Bernt
bae705aa97
- Add NFC ePassport roadmap (ICAO 9303, eIDAS) - Add TensorFlow.js edge face detection (BlazeFace) - Add structured audit logger (GDPR-compliant) - Risk scoring support Part of KYC Apple Native UX v1.1.0
5.9 KiB
UIOS v1.1 – Real World Validation
Version: 1.1.0
Date: 2026-06-28
Status: Specification
Previous: UIOS v1.0 (Architecture)
Overview
UIOS v1.1 shifts focus from architecture to proving the system works with real data. No new core components. Only validation, measurement, and hardening.
"Architecture without validation is just a beautiful diagram."
Five Goals
1. Production Capture iPhone App
Current state: Swift prototype exists (ios/QuixZoomCapture/)
v1.1 requirements:
| Feature | v1.0 | v1.1 |
|---|---|---|
| Manual capture | ✅ | ✅ |
| Auto-upload | ❌ | ✅ |
| Background sync | ❌ | ✅ |
| Real-time GPS | ❌ | ✅ (±3m) |
| Sensor data | ❌ | ✅ (gyro, accelerometer, compass) |
| AI guidance | ❌ | ✅ ("Move closer", "Rotate 30°") |
| Offline mode | ❌ | ✅ (queue locally, sync when online) |
| Battery optimization | ❌ | ✅ |
Key metric: 95% of captures upload within 10 seconds on 4G.
Implementation:
- Background URLSession for uploads
- CoreLocation with kCLLocationAccuracyBest
- CoreMotion for sensor fusion
- Reachability monitoring for sync
- Local SQLite queue for offline
2. Bangkok Field Test
Current state: 6 videos processed, 45 observations, synthetic data dominates
v1.1 requirements:
| Metric | Target | Current |
|---|---|---|
| Verified observations | 1,000+ | 45 |
| Object types covered | 6+ | 3 |
| GPS accuracy < 5m | 90% | Unknown |
| Image quality > 0.8 | 95% | Unknown |
| Multi-zoomer coverage | 3+ people | 1 |
| Time-of-day coverage | Day/evening/night | Day only |
Test protocol:
- Recruit 3 Zoomers in Bangkok
- Assign missions from Mission Planner
- Capture with production app
- Upload to S3 → Pipeline → OIE → UKG
- Human review of all observations
- Measure: precision, recall, merge rate, latency
Success criteria:
- OIE precision > 95% on real data
- OIE recall > 90% on real data
- Average 3+ observations per object
- Coverage score > 60% for all categories
3. End-to-End Pipeline Validation
Current state: Components work individually, limited integration testing
v1.1 requirements:
Test the complete flow:
Capture → Upload → AI Detection → OIE → UKG → Change Detection →
Coverage Update → Mission Generation → Zoomer Notification
Validation scenarios:
| Scenario | Expected Result |
|---|---|
| New street lamp observed | Object created in UKG |
| Same lamp observed twice | Observations merged |
| Lamp with rust vs. repaired | Change detected, health updated |
| Coverage gap identified | Mission generated |
| Mission completed | Coverage score updated |
Performance targets:
| Stage | Target Latency |
|---|---|
| Upload to S3 | < 5s |
| AI detection | < 2s |
| OIE processing | < 100ms |
| UKG update | < 50ms |
| Mission generation | < 1s |
| Total: Capture → Mission | < 10s |
4. OIE Performance Measurement
Current state: Benchmarks on synthetic data show 100% precision/recall
v1.1 requirements: Measure on real data with known ground truth.
Method:
- Select 50 objects in Bangkok (known locations)
- Have 3 Zoomers capture each object 3 times (9 observations/object)
- Run OIE on all 450 observations
- Compare OIE output to known ground truth
Metrics:
| Metric | Target | Measurement |
|---|---|---|
| Precision | > 95% | True merges / All merges |
| Recall | > 90% | Found objects / Known objects |
| False merge rate | < 2% | Wrong merges / All merges |
| Missed merge rate | < 5% | Missed merges / Should merge |
| Latency (P99) | < 50ms | Time per observation |
Tuning:
- Adjust weights based on real-world performance
- Test different configurations
- Document optimal settings per city type
5. Customer Demo Data
Current state: Synthetic data, no customer-facing outputs
v1.1 requirements: Real data that demonstrates value.
Deliverables:
| Deliverable | Content |
|---|---|
| Bangkok City Report | Coverage map, object inventory, health scores |
| Change Detection Report | Before/after comparisons, alerts |
| Mission Effectiveness | Missions completed, coverage improved |
| API Demo | Live endpoints with real data |
Customer pitch:
"We deployed 3 Zoomers in Bangkok for 2 weeks. They captured 1,000+ observations covering 300+ objects. Our AI identified 15 changes, including 3 safety issues. Here's the data."
Timeline
| Week | Focus | Deliverable |
|---|---|---|
| 1 | iOS app hardening | Production-ready app |
| 2 | Bangkok recruitment | 3 Zoomers onboarded |
| 3-4 | Field test execution | 1,000+ observations |
| 5 | Pipeline validation | End-to-end test passed |
| 6 | OIE measurement | Benchmark report |
| 7 | Customer demo prep | Presentation + data |
| 8 | Review & plan v1.2 | UIOS v1.1 retrospective |
Success Criteria
UIOS v1.1 is successful when:
- ✅ iOS app captures and uploads with < 10s latency
- ✅ 1,000+ verified observations from Bangkok
- ✅ End-to-end pipeline processes without manual intervention
- ✅ OIE precision > 95%, recall > 90% on real data
- ✅ Customer demo shows clear value proposition
Risks & Mitigation
| Risk | Impact | Mitigation |
|---|---|---|
| Zoomer recruitment fails | High | Start early, offer competitive pay |
| iOS app crashes in field | High | Extensive testing, offline fallback |
| OIE performs poorly on real data | High | Manual review, iterative tuning |
| Network issues in Bangkok | Medium | Offline mode, batch sync |
| Data quality poor | Medium | Quality control, re-capture missions |
v1.2 Preview
If v1.1 succeeds:
- v1.2: Multi-city deployment (Torrevieja, Stockholm)
- v1.3: Customer API launch
- v1.5: Model marketplace
- v2.0: Autonomous mission planning
End of v1.1 Specification