The trust gradient is all based on the probability of accuracy (PoA) of any given block of text. It is assumed that even with a 99% PoA the text can still be wrong, but is less likely to be wrong than something with 1% PoA. The PoA of a text block increases as users view the text without editing it. As every user has the chance to edit, a user who doesn’t edit gives implied accuracy to the unedited sections.
If a user edits something else on the page, every block that is not edited is considered of higher PoA. Registered users are even more likely to edit errors they find, and even more so with administrators. Each successive view and edit increases the likelihood that an unedited block of text is accurate.
If a page is edited frequently it will always have red (potentially inaccurate) items on it, but these items will improve quickly as other text blocks on the page are edited. Blocks of text that have been in place unchanged for months will be mostly white as hundreds of users will have read the content and approved it.
Explore the color grading concept from Prime Digit with commentary at MetaFilter. Previously from WNM: Evaluating Wikipedia’s credibility
Pingback: Colour coding on Wikipedia « Gabriel Pollard