Cognition-core now includes a real native functioning model path.
adapters/exemplar_adapter.pyscripts/training/train_exemplar_candidate.pyscripts/execution/run_functioning_candidate.pytests/test_functioning_model.pydocs/FUNCTIONING_MODEL_STATUS_2026-04-14.md
The new exemplar adapter loads a deterministic local artifact built from curated JSONL corpora already in the repo and the supporting package ingestions.
Candidate: darpa_functional
- artifact built successfully
- benchmark run completed successfully
- scored output file generated successfully
- current overall weighted score:
0.3458 - full candidate gate still passes for the scaffold lane
This is a real runnable model path and closes the biggest honesty gap in the repo. It is not yet the final GGUF-trained candidate path. The remaining hard blockers are still the true external trainer / merge / GGUF exporter loop and retention validation against a produced quantized artifact.