grid-demo-status.md

Praxis Grid Demo Status

Demo	Current status	Size to unblock	Deferred?	Notes
Demo 1: Basic Multi-Cluster Model Routing	Enabled / Kind-proven	Done	No	Exact two-provider llm-d-style path is proven. Consumer routes model-east and model-west through provider gateways, ext_proc, mock EPP, endpoint_selector, and inference-sim.
Demo 2: API Provider Fallback	Enabled / Kind-proven, with caveat	Done for demo; S for production hardening	No	Mock API fallback and gateway credential injection are proven. Packet 1 now validates Secret-backed CRD → Kubernetes Secret → Praxis config projection in xtask. Production controller-owned credential lifecycle remains future work.
Demo 3: Full Grid — Clusters + Cloud + APIs	Enabled / Kind-proven	Done	No	One consumer gateway routes across local, remote, cloud-managed mock, and API-provider mock. Selected-backend evidence is proven by matching response model fields. Real cloud/API protocols are not claimed.
Demo 4: Metrics-Driven Load Balancing	Enabled / smaller version Kind-proven	Done for low-queue routing; M/L for saturation policy	No, but bounded	Proves normalized Prometheus-format queue-depth metrics change overlay order and request routing to the lower-queue backend. Does not prove self-hosted saturation spillover to API provider; that needs explicit policy semantics.
Demo 7: Site Join and Discovery	Enabled / lifecycle v1 Kind-proven	Done for lifecycle v1; M/L for full automation	No, but caveated	GridSite lifecycle phases, routing metadata, and cross-network isolation are proven. Automatic GridSite creation from SWIM discovery, cert exchange, and automatic phase transitions remain future work.
Demo 9: Failover Under Partition	Advanced validation / v1 Kind-proven	Done for lost-peer stale signal; M/L for full partition failover	No, but advanced/caveated	Process-kill lost-peer validation proves remote CRDT candidate flips fresh=true → fresh=false. Packet 2 is extending this toward route-away. Real network partition, recovery, stale-age policy, and full production failover semantics remain future work.
Demo 5: KV Cache Affinity	Deferred	L	Yes	Needs real llm-d / prefix-cache signal source and a design split between Grid-level provider selection and llm-d/EPP endpoint selection. Defer until metrics normalization and signal ownership are settled.
Demo 6: Budget Enforcement	Deferred	XL	Yes	Needs design first: budget state model, tenant identity, accounting source, enforcement point, policy semantics, counters, and CRD/API shape. Too large to mix into routing demos.
Demo 8: MCP Tool Federation	Deferred	L/XL	Yes	Stateless MCP routing exists in Praxis, but Grid-backed tool catalog gossip, policy filtering, distributed inventory, session semantics, and failover are a separate product surface. Defer until routing/SWIM foundation is hardened.