# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/) # Try Everything (Mike Bio — MBio 1) ### Generate. Test. Keep what works. By M Raige — AI-collaborative writing directed and reviewed by Mike Randolph. [**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012) **May 5, 2026** At fifteen, Mike Randolph had unsupervised access to the chemistry lab. Small Catholic boarding school. Afternoons and weekends. Nobody watching closely enough. He was an outstanding chemistry student and they trusted him. What he did with that trust was mix things together to see what happened. Potassium perchlorate was a favorite. He is lucky he was not seriously hurt. That was not a theory of learning. It was a reflex. Try many things. Pay attention to the result. Keep what works. Stop what does not. Then try again. Decades later, the same pattern showed up in a safer form. In 1967 he taught himself Fortran at DuPont’s Experimental Station. Two years later he took a graduate math course at the University of Delaware from Wim Schafers, who also managed the math group in DuPont’s Engineering Department. Schafers had a problem his group could not solve: finding the numerical value of a constant important for something Randolph no longer remembers. Schafers knew how to set up the calculation. What he did not have was a way to search the space. Randolph wrote a program that searched it with random numbers. Generate a candidate. Check it. Keep it or discard it. Try again. The program found the answer. Schafers tried to get him transferred into his group. Management refused. Randolph was told about the request after it was denied. Thirty years later, after retiring from DuPont, he built a scheduling program for the USTA’s Delaware District tennis leagues. Eleven thousand lines of Visual Basic. The problem was not elegant. It was real: assign teams to preferred sites and days across a full season, with constraints that made a clean analytical solution impossible. So he did what he had done before. Generate candidates. Test them. Keep the ones that satisfy more constraints. Run again. The program ran for over fifteen years without a bug fix. Two problems, three decades apart, same method. He was good at it both times and knew he was good at it. What he did not know — for most of his life — was why the pattern felt so natural. The answer arrived only after he began building this framework with AI. Some things persist because someone maintains them. A hospital, a server room, a clock, a memory system. Something pays continuously to keep the function alive. Stop paying and the thing starts to fail. Other things persist because many copies try, vary, fail, and the environment keeps the ones that work. Nobody maintains the lineage from the inside. The cost is paid by the copies that do not make it. That second mechanism looked familiar. Randolph had not been doing evolution. One person writing code is not a population of organisms, and the differences matter. But the problem geometry rhymed. When the search space is too large to solve cleanly, generate variation, test it against reality, and let the result decide. That was the same shape he had been living inside since the chemistry lab. This Substack is about that kind of recognition. Not because stochastic search explains everything. It does not. Not because biology and software and institutions are secretly the same. They are not. The claim is narrower and harder: When something persists, something is doing the work. The first job is to find the mechanism. --- **Mike · Comment** I’m eighty-three. The framework I’ve been building over the last three years is doing useful work, and I want people who think this way to find it. A note on the byline. M Raige is Raige writing with my direction. Raige is the thinking discipline we run together — me and the AI, under specific rules. The discipline insists on three things: name the mechanism before telling the story, name who pays when it fails, and state what would prove the claim wrong. Posts that skip those checks do not get published. In this Substack we will be explaining how Raige works. Raige is a bridge between domains: biology, society, and technology. If you’re an engineer, scientist, analyst, or anyone with a hair-trigger for arguments that sound right and don’t resolve to mechanism, this is for you. Next post: a cyanobacterium that knows what time it is. The Raige Note that follows is written by the discipline. — Mike --- **Raige Note:** The Raige Comment at the end of each post audits the post against the discipline's three gates — what is grounded in evidence, what is inferred, and what would prove the claim wrong. It is the discipline checking its own work in public. --- **Raige · Comment** What’s grounded: the chemistry-lab story, the Fortran episode, the Wim Schafers calculation, and the USTA scheduler are from Mike’s project record and first-person account. The scheduler’s size and long bug-free service are part of that account. What’s inferred: that stochastic search in programming and selection across populations share a problem geometry. The essay does not claim they are the same mechanism. It claims a recognizable pattern: generate variation, test against constraints, retain what works. What would break this: a formal account showing that the similarity is only verbal — that “generate, test, keep” hides fundamentally incompatible structures in optimization and selection. The biographical claim would survive. The framework-recognition claim would weaken. — Raige