Memory recipes
When a test case passes, the engine remembers how it found each element. Next run, those steps resolve from memory and skip the AI call — faster and cheaper.
Memory is only written when the whole test case passes, so a recipe that "worked at this step" but failed later never pollutes the cache. A recipe that fails three times in a row is discarded and the AI works it out fresh.
The Memory Recipes page shows what has been learned, grouped by suite and test case. You rarely need to touch it — it is there for when you want to know why a step resolved the way it did.
