This week I shipped twenty-two open source packages in roughly twenty-four hours. All MIT or Apache-2.0. All publicly installable. Distributed across three package registries (crates.io, PyPI, npm) and one Model Context Protocol registry. Counting it up: 7 Rust crates on crates.io 1 Python wheel on PyPI (PyO3-bound to a Rust core) 14 Node MCP servers on npm, all listed in the official MCP Registry If that sounds suspicious, it should. Shipping volume is easy. Shipping volume without padding your portfolio with bloat is the harder problem. This post is about how I drew that line and what I refused to ship even when it was tempting. The principle: every package earns its slot The rule I held myself to was this: for each package, before publishing, I had to be able to answer "what specific problem does this solve that the alternative does not?" If the answer was vague or amounted to "well, an LLM can do this but maybe not perfectly," I left it on the floor.…