[lucarlig]

Adds FYRE OpenShift comparisons with 40 GB master/worker roots and no worker data disks. The fully parallel profile runs built-in and external dataplanes at 125, 250, 500, and 1,000 users, with independent target, Locust, and Fast Time pods for each measurement.

Both lanes use the same MCP 2026-07-28 client, server/discover, per-request client metadata, and balanced calls across the same six zero-delay Fast Time tools. Target stacks have 2 vCPU / 2 GiB limits and 1.5 vCPU / 1.25 GiB scheduling requests. Each helper pod has 1.5 vCPU / 0.75 GiB. The three worker roles remain separate, and helper pressure invalidates a comparison when load generation or the backend becomes the bottleneck.

Pods use required node affinity so Kubernetes checks resource availability. Before deployment, the campaign checks actual allocatable CPU/memory, existing system pods, and target registration overhead. A live report container shares Locust's writable report volume and preserves access to artifacts after the master exits. Reports distinguish target requests from limits.

The CLI creates, authenticates, runs, collects artifacts, and deletes only its run-owned cluster. Existing cluster names are refused before a new ownership record is written. Interrupted campaigns can resume from saved state. Outputs include measured-window memory, helper validation, CSV/JSON, a Slack-ready PNG, and a Markdown report. The parallel profile requests 740 GB of disk, including FYRE's fixed 500 GB API VM.

Validation:

• cargo fmt --all --check
• cargo test --locked --all-targets: 437 passed
• cargo clippy --locked --all-targets --all-features -- -D warnings
• Python FYRE tests: 34 passed
• cargo package --locked --allow-dirty: passed
• Live FYRE cluster accepted the revised profile and provisioned seven VMs with the expected CPU, memory, and disk allocations. Full campaign ocp-2v2-check-1001 is still provisioning; end-to-end benchmark and cleanup validation remain pending.