Template change rollout: test on a slice, never the whole set

Editing a template edits thousands of live pages at once. One bad change is a site-wide regression. Roll template edits like code deploys.

The release procedure:
☐ Step 1 — Snapshot. Record current rankings/clicks for a fixed 200-URL cohort before touching anything. Gate: no baseline, no deploy.
☐ Step 2 — Canary slice. Apply the change to 5% of pages — a random, representative sample, not the best performers. Gate: hold 14 days.
☐ Step 3 — Read the canary. Compare canary cohort vs. holdout on impressions and average position. Gate: roll forward only if canary is flat-or-up; any 10%+ drop triggers rollback.
☐ Step 4 — Staged widen: 5% → 25% → 100%, a week between steps, re-checking the gate each time.
☐ Step 5 — Keep the rollback ready. Template changes are versioned; reverting is one deploy, not a re-edit.

Guardrail: never change template structure and content in the same release — you won't know which moved the metric.

Ship gate: don't publish until all boxes are checked.