One way to accidentally modify data has been when a field has become selected without the user noticing it, and the user believes to be scrolling the page, but instead of scrolling the page the value in the input field changes, unbeknownst to the user.
This happens more easily when using a mouse scroll wheel. This goes unnoticed more easily when the value changes without much visual change, for example inline only without a dropdown menu. Busy people, people who are visually challenged, for example wearing reading glasses, and people with large screens, or people with small screens may be more prone to not noticing accidental modification at the time it occurs.
Solutions could include visual indication, and confirmation. Different than generic confirmation, there could be specific confirmation of what item has changed from what value to what value. But, people would tire of confirmations. Hence, bright yellow or red visual indication might be more effective, for those who can see color. To avoid tiring users of visual fireworks, such colorful indication should be more effective if only shown in situations that appear at high risk of having been an unintended change.
A newer way for users to create bad data are unintended actions on a touch screen. For example, instead of scrolling through a feed, a social media app may record a tap on a like which the user would not notice as the user’s mind is on scrolling. Touch screen quality, material wear, humidity, temperature, food deposits, human variability can increase the probability of unintended actions.
In social media, consequences could be social. Even if sooner or later a person is asked by someone well-meaning “did you really like that?” there could be others changing their opinion of the user. Many people believe they can recognize mistakes, but can they? Less obvious, subconscious valuing, weighing, and judging can be insidious.
Then there is big data: Your unintended likes and accidentally modified data about you have a good chance of influencing one or several scores determined relating to you by algorithms.
This may seem trivial compared to our other efforts, and here I merely point out these problems, leaving them for you to solve.
There are two kinds of solutions I generally would not appreciate: Preventing mistakes by disallowing too much of human activity, or solving problems by writing several times bigger code — a recipe for more problems.