[DanBC]

> There's a sort of assumed good-faith for this process.

I don't understand why two people sign off changes if they assume the person making the changes isn't going to make errors or be "malicious". And I'm not sure why two people sign off changes instead of just one. There's a well understood problem when you have more than one person inspecting product - they both assume the other person is competent and will catch anything they've missed, and they each do a lighter inspection.

[magduf]

>And I'm not sure why two people sign off changes instead of just one.

Because more eyes catch more bugs. I work in a time like this where we have 2 or even 3 people review stuff; it's extremely common for one engineer to miss stuff that another one catches. It's particularly notable I think with junior vs. senior engineers: I think you really need at least one senior engineer reviewing everything before a merge, but you don't want to keep junior engineers out of the process because then they'll never learn. With trivial tickets, however, you can relax this rule and just have one person review before a merge.

[DanBC]

> Because more eyes catch more bugs.

But for product inspection we have pretty good evidence, from millions of human hours of factory work, that it's a bit more complicated.

X makes a widget, and sends it for inspection by Y and Z. Y has a look, but assumes Z will catch anything that Y misses. But Z also assumes that Y will catch anything that Z misses.

You end up with two people doing a superficial inspection and missing problems.

Of course, that's only if they're doing the same inspection twice. If they have different and clearly defined inspection roles the two inspector problem doesn't apply.

[singron]

If you need 2 approvals at Google, then usually 1 is for readability in a specific language. I.e. one of the reviewers is understood to mostly be looking at code style and not domain logic.

In my experience, you more often get the other phenomena where all reviewers (and some other people who get automatically CC'd) dogpile the review and you potentially have too many comments. Or someone ignores the review and it never gets approved. Cursory reviews were basically never a problem.

[wolco]

Technically it worked so the approvals would have gone through. These people would not have direct insight into her business objectives or intent.