nx_render_target_match.nx

nx_render_target_match.nx -- did the generated image match the target? Universal RENDERED-VS-TARGET verdict primitive. Given two 16-element feature vectors (target image + rendered image), emits a sealed-enum verdict + per-axis divergence + actionable next-axis pinpoint. USE CASES (all the same substrate, different target sources): 1. RETAIL VIRTUAL TRY-ON (Google Shopping AI Mode pattern): target = product image (lingerie / dress / jacket / shoes) render = model wearing the product (via gen-img) verdict = did the rendered model faithfully wear the product? action = which visual axis (color / texture / silhouette / etc.) needs another iteration? 2. VIDEO-GAME ASSET RENDER: target = concept art reference for a weapon / armor / character render = in-engine model the procgen produced verdict = does the asset match the brief? 3. STORY-DRIVEN SCENE RENDER (legal adult or all-ages): target = reference image for the desired scene mood render = generated scene image verdict = does the render hit the mood? All three use the SAME primitive. Different in the prompt + concept- embedding upstream; identical at the measure-the-output layer. COMPOSES: nx_cosine_similarity -- overall vector alignment per-axis delta math -- normalized |target - rendered| per axis sealed verdict bands -- substrate-honest routing worst-axis pinpoint -- which axis to refactor next render genealogy_id: salton_1971_smart + clip_score_hessel_2021 + ip_adapter_ye_2023 + lpips_zhang_2018 + fid_heusel_2017 lineage_id: render_target_match_q10