The current Dynamic UNL V10 system and the Qwen 3.8 proposal are not two versions of the same scoring prompt. They are different decision architectures.
- V10 is an LLM-assisted multi-factor scoring system. The model assigns five sub-scores; deterministic code combines them into an authoritative final score and applies a cutoff, size cap, and incumbent-churn rule.
- Qwen proposes a deterministic evidence-qualification system. It rejects scalar validator-quality scores, ranks eligible candidates primarily by 30-day agreement, and optionally constructs a set under concentration and continuity constraints.
Qwen is stronger about epistemic boundaries: it cleanly separates candidate observations from set-level safety, refuses to treat endpoint metadata as proof of independent control, and makes missing threat-model parameters explicit. But it is not a safe drop-in replacement for V10. Its default ranking gives r