Analytics

We use PostHog for analytics. Focus on emotional engagement, not traditional game metrics.

Current Setup

• Client: /web/src/lib/posthog.js
• Server: /web/src/lib/server/posthog.js
• LLM calls auto-tracked via ask()
• Only manual event: level_evaluation

Adding Events

Use category_object_action snake_case format:

import {client as posthog} from '$lib/posthog'

posthog.capture('scenario_level_complete', {
  level_number: levelNumber,
  emotional_resonance_score: evaluation.score
})

Follow PostHog conventions: lowercase, present tense verbs, snake_case.

Events to Track

Game Flow:

• intro_new_game - User starts new game
• intro_continue_game - User continues existing game
• prechat_complete - User completes prechat interaction
• call_accept - User accepts/connects to call
• call_complete - User finishes call

Scenario Progression:

• scenario_start - First time entering scenario
• scenario_level_start - Enter level within scenario
• scenario_level_complete - Complete level within scenario

Engagement:

• engagement_long_pause - User hesitates >30s
• connection_emotional_moment - Breakthrough moments

What NOT to Track

• Every message (LLM pipeline covers this)
• UI clicks
• Debug logs from console.log

Track meaningful impact, not noise.