detect_drift previously ignored columns present in the source but absent from the target. The docstring claimed "they appear naturally via SELECT *", but the runtime's incremental INSERT path then issues INSERT INTO target SELECT * FROM source against a target whose schema is fixed — and BigQuery / Snowflake / Databricks all reject with Inserted row has wrong column count.

The natural "source schema evolves; replicate again" workflow was structurally broken.

What this adds

• DriftResult.added_columns: Vec<ColumnInfo> populated in the same single pass over source columns. Reuses the existing ColumnInfo shape (name + data_type) so no codegen ripple.
• New helper drift::generate_add_column_sql mirroring generate_alter_column_sql. Standard ALTER TABLE … ADD COLUMN works across all four adapters; no dialect override needed.
• Runtime change in run.rs: when the drift result reports added columns (and DropAndRecreate isn't already firing for type drift), execute the ALTER statements before the regular INSERT path. Surfaces as action: "add_columns" in drift.actions_taken so orchestrators can observe schema evolution alongside data movement.

Verification

Extended live/drift/run.sh into a three-stage flow:

1. Stage 1: initial replication of a 3-column source. No drift.
2. Stage 2: ALTER source ADD COLUMN region; rerun — asserts add_columns action fires and target gains the column without a full refresh (historical rows stay, new rows include the source value).
3. Stage 3: DROP + CREATE source with id type changed INT64→STRING; rerun — asserts drop_and_recreate action fires and target's id column is now STRING.

==> stage 1 drift: tables_checked=1, tables_drifted=0 OK
==> stage 2 drift: action=add_columns OK (["added column 'region' (STRING)"])
==> stage 3 drift: action=drop_and_recreate OK (["column 'id' changed STRING → INT64"])
==> target columns: id,name,_updated_at,region
==> target customers.id is now STRING (drop_and_recreate took effect)

Idempotent across consecutive runs.

Sibling gap (not fixed here)

DriftAction::AlterColumnTypes is detected in drift.rs for safe widenings (e.g. INT→BIGINT) but the runtime at run.rs:4105 still only wires DropAndRecreate. Safe widenings silently fall through to the next INSERT. Documented as finding #9 in live/README.md.

Test plan

• cargo test -p rocky-core --lib drift — 35 passed (includes 4 new tests)
• cargo clippy -p rocky-core -p rocky-cli --all-targets -- -D warnings clean
• cargo fmt -p rocky-core -p rocky-cli --check clean
• live/drift/run.sh against the BQ sandbox — exits 0, all three stages pass
• Two consecutive runs both pass (idempotency)