Fresh context review is not bureaucracy

Fresh context review is what keeps the builder from reviewing their own story.

The builder always has extra information. They remember the false starts, the awkward workaround, the reason a strange line exists, and the command that almost worked. That memory is useful while building. It is a liability when it is time to decide whether the work actually stands on its own.

I prefer a reviewer who gets a thin packet: goal, diff, verification, risk, question. If the reviewer needs the whole history, they can ask for it. Starting cold makes the important thing harder to dodge.

why the builder is the wrong final judge

After a long session, the intended solution starts to look more real than the actual artifact. The missing edge case feels handled because it was discussed. The test feels run because a similar command happened earlier. The requirement feels satisfied because the patch moved in the right direction.

That is exactly the bias fresh review is supposed to break.

the handoff should be small

goal:
  one sentence about the change

changed:
  files that matter

verified:
  commands that actually ran

risk:
  what still makes me uneasy

question:
  the one thing the reviewer should decide

If the handoff cannot fit in that shape, it is probably trying to smuggle in too much explanation.

receipts matter more than confidence

“Tests pass” is not a receipt. Which tests. Which command. Did it exit zero. Did it cover the changed path or just the whole project by habit.

That sounds fussy until you see how often agent output sounds more complete than the verification behind it. Fresh review is the part that checks whether the claims and the evidence are actually in the same room.

pay the cost where it buys safety

I would use fresh review for public writing, auth changes, migrations, automation, and anything with a wide blast radius. I would not spend the same effort on a typo fix.

The point is not to create ceremony. The point is to make sure the diff has to satisfy the goal without leaning on the builder’s memory to explain away the gaps.