Claude auto-memory chunking spike results
Generated from the synthetic corpus by mise run bench:auto-memory-spike.
This measures deterministic FTS5 behavior; semantic/vector quality requires the later live-model benchmark.
| Strategy | Chunks | Hit@1 | MRR |
|---|---|---|---|
| whole-file | 6 | 1.000 | 1.000 |
| markdown-heading | 13 | 1.000 | 1.000 |
| token-window | 12 | 1.000 | 1.000 |
Per-query results
whole-file
q-architecture: expectedproject_architecture.md, topproject_architecture.md, reciprocal rank 1.000q-busy: expecteddebugging.md, topdebugging.md, reciprocal rank 1.000q-release: expectedworkflow.md, topworkflow.md, reciprocal rank 1.000q-candor: expectedfeedback_candor.md, topfeedback_candor.md, reciprocal rank 1.000q-index: expectedMEMORY.md, topMEMORY.md, reciprocal rank 1.000
markdown-heading
q-architecture: expectedproject_architecture.md, topproject_architecture.md, reciprocal rank 1.000q-busy: expecteddebugging.md, topdebugging.md, reciprocal rank 1.000q-release: expectedworkflow.md, topworkflow.md, reciprocal rank 1.000q-candor: expectedfeedback_candor.md, topfeedback_candor.md, reciprocal rank 1.000q-index: expectedMEMORY.md, topMEMORY.md, reciprocal rank 1.000
token-window
q-architecture: expectedproject_architecture.md, topproject_architecture.md, reciprocal rank 1.000q-busy: expecteddebugging.md, topdebugging.md, reciprocal rank 1.000q-release: expectedworkflow.md, topworkflow.md, reciprocal rank 1.000q-candor: expectedfeedback_candor.md, topfeedback_candor.md, reciprocal rank 1.000q-index: expectedMEMORY.md, topMEMORY.md, reciprocal rank 1.000