The Chunking Lab.

Two chunking strategies × three retrieval modes, run against 12 questions whose gold filing sections are known. Hit-rate@4 asks whether a chunk from the right section reached the top four; MRR rewards it landing first; latency is the mean per question on this machine. Gold marks the app's default configuration.

Measured on the bundled corpus (AAPL · MSFT · NVDA snapshot, Oct 2026). Session-ingested companies are not part of these measurements.