Planning, routing, and robust eval

Unlike end-to-end audio models, it explicitly plans readable scores before full-song rendering.

A 400M model trained only on synthetic SCM tables beats tuned AutoGluon zero-shot.

Unlike routed sparse attention, it guarantees full causal coverage collectively across heads, preserving 8K recall while cutting 128K train latency 7.4×/8.6× and decode 3.0×.

Unlike single-pass decision models, it adds reasoning to a Jev-compatible 9B classifier, raising JevBench from 0.866 to 0.935 while keeping calibrated outputs and optional 0.3-second no-think inference.

Unlike static benchmarks, it append-only tracks post-launch model drift daily with pinned prompts and pre-registered statistics.

Unlike typed-decision classifiers, it scores state-action pairs for routing, ranking, and verifiers at up to 9× lower latency than Jev.

It pairs 74.0M field-instance masks with a 100-country zero-shot split for cross-country generalization.

Unlike navigation datasets, it pairs 48,264 egocentric steps with verified first-person rationales and closed-loop state-based evaluation in seven unseen environments.

Unlike typical radiology datasets, it links 11K PET/CT lesion localizations directly to aligned free-text findings.