Model Fit Deck

Every time a new model drops, you lose hours scanning leaderboards, quantization threads, and Reddit benchmarks—then still ship a guess into production. Model Fit Deck ends that loop. You paste in your GPU profile, your workload constraints, and the candidate models you are considering. The skill runs a structured fit check against your actual VRAM ceiling, latency floor, and context requirements and returns a ranked decision packet: one primary recommendation, one fallback, and explicit rejection reasons for every model that does not fit. No benchmark averaging across unrelated hardware. No generic 'it depends' output. A concrete go or no-go you can put in a deployment runbook or share with your team in under 20 minutes. Built for engineers who are tired of repeating the same manual research loop every release cycle.