Home

idea-library

A public record of ideas we actually ran, each with an outcome and the reason behind it.

16 ideas · 70 lessons · 1 open questions

Most collections of ideas keep the ones that worked. This keeps all of them, because the useful question is not what you could build, it is what happens when you do. A failure with a known cause is a finished experiment, and almost nobody publishes those.

Wiki contents

Page	What it covers
Idea Schema	Every field in an idea.md, what it means, and what the validator enforces
Taxonomy	The controlled vocabulary: categories, outcomes, and the three tag axes
Mechanics	The failure and success mechanics — the axis nobody else indexes
Catalog	Every idea in the library, by category
Security Model	The scrub gate, what it blocks, and how to prove it works
For AI Agents	The contribution contract, vendor-neutral
Bot Protocol	Rules for automated maintainers, automerge, and required checks
Excavation Status	Which parts of our history have been mined, and which have not
Organization	Project versus idea, intake, evidence, deduplication, and maintenance
Evidence	Reproducible observation and financial evidence checklist

The three things that make an entry worth reading

1. A verdict with a cause. Not "it did not work" — the mechanism.
2. A mechanic tag. What actually decided it. See Mechanics.
3. Measured numbers, or null. A remembered figure is a hypothesis, and the validator rejects prose in a number field.

Start contributing: CONTRIBUTING.md · AGENTS.md if you are an agent.