taxonomy

Taxonomy

When to Use

• Auditing an existing Amplitude project for naming inconsistencies, duplicates, and orphaned events
• Establishing a taxonomy standard before a major product launch
• Onboarding a new analytics engineer who needs to understand the event schema
• Quarterly analytics hygiene review
• Before migrating analytics to a new SDK or platform
• When event volume is unexpectedly high (may indicate duplicate tracking)
• When reports break after a code change (may indicate event renamed/removed)

Core Jobs

1. Naming Consistency

Every product needs exactly one naming convention. Enforce it ruthlessly.

Standard convention: verb_noun in snake_case

page_viewed ✓
button_clicked ✓
form_submitted ✓

PageViewed ✗        → rename to page_viewed
buttonClick ✗       → rename to button_clicked
form-submit ✗       → rename to form_submitted
FormWasSubmitted ✗  → rename to form_submitted

Common verb vocabulary (use these, don't invent synonyms):

• Views: viewed, rendered, displayed
• Actions: clicked, tapped, submitted, applied, selected, toggled
• Lifecycle: started, completed, abandoned, cancelled, failed
• Navigation: navigated, redirected, scrolled

Rule: If two events describe the same action, they must use the same verb. button_clicked and cta_pressed are synonyms — eliminate one.

2. Property Standardization

Define a set of shared properties that appear on every event. Call these "global context properties."

Required global properties:

Property	Type	Source	Notes
user_id	string	auth system	null for anonymous users
session_id	string	SDK	auto-populated by Amplitude
platform	string	app config	web, ios, android
app_version	string	build system	1.4.2
environment	string	env var	production, staging

Prohibited property anti-patterns:

• userId — use user_id (consistent snake_case)
• ts — use timestamp (readable)
• v — use app_version (no abbreviations)
• flag — too generic; name what the flag is

Use mcp__Amplitude__get_event_properties to audit actual property names across events.

3. Deduplication

Identify events that track the same user action under different names:

Detection signals:

• Similar event names: checkout_complete vs checkout_completed vs purchase_completed
• Similar volume curves: events with identical daily trends are likely duplicates
• Same trigger code in different branches or pages

Resolution process:

1. Identify canonical name (follows naming convention, most descriptive)
2. Map old names → canonical name in code
3. Set sunset date for old event names (keep 90 days for backfill continuity)
4. Document the migration in the taxonomy changelog

4. Lifecycle Management

Every event has a lifecycle: proposed → active → deprecated → removed.

Lifecycle states:

State	Description	Action Required
proposed	Defined but not yet implemented	Implement or discard
active	Currently firing in production	Monitor volume
stale	Volume declining but still active	Review if still needed
deprecated	Replaced by new event, still firing	Set end date, update code
removed	No longer firing	Archive in taxonomy doc

Orphaned event detection (events potentially abandoned):

• Volume < 100/day AND no dashboard references → likely orphaned
• Last seen > 90 days ago → likely removed from code but still in Amplitude
• Zero properties → likely a placeholder or misconfigured event

Use mcp__Amplitude__query_amplitude_data to check event volumes and last-seen dates.

5. Documentation Standards

Every active event must have:

• Description: one sentence explaining what user action this captures
• Trigger: when exactly this fires (not just "when user clicks" — be specific about which button)
• Owner: team or person responsible for maintaining this event
• Properties: full definition with types and valid values
• Funnel membership: which funnel(s) does this event belong to?
• Created date and last modified date

Template:

event: checkout_confirmation_viewed
description: "Fires when a user successfully reaches the order confirmation screen after completing payment"
trigger: "Page component mounts after order API returns 200 with valid order_id"
owner: "growth-team"
funnel: ["purchase-funnel", "first-purchase-funnel"]
properties:
  order_id: {type: string, required: true, example: "ord_abc123"}
  order_total: {type: number, required: true, unit: "USD", example: 49.99}
status: active
created: "2024-01-15"
last_modified: "2024-11-03"

6. Governance Process

Define how new events get approved and who can add/change events:

Approval tiers:

• Tier 1 (fast track): New event that follows naming convention + has owner + has all required properties → can be added without review
• Tier 2 (review required): New event in a critical funnel, or an event that renames an existing event → requires analytics team sign-off
• Tier 3 (committee required): Removing an active event with dashboard dependencies, or changing a property on a widely-used event → requires PM + analytics + engineering sign-off

Taxonomy changelog: Every change to the taxonomy is logged with date, change type, rationale, and owner.

Taxonomy Audit Workflow

1. Pull all events from Amplitude via mcp__Amplitude__get_event_properties
2. Use mcp__Amplitude__query_amplitude_data to get volume per event for last 30 days
3. Group events by naming pattern (snake_case, camelCase, etc.)
4. Flag inconsistencies: events not following dominant convention
5. Identify low-volume events (< 100/day) as potentially orphaned
6. Identify events with identical volume curves as potential duplicates
7. Check for events with no properties (likely misconfigured)
8. Produce remediation plan with priority ordering

MCP Tools

• mcp__Amplitude__get_event_properties — pull the full event and property schema for the project
• mcp__Amplitude__get_project_context — get project metadata and configuration
• mcp__Amplitude__query_amplitude_data — check event volumes, last-seen dates, and trend data to detect orphaned or duplicate events

Key Concepts

• Taxonomy: The complete schema of all analytics events, their properties, and their relationships — the "data dictionary" for analytics
• Global context property: A property included on every event automatically — provides cross-event analysis capability
• Orphaned event: An event that appears in Amplitude but is no longer fired from production code — creates confusion and inflates event counts
• Semantic duplicate: Two events that capture the same user action under different names — the #1 taxonomy problem in growing products
• Governance tier: The level of approval required to make a taxonomy change — protects data integrity in high-volume systems

Output Format

Taxonomy audit report:

## Taxonomy Health Score: <score>/100

### Issues by Severity

CRITICAL:
  - 3 funnel events missing (cannot compute purchase conversion rate)

HIGH:
  - 7 events violate naming convention (camelCase instead of snake_case)
  - 2 duplicate event pairs: checkout_complete / checkout_completed

MEDIUM:
  - 12 events have no owner assigned
  - 5 events have no trigger documentation

LOW:
  - 8 events with volume < 100/day (possible orphans)

### Remediation Plan (priority ordered)
1. Add missing funnel events (week 1) — unblocks conversion analysis
2. Rename 7 non-convention events (week 2) — coordinate with all clients
3. Merge duplicate pairs (week 3) — 90-day overlap period
4. Assign owners to undocumented events (ongoing)

### Migration Plan
<event rename mapping table with sunset dates>