Function with_busy_retry_policy 

pub fn with_busy_retry_policy<T, F>(
    max_retries: u32,
    base_delay_ms: u64,
    op: F,
) -> Result<T, AppError>
where
    F: Fn() -> Result<T, AppError>,

with_busy_retry with the schedule supplied instead of resolved.

The ambient resolution is correct for the binary and wrong for a test. The schedule is exponential, so its total duration is governed by whatever the DEVELOPER happens to have in XDG: at the compiled defaults of 5 retries and 300 ms a contention test costs about nine seconds, while a machine carrying db.busy_retries = 12 and db.busy_base_delay_ms = 600 pays roughly THIRTY MINUTES for the same test. Measured, not theorised — it is why the suite appeared to hang for four consecutive sessions on this workstation, and the two tests involved were reported as “running for over 60 seconds” while they were in fact working exactly as written.

A test that reads ambient configuration is the same confused-deputy shape GAP-SG-205 describes: behaviour governed by state nobody declared. So the contention tests state their own schedule and the binary keeps resolving its.

Errors

Returns AppError::DbBusy once max_retries attempts are exhausted, and propagates any non-busy error unchanged.