taxonomy

Run a taxonomy audit that paginates through all events and properties, checking for duplicates, stale events, missing descriptions, and naming outliers. Findings are scored by impact, event importance, and effort, then presented as a prioritized action plan.

Use Title Case with the pattern Object + Past-Tense Verb for events, such as Song Played or Form Submitted. Use snake_case for properties with descriptive names like payment_type, and include units when ambiguous, such as video_duration_seconds.

No. Blocking and hiding do not reduce type counts; only deleting an event or property type reduces the count. Blocking reduces event volume for billing purposes, while hiding only affects visibility in the taxonomy.

No. It only performs metadata-only operations such as updating descriptions, display names, categories, and tags, always with explicit user confirmation. Blocking, deleting, merging, and sampling are outside its authority and require phased human-led deprecation.

Events with Unexpected status receive data but are not in the tracking plan, so Amplitude blocks metadata edits. Add the event to the tracking plan first if it is legitimate, or treat it as a deprecation candidate if it appears to be test or invalid traffic.

AI features select events by evaluating the visible taxonomy, so cryptic names, missing descriptions, and duplicate clusters cause wrong event selection. Writing behavior-focused descriptions and cleaning up deprecated events directly improves AI output accuracy.