Real Data Pipeline Verification

Overview

This document describes how to verify the complete GlycemicGPT data pipeline with real Dexcom G7 and Tandem t:connect data. The verification script exercises every stage from credential configuration through data ingestion, dashboard display, and AI analysis.

Prerequisites

1. Backend running with database and Redis (Docker Compose or local dev)
2. Registered user with accepted safety disclaimer
3. Dexcom G7 credentials (Share username and password)
4. AI API key (Anthropic or OpenAI) for AI analysis verification
5. (Optional) Tandem t:connect credentials for pump data verification

Quick Start

# 1. Start backend services
docker compose up -d db redis
cd apps/api && uv run uvicorn src.main:app --host 0.0.0.0 --port 8000 &

# 2. Start frontend
cd apps/web && npm run dev &

# 3. Configure integrations via the web UI:
#    - Navigate to /dashboard/settings/integrations
#    - Enter Dexcom G7 credentials and test connection
#    - (Optional) Enter Tandem credentials
#    - Navigate to /dashboard/settings/ai-provider
#    - Configure Claude or OpenAI with API key

# 4. Run the automated verification
./scripts/verify-data-pipeline.sh --email [email protected] --password yourpassword

What Gets Verified

Stage 1: Authentication

• Login with real user credentials
• Verify user role and session

Stage 2: Integration Status

• Dexcom G7 connection status (connected/error/unconfigured)
• Tandem t:connect connection status

Stage 3: Data Sync

• Trigger manual Dexcom sync and verify readings are fetched
• Trigger manual Tandem sync and verify pump events are fetched
• Report counts: readings fetched vs. stored (deduplication)

Stage 4: Dashboard Data

• Current glucose: Value, trend direction, freshness (minutes ago)
• Glucose history: Number of readings in last 3 hours
• Staleness check: Warns if reading is > 10 minutes old
• Target range: Confirms low/high threshold configuration

Stage 5: Insulin on Board (Tandem only)

• IoB projection from pump data
• Staleness check on IoB data

Stage 6: Control-IQ Activity (Tandem only)

• 24-hour activity summary
• Automated vs. manual event counts

Stage 7: Real-Time SSE Stream

• Connect to SSE endpoint and verify data events arrive
• Parse glucose value from first event

Stage 8: AI Provider

• Confirm AI provider is configured (Claude or OpenAI)

Stage 9: AI Chat

• Send a test message asking about glucose status
• Verify AI responds with contextual analysis

Stage 10: Daily Brief Generation

• Generate a 24-hour daily brief with AI
• Verify time-in-range, average glucose, and AI summary

Stage 11: Insights Feed

• Verify insights (briefs + analyses) appear in the feed
• Report unread count

Stage 12: Alert System

• Verify alert thresholds are configured
• Check for active alerts

Data Flow Diagram

Dexcom G7 Cloud                 Tandem t:connect Cloud
     |                                |
     v                                v
  POST /api/integrations/dexcom/sync   POST /api/integrations/tandem/sync
     |                                      |
     v                                      v
  glucose_readings table              pump_events table
     |                                      |
     +------------- merged --------------->+
                      |
                      v
         GET /api/integrations/glucose/current   (Dashboard: GlucoseHero)
         GET /api/integrations/glucose/history   (Dashboard: TrendArrow, TimeInRange)
         GET /api/v1/glucose/stream              (Dashboard: Real-time SSE)
                      |
                      v
         POST /api/ai/briefs/generate            (AI: Daily Brief)
         POST /api/ai/chat                       (AI: Chat with context)
         POST /api/ai/meals/analyze              (AI: Meal patterns)
         POST /api/ai/corrections/analyze        (AI: Correction factors)
                      |
                      v
         GET /api/ai/insights                    (Insights Feed)

Manual Verification Checklist

After running the automated script, verify visually via the web UI:

Dashboard (/dashboard)

• GlucoseHero card shows current glucose value and trend arrow
• Value updates in real-time via SSE (watch for ~60 seconds)
• TimeInRangeBar shows percentage based on real data
• FreshnessIndicator shows "X minutes ago"
• No "Live updates paused" banner (SSE connected)

Daily Briefs (/dashboard/briefs)

• AI Insights page shows generated briefs
• Clicking a brief shows full AI analysis
• Unread badge appears on sidebar "Daily Briefs" link
• Filter tabs work (All Insights / Daily Briefs)

AI Chat (/dashboard/ai-chat)

• Chat interface loads with input field
• Sending "How am I doing today?" returns contextual response
• AI references actual glucose data in its analysis
• "Not medical advice" disclaimer visible

Alerts (/dashboard/alerts)

• Alert feed shows any triggered alerts
• Alerts reflect real glucose threshold crossings

Settings (/dashboard/settings)

• Integrations page shows Dexcom as "Connected"
• AI Provider page shows configured provider
• Glucose Range settings persist after save

Troubleshooting

Dexcom sync returns 0 readings

• Dexcom Share may have a delay (5-15 minutes after CGM reading)
• Verify credentials work on the Dexcom Clarity app
• Check the Dexcom account has Share enabled

AI chat returns empty response

• Verify API key is valid (test in Settings > AI Provider)
• Check backend logs for AI provider errors
• Ensure there is glucose data available for context

SSE stream shows no data events

• SSE sends heartbeats every 30 seconds (verify those arrive first)
• If only heartbeats: sync Dexcom first, then reconnect SSE
• Browser DevTools > Network > EventSource shows live events

Dashboard shows stale data

• The scheduled sync runs every 5 minutes by default
• Trigger manual sync: curl -X POST http://localhost:8000/api/integrations/dexcom/sync
• Check backend logs for sync errors

Time-in-range shows 0%

• Need at least a few readings to calculate meaningful TIR
• Default range is 70-180 mg/dL (configurable in Settings > Glucose Range)