Reviewing Large Changes

In one sentence

Reviewing a large change means prioritizing risk and evidence first, then scanning the remainder with explicit coverage limits.

Why it matters

Large human and agent changes are common. Linear top-to-bottom reading burns attention on low-risk files and leaves dangerous paths unchecked.

How it works

• Inventory files by risk: auth, data, money, permissions, migrations, networking, crypto.
• Read those paths fully before bulk refactors and formatting.
• Check tests and config that claim to cover the risky behavior.
• Skim mechanical files last; do not let them consume the review budget.
• Document what you did not deeply review if time forced a tradeoff.

Example

For a 120-file agent refactor, open the auth and payment directories first, then migration SQL, then tests named for those areas. Only afterward sample the rename-heavy packages.