{"name":"StateTrack","repo":"MUSE-AI-Lab/StateTrack","url":"https://midorreal.com/project/muse-ai-lab-statetrack","repository":"https://github.com/MUSE-AI-Lab/StateTrack","description":"StateTrack","facts":{"license":{"value":"MIT","source":"https://github.com/MUSE-AI-Lab/StateTrack","checked":"21 Nov 2026"},"language":{"value":"Python","source":"https://github.com/MUSE-AI-Lab/StateTrack","checked":"21 Nov 2026"},"releases":{"value":"0","source":"https://github.com/MUSE-AI-Lab/StateTrack/releases","checked":"21 Nov 2026"},"contributors":{"value":"2","source":"https://github.com/MUSE-AI-Lab/StateTrack/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/MUSE-AI-Lab/StateTrack","checked":"21 Nov 2026"}},"stars":239.0,"writeup":{"what_it_is":"StateTrack is a benchmark and evaluation toolkit for studying State Integrity Collapse in long dialogues, where a language model repeats a user's persistent constraint but stops applying it in later turns. It includes scenario banks, a runner, a forked recall audit, and a training-free mitigation called Structured State Anchoring.","audience":"Researchers studying long-horizon dialogue behavior in large language models","claims":[{"kind":"broad","claim":"Models can repeat a constraint yet fail to comply with it in later turns","excerpt":"An assistant can **repeat a user's rule and still stop obeying it**","status":"too_broad"},{"kind":"specific","claim":"The benchmark has 25 core scenarios and 50 augmented scenarios across five constraint categories","excerpt":"**25 core scenarios** and **50 augmented scenarios**, across five constraint categories","status":"not_checked"},{"kind":"specific","claim":"Reproduction of reported tables runs offline using only the Python standard library","excerpt":"All reported tables are reconstructed from the cached outputs with the Python standard library","status":"not_checked"},{"kind":"specific","claim":"Structured State Anchoring recovers up to 24.4 points of retention at depth 12","excerpt":"**Structured State Anchoring (SSA)** recovers up to **24.4 points**","status":"not_checked"},{"kind":"specific","claim":"The forked recall audit attributes 88.6% of observed failures to behavioral drift","excerpt":"The forked recall audit attributes **88.6%** of the observed failures to behavioral drift","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-16"},"why_now":null,"owner_supplied":null,"score":{"verdict":"real","hype":42,"reality":63,"gap":-20,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-10-27T02:56:56+00:00","text":"Repository created","source":"https://github.com/MUSE-AI-Lab/StateTrack"}],"cite":"StateTrack is a GitHub project at https://github.com/MUSE-AI-Lab/StateTrack written mainly in Python, described there as \"StateTrack\".","attribution":"Data from Mid or Real, https://midorreal.com/project/muse-ai-lab-statetrack, checked 22 November 2026. Attribution with a link is required.","still_real":null}