Fresh context review is not bureaucracy

Fresh-context review keeps the builder from reviewing the story they told themselves.

The builder remembers false starts, workarounds, and the command that almost worked. That context helps create the change. It can also make a missing edge case feel handled because it was discussed.

A reviewer should start with a thin packet:

goal:
  one sentence about the change

changed:
  relevant files

verified:
  commands that actually ran

risk:
  what remains uncertain

question:
  the decision the reviewer should make

A cold start makes the artifact carry more of the argument. If the reviewer needs the full history, they can ask for it. The first pass should not depend on the builder’s memory.

“Tests pass” is not a receipt. Which command ran? Did it exit zero? Did it cover the changed path or merely exercise the project broadly? Agent output often sounds more complete than its evidence. Fresh review checks whether the claim and proof are in the same room.

Use the extra review where it buys safety: public writing, authorization, migrations, automation, and changes with a wide blast radius. A typo fix does not need the same ceremony.

The point is not ritual. It is making the diff satisfy the goal without borrowing credibility from the person who built it.