[canonical 4.2 harness] session=C1
[canonical 4.2 harness] protocol: 10 warmup + 100 continuous timed iters
[canonical 4.2 harness] inter-shape settle: 5.0s (NOT section-4 cooldown)
  lcsa_small_seq4k                 d=0.239 V2 p50= 0.970ms SDPA p50= 2.578ms ratio= 2.66x
  lcsa_small_seq4k_sparse          d=0.067 V2 p50= 0.715ms SDPA p50= 2.586ms ratio= 3.62x
  lcsa_mid_seq8k                   d=0.119 V2 p50= 1.219ms SDPA p50= 6.441ms ratio= 5.29x
  lcsa_mid_seq8k_sparse            d=0.030 V2 p50= 0.612ms SDPA p50= 6.443ms ratio=10.53x
  lcsa_large_seq16k                d=0.120 V2 p50= 2.031ms SDPA p50=12.696ms ratio= 6.25x
  lcsa_large_seq16k_sparse         d=0.030 V2 p50= 0.920ms SDPA p50=12.696ms ratio=13.81x
  lcsa_mid_seq8k_very_sparse       d=0.011 V2 p50= 0.517ms SDPA p50= 6.435ms ratio=12.44x

[canonical 4.2 harness] session 'C1' -> docs/methodology/canonical-bench-data.json
