Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2607.11948.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

The paper combines an underpowered proof-of-mechanism test of ontology-amplified distillation, training a Qwen3.6-27B student on a single Apple M5 Max, which matched but did not exceed GPT-5 on 40 Vietnamese financial tasks, with a contextuality-audit method that found zero residual contextuality, indicating direct influence and construct coupling as the useful signals.

Reports honest negative results on contextuality-based routing and underpowered distillation, useful for anyone sizing sovereign LLM pilots or designing governance diagnostics for enterprise agent systems.

Cited by

No citations on record.