Evaluation metrics and harness for mnemonist.
- [
search] — retrieval quality (MRR, NDCG, precision@k, recall@k) - [
embedding] — embedding space quality (anisotropy, discrimination gap, intrinsic dimensionality) - [
quantization] — quantization fidelity (MSE, cosine distortion, recall impact) - [
dataset] — synthetic benchmark dataset generation - [
harness] — end-to-end eval runner producing structured reports