FRONTIER DESK
Research
Not a write button. A research loop: search the literature, type every claim, write a discovery, propose integrations. The model does not silently retrain. Observer memory learns only through the gate.
PROGRAM
Live search when available. One user-started run. No background loops.
LEARNING LEDGER
Proposed findings enter unverified observer memory. Integrate to inject them into the next MIO run. Hold or reject to stop. This is continuous learning as an inspectable trail — not weight updates.
No proposed integrations yet.
LIBRARY
UNKNOWN REGISTER
Unknown is a state, carried across discoveries and runs. Program unknowns stay until the experiment closes them. Harvested unknowns can be resolved.
The register is empty. That is itself a claim to inspect.
CITATION GRAPH
Metzinger, the MPE Project, Lopez-Sola 2026, Mago 2026 — linked to claims, not dumped in a bibliography.
Measure the instrument on MIO-Bench: ladder, ablation, cross-model invariance.