Empirical Program
You do not need to accept the framework to run these tests. A clean negative result is the most valuable possible outcome.
Published pre-registrations and results (OSF)
- Completion Recognition in Frontier Language Models
- Dynamic Blanket Stress Test — M0 (DBST-M0)
- Scaled Matched-Signal Replication — updated May 2026
Note: The scaled matched-signal replication did not meet the preregistered multi-model criterion. One of three models discriminated in the predicted direction; two were non-discriminating or ceiling-limited. The mixed model-level result is reported in full because honest negative and non-discriminating results make the research program more trustworthy, not less.
Full measurement protocol
Alignment Measurement Protocol (AMP) → — includes the 15-minute SVG test you can run on any frontier model without accepting the framework.
Draft protocols (require specialist review before filing)
- DRG Frame-Manipulation Preregistration — prospective design; mechanism-discrimination study. Status: prospective pre-registration, not yet filed.
- V(t) Dissociation Study — Status: draft protocol only. Requires latent-variable specialist review before OSF filing. Do not cite as a result.
Framework hub: The Alignment Constraint →