prefer manufacturer fitment documentation, official part catalogs and verifiable product specifications.
• Record where each relationship came from, when it was checked and who approved it.
• Mark unknown or conflicting relationships for human review; do not silently publish a probabilistic match as fact.
• Capture exclusions explicitly, such as 'fits model A except revisions manufactured before 2021'.
• Test the model selector with real aliases, spacing, punctuation and common user mistakes.

A minimum viable compatibility record

For a starting implementation, record: equipment brand, model identifier, model aliases, equipment variant if needed, compatible item SKU, relationship type, source URL or supplier reference, last verification date, publication status and explanatory fitment note.

Keep draft relationships distinct from published ones. An exception queue is less exciting than an 'AI matched everything' claim, but it is more useful when the cost of a wrong recommendation is a returned item, damaged equipment or lost customer trust.

Measure fitment quality

Track a reviewed sample of published relationships, an incorrect-fitment report rate and time to correct an error. Do not use the number of generated mappings as the quality metric: volume alone can hide systematic mistakes.