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.

StrategyChunksHit@1MRR
whole-file61.0001.000
markdown-heading131.0001.000
token-window121.0001.000

Per-query results

whole-file

markdown-heading

token-window