Here's how the search giant tries to improve results:
1. PROBLEMAn engineer comes up with an idea for improving a set of search results, such as removing irrelevant sites or discovering new sites that cannot yet be found.
2. EXPERIMENTATIONUsing a copy of the entire Web on Google's computers, a team tries out various tweaks to Google's mathematical formulas to produce more relevant results.
3. EVALUATIONEngineers send the proposed improvement to "eval," the evaluation unit, for more rigorous tests, including surveys of human raters and limited live tests on the public site. Eval suggests further improvements to the fix and creates a report on how they will affect results overall.
4. IMPLEMENTATIONThe proposal is reviewed at a weekly launch meeting and, if approved, soon rolled out to all 720 million Google users.