Horizon installation page describes planned capacity, not achieved research outcomes

Engage research-computing directors, principal investigators, finance and grant administrators around jobs constrained by existing resources. Ask when an allocation is needed, whether software can use the architecture, and which downstream laboratory or data bottlenecks remain. Offer a bounded workload-readiness assessment using one reproducible pipeline. The value hypothesis is meeting a research deadline with appropriate shared capacity. Do not convert system specifications into promised application speedups or imply that production access is already available. Compare access to a national facility with local and rented capacity using actual eligibility and support requirements.

Inventory compiler, container, numerical-library and accelerator dependencies, then prepare a portable test workload with known outputs. Map campus data staging, identity and artifact return paths. Obtain current allocation and platform documentation before estimating integration effort. Test precision sensitivity, I/O and restart behavior in an approved environment, with least-privilege data access. The proof of value should compare end-to-end turnaround and scientific equivalence against the existing workflow. Hardware peak figures alone cannot validate an application migration; isolate any external model endpoints from the HPC execution boundary.

Assign the research-computing service owner to coordinate platform staff, scientific users and allocation administrators. Prepare onboarding instructions, reproducible job recipes and a staged migration checklist. Dependencies include confirmed service availability, usable software, data-transfer capacity and researcher training. Proposed acceptance criteria are reproducible output within agreed scientific tolerance, successful checkpoint recovery and documented support escalation for a representative job. Measure queue and transfer time alongside compute time. Preserve a fallback execution path until acceptance; risks include schedule changes, portability work and research deadlines outpacing available allocations.