Observe
What is not working, what is slowing down, or what is approaching the limit? I start from a real business operation.
The method
A method for finding the real cost
I do not start with a fashionable tool. I start with the operation, the volume, the verified result, and the place where work begins to multiply.
Separate
I divide reading, processing, external calls, validation, and writing so each cost becomes visible.
Test
I change one dimension at a time: fields, rows, repetitions, instances, lookups, or concurrent work.
Change
I choose the smallest structural change that removes unnecessary work while preserving the functional result.
Verify
I measure every stage again. An optimization moves the critical path; it does not make it disappear.
Transfer
I leave logs, tests, conventions, models, and a clear proof that another person can repeat.
Judgment before ritual
Not every problem needs the entire method
A one-day script, a disposable prototype, or a page that reads ten rows may not justify full instrumentation. The decision is still measured against risk, lifetime, volume, and the cost of returning later.
Public laboratories
The same result, reached through different paths
The repository compares repeated reads with one set-based read, LINQ/Where with dictionaries and direct arrays, line-by-line writes with batches, IndexOf with Aho-Corasick, serial work with controlled parallelism, and wide paging with ID-first paging.
18demos
5×measured runs
[PHASE]execution map
SHA-256evidence manifest