The version that worked until it didn't
Some experiments fail immediately. They're the kindest ones: they don't waste much time.
The dangerous ones look right for a few hours. The texture is good, the behaviour seems predictable and we're already mentally naming the jar. Then something separates, thickens, forms lumps or decides to develop a personality of its own.
This one went through five versions.
Each correction solved one problem and introduced another. One became too thick. Another lost stability. Another finally seemed perfect, but remembered the previous versions and made sure to demonstrate it.
What the experiment left behind
The fifth version didn't come from a brilliant formula. It came from carefully observing everything the others did wrong.
That's an important distinction: experimenting isn't mixing things at random. It's changing one variable, waiting, observing and resisting the urge to “fix” five problems at once.
The contents of the jar remain private. The conclusion can be public:
A stable formula is just a collection of mistakes that no longer need to happen again.