Capabilities · Recommendations
Product Recommendations
HealthyIn-context predictionEndpoint product-recommendationsProduction v12Challenger v13
Product Recommendations / Deployments
Simulated
BLUE— Production
Healthyv12
- Traffic
- 100%
- NDCG@10
- .412
- P95
- 58ms
- Cost / 1K
- $0.22
- Engine
- Managed A
- Created
- 2 weeks ago
GREEN— Candidate
Readyv13
- Traffic
- 0%
- NDCG@10
- .421
- P95
- 61ms
- Cost / 1K
- $0.23
- Engine
- Managed A · session events added to context
- Created
- 4 days ago
Live traffic
BLUE v12100%
GREEN v130%
Deploy v13
Choose how the challenger reaches production
Production
v12
Challenger
v13
Customer responses still come from v12. v13 predictions are evaluated silently on 10% of mirrored requests.
No customer impact. Shadow traffic is billed at challenger cost.
Shadow results — 48 hoursRequests evaluated 1.5M
Simulated| Metric | v12 | v13 |
|---|---|---|
| NDCG@10 (replay) | .412 | .421 |
| Recall@20 | .318 | .327 |
| Catalog coverage | 71% | 73% |
| P95 latency | 58ms | 61ms |
| Cost / 1K | $0.22 | $0.23 |
✓P95 latency 61ms < 100ms
✓Availability 99.995% > 99.95%
✓Catalog coverage 73% > 60%
✓Cost / 1K $0.23 < $0.50
v13 satisfies all configured production constraints.
Latency and cost are measured on live shadow traffic. Quality rows come from replaying both versions on predictions with matured outcomes (Sep 17 – Sep 23, 94% coverage) — outcomes for this shadow window arrive in ~7 days.
Version timeline
Today
- 07:30Experiment v12 vs v13 — day 4 of 14, 20% challenger traffic
4 days ago
- 13:05v13 created — session events added to context
Last week
- 09:12Managed B rejected — P95 112ms violates constraint
2 weeks ago
- 15:40v12 promoted to production