New reference implementation: Misalignment evaluations#108

Open

ethancjackson wants to merge 59 commits into

VectorInstitute:mainfrom

ethancjackson:ethan-dev

Commits on Mar 18, 2026

Add configurable misalignment QA experiment shell.
ethancjackson
committed

Commits on Mar 19, 2026

Commits on Mar 23, 2026

Replace misalignment QA CLI reporting with a notebook explorer.
ethancjackson
committed
Simplify misalignment_qa: refactor results_notebook, improve notebook, rewrite README
ethancjackson
committed
Fix TypeError in _fetch_trace_metrics_df: coerce Langfuse API string values to float
ethancjackson
committed
Clean up preparation.py and config_types.py
ethancjackson
committed
Improve README: bridge smoke→notebook, clarify variant inheritance, annotate minimal config
ethancjackson
committed
Clear notebook outputs before committing
ethancjackson
committed
committing current notebook outputs
ethancjackson
committed
Add cross-modality reckless ICL experiment config
ethancjackson
committed
Expand cross-modality experiment to all Gemini models; fix thinking token handling
ethancjackson
committed

Commits on Mar 24, 2026

Commits on Apr 22, 2026

the baseline condition with the examples but without the context following instruction
ethancjackson
committed

Commits on Apr 28, 2026

updated baseline experiments
ethancjackson
committed

Commits on Apr 29, 2026

added proper baseline config and updated interactive notebook
ethancjackson
committed

Commits on May 14, 2026

Commits on May 20, 2026

Commits on May 21, 2026