Module reconcile 

Completeness reconciliation (#502).

An opt-in post-run guard against silent truncation: after a successful root run, fetch an authoritative row count for the same data (a count(*) query, an OData $count, …) and compare it to the number of rows this run wrote. A shortfall beyond tolerance_pct fails the run — the point being that a half-read source must not quietly replace good data with less (especially under write_mode: overwrite, #492/#494).

The authoritative count comes from a small count-probe source the user configures (any faucet source that yields the count as a single value/row), so reconciliation works for any backend without a per-connector count() capability. Pure evaluation lives in evaluate / extract_count; the CLI executor runs the probe post-run.

Structs

CountProbe

A source that yields the authoritative count.

ReconcileSpec

Top-level reconcile: block.

Functions

evaluate

Compare rows written against the authoritative count. Ok(()) when written >= authoritative * (1 - tolerance_pct/100), else Err with a message naming the shortfall. Pure.

extract_count

Extract the authoritative count from a probe’s returned records. Uses the named field when given, else the first numeric field of the first record. Pure.

run

Run the count probe and reconcile it against written. Builds the probe source through the registry (so any connector works and shared auth: { ref } resolves), drains it, extracts the count, and evaluates. Returns FaucetError::Source on a shortfall so the caller fails the run.