SparkInfer / genie-ai-runtime narratives on the site describe engineering direction—not automatic customer-accepted results. Prefer an acceptance-first conversation over multiplier shopping.
Typical path: review #/mrf or #/exclusive → submit trial interest → commercial human review → guidance / tech pack → retest on agreed hardware and baselines → then PoC milestones and commercialization talks. Human review clarifies scenario, stack, and delivery-party boundaries so open-source ecosystem references are not mistaken for the contracting counterparty.
Write the retest brief early: same GPU / same quant weights, baseline (e.g. a generic stack), separate decode/prefill KPIs, concurrency and stability protocol. Public numbers are illustrative · subject to retest; acceptance is reproduction in the customer environment. Start from #/mrf (trial) or #/exclusive.
Method note only—no promise of trial slots, timelines, or performance outcomes.