// the find
danielmiessler/Ladder
A system for autonomous creation and optimization.
Ladder is a markdown-based pipeline for tracking ideas through six stages (Sources → Ideas → Hypotheses → Experiments → Algorithms → Results), driven by a small bun/TypeScript CLI. It's aimed at people who want a structured, git-native way to log a research/improvement loop for themselves or point an AI agent at.
The Sources/Ideas/Hypotheses/Experiments/Algorithms/Results structure with ID-prefixed markdown files is a genuinely clean way to track a pipeline — no database, fully diffable, easy to grep or script against. The CLI commands (add/list/status) are simple enough that wiring an agent to call them is trivial. Forking as the distribution model (your fork = your instance) avoids any shared-state or multi-tenant complexity.
Despite the 'autonomous' framing, there's no code that actually runs the CONSUME/DREAM/STEAL/TEST phases — the CLI only adds and lists markdown files; a human or external agent still has to do the thinking and typing. The one example Result in the repo is literally titled 'Pending experiment not yet started,' so there's no evidence the loop has produced a single verified outcome. The Tools directory is two files (cli.ts, pai-integration.ts) with no tests, and the README spends most of its length on philosophy and links to the author's own blog rather than documenting the frontmatter schema or how scoring/mutation in the EVOLVE phase is supposed to work.