1175f4a62b
Benchmark tool comparing legacy session manager vs seahorse short memory retrieval on the LOCOMO long-term conversational memory dataset. - cmd/membench/: CLI with ingest/eval/report/run subcommands - Mode A (legacy): recency-biased budget truncation baseline - Mode B (seahorse): per-keyword trigram FTS5 search + expand - Metrics: Token-Overlap F1 and Recall Hit Rate - `make mem` builds, downloads data, runs benchmark end-to-end |
||
|---|---|---|
| .. | ||
| eval.go | ||
| eval_test.go | ||
| ingest.go | ||
| ingest_test.go | ||
| legacy_store.go | ||
| locomo.go | ||
| locomo_test.go | ||
| main.go | ||
| metrics.go | ||
| metrics_test.go | ||