SEDIM
The knowledge and evidence architecture of Nage AI OS, serving as the reference implementation of the KLM specification to support reasoning records and evidence visibility.
Timeline (last 90 days)
SEDIM is the knowledge and evidence architecture of Nage AI OS and exists as the reference implementation of KLM; the version currently cited is KLM v0.3 (repository tag v0.3.0).
In a historical platform-domain routing benchmark, the system completed 137/150 correct classifications across 15 categories, for an accuracy of 91.3%, with a nominal 95% Wilson interval of 85.7-94.9%.
The numerator of the above routing diagnostic benchmark was recovered from a rounded summary, and the attribution of the historical samples has not been verified, so the 91.3% accuracy carries uncertainty and cannot be cited directly as a precise performance metric.
In the corrected two-source test, the native model weight path configuration had 0/15 successful answers.
The capacity growth comparison could not establish a reliable effect because the legacy scorer was flawed and the original outputs were missing.
In legacy packing, 118/221 selected cases lost content containing the gold-standard text; the fixed packing retained this text in 221/221 cases, but this measures text presence rather than answer quality.
The Python and TypeScript validators reportedly agree on six checks per synthetic fixture suite, but an independent production disclosure was not accepted.
Recovery, re-derivation, human rubric support, boundary isolation, and bounded response evaluation remain at the proposal stage.
The authors emphasize that evidence visibility does not establish correctness, causal dependence, safety, or regulatory compliance; independent reproduction and a complete public audit package remain future work.
All Facts (9)
SEDIM is the knowledge and evidence architecture of Nage AI OS and exists as the reference implementation of KLM; the version currently cited is KLM v0.3 (repository tag v0.3.0).
50%UnverifiedIn a historical platform-domain routing benchmark, the system completed 137/150 correct classifications across 15 categories, for an accuracy of 91.3%, with a nominal 95% Wilson interval of 85.7-94.9%.
50%UnverifiedThe numerator of the above routing diagnostic benchmark was recovered from a rounded summary, and the attribution of the historical samples has not been verified, so the 91.3% accuracy carries uncertainty and cannot be cited directly as a precise performance metric.
50%UnverifiedIn the corrected two-source test, the native model weight path configuration had 0/15 successful answers.
50%UnverifiedThe capacity growth comparison could not establish a reliable effect because the legacy scorer was flawed and the original outputs were missing.
50%UnverifiedIn legacy packing, 118/221 selected cases lost content containing the gold-standard text; the fixed packing retained this text in 221/221 cases, but this measures text presence rather than answer quality.
50%UnverifiedThe Python and TypeScript validators reportedly agree on six checks per synthetic fixture suite, but an independent production disclosure was not accepted.
50%UnverifiedRecovery, re-derivation, human rubric support, boundary isolation, and bounded response evaluation remain at the proposal stage.
50%UnverifiedThe authors emphasize that evidence visibility does not establish correctness, causal dependence, safety, or regulatory compliance; independent reproduction and a complete public audit package remain future work.
50%