The best answer is: you judge the performance of a developer the same way you would judge the performance of any employee. You gather feedback and apply your own personal assessment of that individual's performance given the tasks and objectives you've assigned. There is no magic bullet.

Another demotivator is management hysteria to get the bug count to zero. Bugs need to be reported no matter what the goal is.

The best way to understand what is is going on is to talk to people informally (definitely not regularly scheduled meetings).

I've seen this kind of stuff try to be deployed in the literary world and it always falls flat. One guy can say in 100 words what someone else wouldn't be capable of getting out in 10,000. Another might be writing 5 page features and the rest of his team are writing half-pages or less. Then there's creativity and personal voice.

I'm sure there are thousands of developers out there, likely hundreds on HN who are very elegant coders. I'm sure there are many who aren't.

People need to be judged on an equal playing field, if your metrics are bias you're going to be judging someone unfairly, and I don't think anyone here would want to go to trial over an unfair dismissal. A small company wouldn't survive losing to that. So my advice would be to be exceptionally careful in the ways you judge your workers and don't distance yourself from the judging. If you're relying on a computer to decide who is a good worker and who isn't, then you're in risky territory.

Don't do this! You will get disastrous results. For example, some programmers have more bugs assigned to them because they care more about fixing things. People know this, so when they want something fixed they will try to assign it accordingly. If you only look at the metric, you won't be able to distinguish this case from a bad programmer who generates more bugs.

It's well known that metrics like this don't work. The moment you settle on one, everybody will begin optimizing for that (instead of, say, getting their job done). This inevitably leads to unintended side-effects.

Your best bet, assuming you're a manager who isn't working on the program directly, is to pay a lot of personal attention to what's happening on your team. Over time, you'll notice patterns.

There is a related pitfall I've seen many people fall into: watch out for the case where someone looks like a star performer because he's cranking out shoddy work. This guy often becomes the manager's favorite because he seems to get features done quickly, when what's really happening is he's getting them half-done and making a mess in the process. Worse, truly good programmers then compensate by taking time to clean up this guy's messes, which makes them appear less productive. One bad programmer like this can easily suck up the value of half a dozen good ones. The solution is to make sure not to take things at face value, but to watch carefully to see what effects someone's work is having on the product and team as a whole. This is hard but doable; everything that's going on in the code is ultimately visible at the human level if you know how to read it.

My co-workers have a legend that backs this up, from the time before I joined. They had a quarterly bonus system that would be adjusted for defect rate. If you had defects against your code, you'd lose on the bonus.

Needless to say that this resulted in a "black market" for defects, with little gain in quality or productivity resulting.

If someone is a perfectionist takes a year to complete a small item then it may be perfect, but by the time it is done (0 bugs to fix) the company may be bust.

The opposite, someone that is extremely productive but so sloppy that you are basically thrown in to 'crisis mode' by default is also untenable and not value for money.

The sweet spot is somewhere in the middle, good enough to go to production in an acceptable time.

Things like 'bugs fixed', lines of code or anything like that are not going to cut it because they can be inflated or spoofed with great ease.

When you hire a developer you hire time from somebody else for your dollars, as long as it is worth to you what you pay for it you have a good developer.

You can start 'qualitative' comparisons by employing several developers in parallel, seeing which ones work best and then slowly over time try to improve your 'mix'.

There are no absolutes in this.

Code reviews might be more applicable. But, like bug reports, the code being reviewed could be as much the fault of past developers as the person trying to make changes. There is also a tendency for peers to focus on minor issues like coding style, so make sure your reviews are covering real problems before you attach too much weight to the results.

Having said that, this is what I would do:

First, judge a programmer based only on the experience he claims to have. It's just not fair to dump a new programmer onto a team and expect brilliance; if his team is relatively brilliant, he will always appear incompetent. You should, however, expect improvements over time (such as, less time required to fix things, or more instances where he speaks up with new ideas).

Next, judge a programmer based on individual work. Every programmer should have some task, even a small task, that is only theirs, even if the majority of the work requires teams. This makes it easier to judge how the programmer tackles problems, how code looks, etc. This also avoids the pitfalls above, as there are no excuses. Also, some programmers work best alone; you should be seeing their best work, and if that is shoddy, you should get rid of them.

Only after considering the above, should you look at team work on large code bases.

Using quantitative measurements to track developers is great. But not when done stupidly. Using only one measurement is dumb (ie, # of bugs alone, or lines of code, or number of checkins). And removing human participation is dumb too. You're a manager, not a spreadsheet.

But, using quantitative measurements along with human feedback, peer review, and other traditional measurements gives you an early warning setup, and a better way of tracking changes in behavior.

This then allows you to have discussions with an employee like "I'd like you to be more productive, it feels like you've been slacking a bit recently". Then, you have both visual clues (they're at the desk, not with HN open), and also graphical clues (hmm, your story points completed per week went up 10%, awesome).

It's another tool. Don't confuse it with the only tool.