MUSE-AI-Lab / StateTrack
git clone https://github.com/MUSE-AI-Lab/StateTrack.git
Benchmark and evaluation protocol for measuring whether LLMs keep obeying a persistent user constraint across long dialogues.
Why you might care
Separates forgetting from remembering-without-complying using a forked recall audit, and includes a training-free mitigation (Structured State Anchoring). Reported results are reproducible offline from cached outputs using only the Python standard library. Suits researchers evaluating multi-turn instruction following.
+-22
stars, 24h · #75,003
+106
stars, 7 days · #2,968
+239
stars, 30 days · #4,666
239
★ total · +80% this week
1
forks · 6 open issues
Star history (as recorded here)
LanguagePython
LicenseMIT
Created2026-10-27 (11d ago)
Last push8d ago
Watchers0
OwnerOrganization
Links
README
Loading README…
Blurb generated by claude-haiku-5-5 6d ago. Star deltas marked ~ are extrapolated from partial history.