# Bench questions — one per line, ignoring blank lines and lines starting with #.
# Mix of: code-discovery, doc-retrieval, cross-modal (code↔doc), comparative.
# Edit freely.

Where in the storage layer is foreign-key validation performed on an INSERT?
What does the WASM plugin architecture do, per the project's documentation?
Find the function that handles WAL flushing — what's its signature?
Does this codebase support time-travel queries? If yes, where is the API entry point?
How does the engine handle multi-tenancy? Point to the relevant module and explain in 2-3 sentences.
What's the default WAL sync mode and where is it set?
In the docs, what does HIGH_AVAILABILITY_OVERVIEW.md cover?
What's the difference between the heliosdb-cache and heliosdb-storage crates?
Where is the HNSW index implementation and how are inserts maintained?
List the public types exposed by the heliosdb-vector crate's main module.

## broad-scan — questions that force the no-MCP baseline to walk many files.
## Designed for the 1-10 GB `large` / `huge` corpora where Read+Grep
## scales linearly while MCP wrappers stay flat.

Describe the overall architecture of this codebase in 3 paragraphs.
Where is rate limiting implemented, and what are the configurable knobs?
List every place that handles user authentication or session validation.
What modules depend on the storage layer, and how do they consume it?
Summarise the public API surface a downstream user would interact with.
