limen-runtime-audit / docs /SCIENTIFIC_SCOPE.md
jeanbatuli's picture
Sync from GitHub main
d22e889 verified
|
Raw
History Blame Contribute Delete
1.28 kB
# Scientific scope
## Claim levels
- **[I] Observation:** directly measured from the supplied arrays.
- **[II] Controlled result:** supported by appropriate statistics and controls.
- **[III] Interpretation:** compatible with the observations but not uniquely
established by them.
- **[IV] Hypothesis:** a proposed explanation requiring a new test.
All metrics emitted by v0.1 are [I]. The software does not automatically
promote them to [II].
## Current supported use
LIMEN v0.1 is suitable for descriptive comparison of small open-weight models,
checkpoints, prompt groups and extraction pipelines. It can expose numerical
errors, profile differences and outlying runs.
## Unsupported conclusions
The following conclusions are outside the evidence produced by this package:
- a trajectory region is a semantic thought state;
- decodability identifies functional localization;
- smoothness means correctness;
- irregularity means hallucination;
- an activation perturbation caused an output improvement;
- a metric is invariant across models or transformations;
- a layer-wise path is a time-domain physical trajectory.
## Publication boundary
The public package is LLM-only. Non-LLM experiments and unaudited control or
compression claims are outside its scope.