Benchmark
Native Markdown Context Benchmark
Compare agent retrieval accuracy, token use and latency across SQLite and Markdown context stores.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Richard Ng describes forty work questions, each run three times per setup, over approximately 11,000 records projected from Native’s own event log. This is the vendor’s own atypical workload, not a representative sample of companies.
Methodology
Use OpenCode with DeepSeek v4.1 Flash at High effort across SQLite through MCP, structured Markdown, Markdown with wikilinks, and plain Markdown. Compare correctness, elapsed time and tokens per fully correct answer, including unsuccessful attempts.