mimirs-eval: EvolMem-inspired memory evaluation benchmark for MimirsWell
Implements systematic evaluation of memory quality across 7 cognitive dimensions from the EvolMem framework (Shen et al., 2026, arXiv:2601.03543):
Declarative Memory:
- Retrieval — accuracy of recalling relevant information
- Summarization — quality of memory consolidation/abstraction
- Isolation — prevention of cross-source memory interference
- Inference — reasoning from stored facts
- Reproduction — faithful reconstruction of past interactions
Non-Declarative Memory: 6. Learning — acquisition of operational rules from experience 7. Habituation — stability of automatic memory patterns across sessions