The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
Externalized Intelligence and Model Substitution: When a Correct Answer Is Not Enough for Human-Centered AI Work
ADI Research Series — Paper 2
Branden Laskowski · 2026 · Public web edition
Research question
Can authority, scope, provenance, and required abstention behavior remain external to a reasoning model when a bounded worker is handed between tested local models?
Contribution
The public research edition reports three frozen synthetic studies of governed scoped context, conventional metadata-filtered lexical retrieval, an authorized-record oracle ceiling, and bounded cross-model handoff. It distinguishes a correct answer from a complete work contract that also preserves authorization, provenance, scope, and required HOLD behavior.
Source of record
- Public research edition: https://github.com/brandenlaskowski7-bot/cortex-adi-research/tree/main/papers/paper-2
- Evidence notes: https://github.com/brandenlaskowski7-bot/cortex-adi-research/tree/main/papers/paper-2/evidence
Review boundary
This is a first-party study using synthetic data. It is not a public archival PDF/DOI release, independent replication, a production evaluation, or a general claim about model quality.
- Downloads last month
- -