Every time you report results, your Oddling compares formats against each other and looks for a real gap. When it finds one, it tells you and it changes what it suggests from then on.
It will only claim a pattern when
- At least two posts have reported numbers.
- The best format is beating the worst by at least 12%.
- There is more than one format to compare.
On the first read, with only one post per format, it says "early read" and suggests trying another to see if it holds — rather than announcing a trend from a single comparison. As more posts come in the language firms up and the confidence score rises.
Whatever it has learned goes into the prompt for every idea after that. This is the part that makes it feel like it is getting smarter, because it genuinely is: same business, different suggestions, because the evidence changed.