Ladder — AI Guidance

What Is This?

Ladder is an open source pipeline for systematic improvement: Sources → Ideas → Hypotheses → Experiments → Results, with results feeding back as sources (the loop).

Key Conventions

• IDs: Each collection uses a prefix: SR-, ID-, HY-, EX-, AL-, RE- followed by 5-digit zero-padded number
• Format: All entries are markdown with YAML frontmatter
• Status values: draft, active, testing, complete, archived
• Linking: Entries reference upstream items (e.g., a hypothesis links to its parent idea via idea: ID-00001)
• CLI: Use bun run ladder <command> — never npm/npx

Directory Structure

• Sources/ — Raw inputs (papers, articles, observations, telemetry)
• Ideas/ — Candidate solutions generated from sources
• Hypotheses/ — Testable predictions from ideas
• Experiments/ — Structured tests with methodology
• Algorithms/ — Proven methods for specific tasks
• Results/ — Verified outcomes from experiments
• Tools/ — TypeScript CLI tooling (bun-based)

When Adding Entries

1. Use the CLI: bun run ladder add <type> --title "..."
2. Or create manually following the TEMPLATE.md in each directory
3. Always include all required frontmatter fields
4. Link to upstream entries where applicable
5. Use the next available ID number

Pipeline Flow

Sources → Ideation → Ideas → Hypotheses → Experiments → Results
   ↑                                                        |
   └──────────────────────── Loop ──────────────────────────┘

Results feed back as sources. This is the core mechanic.

Code Style

• TypeScript only, bun runtime
• No Python, no npm/npx
• Minimal dependencies
• CLI-first philosophy