Post by Martin Fredriksson

“Premature optimization is the root of all evil.” No. Premature mediocrity is. Knuth never said “write naive code.” He said measure before you optimize. But the industry mutated this into a permission slip for architectural laziness. The result? Systems that need Kubernetes clusters to do what fits on a Raspberry Pi. “Clean Architecture” that means eight layers of indirection, every dependency injected, every decision deferred. Fourteen heap allocations per API call — hidden behind NuGet packages so nobody has to feel bad about it. That’s not clean. That’s debt disguised as design. I’ve spent 50 years building systems and 13 years working with RDF. Last year I built a SPARQL engine: 72K lines of C#, zero garbage collection, BCL-only, 100% W3C conformance. People asked: “Why that level of performance for a triple store?” Now I know: because a 3B-parameter language model running on a Pi 5 needs semantic context faster than it generates tokens. Microsecond triple lookups aren’t premature optimization — they’re an architectural precondition that naive design can never retrofit. The real lesson Knuth taught isn’t “don’t optimize.” It’s “don’t optimize the wrong thing.” Memory model, allocation patterns, dependency topology — these aren’t optimizations. They’re foundational decisions. Get them wrong on day one and no amount of profiling saves you later. I call this E-Clean Architecture: → Code IS the model — executable and extractable, not documented elsewhere → Invariants ARE the specification — falsifiable, not aspirational → Performance IS the architecture — not a layer you add when things get slow → Names ARE semantics — if you need a class called “Handler” or “Manager” or “Helper,” you haven’t understood your domain The industry has normalized architecture for mediocrity and called it best practice. Maybe it’s time to stop optimizing for naive developer comfort and start optimizing for systems that actually work. #SoftwareArchitecture #EClean #SemanticArchitecture #RDF #Performance