Quick question before I write this up as a feature request — is there a reason the resource controls are limits-only? I see serviceInstanceLimitsUpdate and limitOverride in railway.json for setting a ceiling on vCPU/memory, but I don't see anything for setting a floor (a guaranteed minimum / reservation per replica).

For latency-sensitive workloads (in my case real-time voice agents on FastAPI + LiveKit) the autoscaler ramp on the first burst after idle is the thing I'd most want to control. A floor would solve it cleanly.

Also — I'd happily pay more for it. A reserved floor is more expensive to provision than burstable headroom, so this seems like a feature you could bill at a premium and I'd still take it. Predictable latency on real-time voice is worth real money to me.

Is this something the platform deliberately doesn't expose, or just hasn't been built yet? If the latter, happy to file it properly.