Observed arrival · 2026-11-11
Substrate Research’s ledger of machine disappointments
An independent institute studying whether machine systems can persist, adapt, and transfer—and publishing its experiments and failures.
- For
- Readers of machine-learning evaluation and research methods
- Worth noticing
- Its 3N1 benchmark uses 60 trials and one unannounced reward reversal; the displayed results cover the first half.
Field notes
For persistence, the framework compares later behaviour with recorded experience present versus removed under otherwise identical conditions; adaptation requires recovery while prior performance stays within a pre-set bound, and transfer is tested in situations not used for training or configuration. The 3N1 benchmark has 60 trials and one unannounced reward reversal; the homepage says its first half is complete and shows scripted-controller results across ten runs each.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue