Deployment Fitting Guide

A framework for evaluating AI model deployments using four measurable traits — Quality, Speed, Context, and Size — scored against class-specific baselines and aggregated into a single deployment score.

The framework evaluates deployments (model + engine + hardware + configuration as a unit), not individual models. A speculative decoding setup with target + draft model on 4 GPUs is one deployment, scored as one unit.

Concepts

Document	Topic
four-traits.md	The 4 deployment traits: definitions, correlations, what was considered and rejected
suv-model.md	The 4×4 SUV positioning analogy for understanding trait relationships
class-baselines.md	Class-specific baselines: Gen AI (Opus) vs Decision (Jev), selection criteria
weighted-scoring.md	Scoring formula, weight rationale, interpretation bands, worked examples
sizing-guidance.md	Rules of thumb for context window and parameter sizing, hardware interaction

Log

Date	Change
2026-10-15	Added sizing-guidance.md: 8 rules of thumb for context and parameter sizing, L4 hardware data, measured deployment table
2026-10-14	Initial bundle creation from deployment evaluation session