What it is
Personal language model forked from AXON, trained on 5GB of Notion exports. Currently exhibits ~2x train/val loss gap suggesting overfitting.
Open architecture decisions: add a RAG layer, audit chunking strategy, or repurpose as an embedding model rather than a generator. The first instrument that points back at the operator: a model trained on the documented self.
Where it sits
ARCHIE is catalogued under Cognitive Systems in the instrument registry. Instruments in this group share a concern and are frequently composed rather than used alone; the related entries below are the nearest ones.
This is a registry entry: a real instrument, documented here, without a public endpoint a stranger can open. Pages carrying source excerpts and measured results are marked as such in the catalog. If you want detail on this one specifically, ask about it directly.