Synthetic chunking benchmarks show a massive improvemen...

Synthetic benchmarks often rely on concatenated documents that create sharp, artificial topic boundaries which semantic chunkers can easily detect. Real-world documents rarely have such distinct boundaries. You should validate your findings by testing on a sample of actual production data before trusting the synthetic benchmark results.