Should the agent roll out this model?
A deployment agent can shift traffic between two versions of a demand model. Try a rollout against a simulated latency budget before granting wider traffic.
Simulated online endpoint · 10,000 requests. The stable model misses the 100 ms deadline on 1% of requests. The canary budget is 3%.
v2 serves 500 requests. Its late rate is 2%. Across both versions, 1.05% are late.
The latency gate passes. Check forecast quality before approving the next traffic increase.