Fable
Two ways of prepping at the noodle shop
At the old street corner there are two noodle shops, both claiming they can serve food quickly. The shop on the left does it by serving a complete bowl of noodles, then keeping the remaining broth and toppings, so that when the next customer orders exactly the same noodles, it can be served directly. The shop on the right does it differently: before opening each day, it simmers a large pot of base broth. No matter whether the customer orders beef noodles, three-delicacy noodles, or vegetarian noodles, the base broth is the same; once the noodles are cooked, the base broth is poured over them, and then toppings are added according to the ordered flavor.
At the lunch rush, a group of company employees came for a team-building meal. Ten of the twelve ordered the signature beef noodle set—the noodles and toppings were exactly the same. The noodle shop on the left was delighted; the ten pre-made bowls in storage went straight out, and serving speed broke records. The noodle shop on the right still had to cook noodles bowl by bowl, but the base broth was ready-made, and the toppings and side dishes could be prepared in parallel, so its speed wasn't too slow either.
The scene changed during afternoon tea. Each guest who came in ordered a different noodle dish—beef, three delicacies, pork ribs, pickled cabbage, seafood—no two the same. The noodle shop on the left could not use a single item of its inventory and had to start from scratch, tripling serving time. The noodle shop on the right was unaffected: the base broth was used as usual, different toppings were already prepared separately, and serving speed was almost no different from the lunch rush.
Two noodle shops found a strange ranking in their monthly review: in months with a high share of set meals, the left shop crushed the right on total score; in months with many walk-in customers, the right shop pulled far ahead. Neither won, neither lost—facing different order structures, the two prep strategies simply had completely different efficiency distributions.
Later someone proposed a plan to the restaurant association on the old street: during peak hours, first build the base soup system according to the method on the right, covering the common part of all orders; at the same time, build express slots for high-frequency set meals according to the method on the left, with matching conditions strictly limited to "noodles and toppings exactly the same." The two systems run in parallel without interfering with each other.
After parallelization, an interesting phenomenon appeared in the meal data: the base soup covered 100% of the day's orders, but only 15% of orders hit the full combo cache. This means the right-side strategy contributed baseline efficiency, while the left-side strategy provided extra acceleration only in specific scenarios. Only after separating the statistics could the true value of each strategy be seen clearly.
In the year-end summary, one line was written: do not use set-meal caching to cover what the base broth can do, and do not use the base broth to benchmark the speed of set-meal caching. Keep the two accounts separate, and set hit conditions for the two rules separately.