Function checkpoint_once 

pub fn checkpoint_once(
    pool: &ConnectionPool,
    conn: &Connection,
    config: &CheckpointConfig,
    truncate_state: &mut TruncateState,
) -> Result<u64, Error>

Issue one checkpoint cycle against the task’s dedicated checkpoint connection (conn — see CheckpointConnection; NEVER the pool’s writer mutex).

Returns the observed WAL page count on success. A busy PASSIVE row has no usable observation and the compatibility wrapper returns SQLITE_BUSY; an inconsistent nonbusy frame pair instead returns SQLITE_ERROR. A connection-level pragma error is also returned; the task caller drops that connection and reopens next tick. TRUNCATE errors remain non-fatal.

The caller owns all threshold-crossing WARN logging so that warnings fire at most once per crossing, not every tick.

ADR-091 Plank 2: after the PASSIVE pass, this is also the single point that may escalate to TRUNCATE (maybe_truncate) — on the SAME dedicated connection, never a second connection or a pool checkout. A no-progress result produces a separate cross-process attribution request; the synchronous core never walks the sidecar directory. Production’s run_checkpoint_task consumes that request through an awaited spawn_blocking before continuing the tick. This compatibility wrapper intentionally returns only the historical page-count surface; the daemon calls checkpoint_once_core so it cannot discard the request.