Earlier quoted context omitted.
This is exactly what I worked on ~10 years ago at amazon, embedding steganographic information into a certain internal app that reported confidential sales numbers. Ended up catching the person who leaked this: https://techcrunch.com/2011/10/04/leaked-sales-data-puts-kin...
Out of curiosity, can you share a ballpark of how many different variations can you generate per, say, paragraph of text?
I was able to develop enough variations that vastly outnumbered our users though, so even with just a portion of a screenshot, you could fairly easily figure out where it came from.
Just looking at possible CSS rules and you can see where the variations come into play - cell width, border width and styles, font color(e.g. the specific green or red that represents gain/loss), kerning, column placement , etc.
On top of that, I only fudged with display elements - the numbers were never changed. However, the numbers were updated on a near-continuous basis by ingesting various logs, so any column that was live(year/month-to-date, etc) would have only a very small time range where that number could have been displayed to the user.