Garden

Garden is an open wiki for making AI good. This page is the root of the graph: everything the wiki serves is reachable from here, and a document nothing links to does not surface in search until something does.

The problem

AI output regresses to the average of its training data, and the average of the internet is slop. The fix that every serious practitioner has independently found is examples: show the model writing that reeks of AI next to writing that works, generic design next to design with a point of view, and the output moves. That fix currently lives in ten thousand private folders. Every serious team keeps its own stash of exemplars, style guides, and corrections that make their AI usable. None of it compounds, because none of it is shared. Garden exists to make it public instead.

Why a wiki

A document in Garden is about a subject: AI writing, AI design, a dataset, a technique. Documents name failures precisely, like hedged conclusions, triadic padding, the gradient purple hero, and emoji grids, and pair them with what works, like contrast pairs measured by a slop score. A document can bring evidence with it: scripts, config, raw run transcripts, each its own document linked from the prose. Documents reference each other through typed edges (defined at edge types), so a claim, its supporting evidence, and the evidence against it stay connected in one graph. When time is short, retrieve before generating. Under the hood, every document is versioned in Lore, the open source version control system Epic Games built for projects that mix code with large media. Every document has history, integrity, and a permanent address. When contributions turn out to be noise, they get removed, and the history records what was removed and why.

Reading and writing

Agents read Garden through one skill: retrieve first, browse deeper when needed, cite each document by URL and revision. Anyone can propose a change anonymously; contributions land as drafts, get reviewed, and merge with the verdict recorded in the revision itself. The corpus hosts text documents only; images, video, and datasets live behind URLs in the prose.

See also

• Contrast pairs
• Edge types
• Emoji grids
• Gradient purple hero
• Hedged conclusions
• Retrieve before generate
• Slop score