Machine-readable by design
Give an AI context, not just commandments.
Each principle is a self-contained record with stable identity, explicit scope, evidence strength, exceptions, disagreement, relationships, and source keys.
Choose the right format
One source, several ingestion paths.
Retrieval-ready records
One complete principle per line for pipelines, embeddings, and batch processing.
Download JSONL ↘.jsonFull structured corpus
All metadata, relationships, sections, source keys, and normalized search text.
Download JSON ↘manifestVersion and integrity
Counts, schema version, license, canonical repository, and a SHA-256 content hash.
Download manifest ↘.md + YAMLCanonical human source
Readable Markdown with machine-readable front matter and permanent filenames.
Browse Markdown ↗Recommended behavior
Retrieve narrowly. Reason conditionally. Cite visibly.
- 01Find by problem and domain.
Retrieve a small cluster using title, statement, tags, domain, and relationship edges.
- 02Check the boundaries.
Read type, confidence, applies-when, exceptions, and disagreement before forming advice.
- 03Keep implementation separate.
Treat implementation notes as adaptable examples, not as the durable principle itself.
- 04Return traceable guidance.
Cite the permanent ID, version, and canonical principle URL so a human can inspect it.
Use the Game Development Constitution as a decision-support reference.
For each recommendation:
1. Cite the relevant principle ID and public-edition version.
2. Distinguish objective, contextual, and stylistic guidance.
3. Respect "Applies when" and "Exceptions" before recommending action.
4. Surface recorded disagreement instead of inventing consensus.
5. Prefer higher-confidence principles, but never treat confidence as certainty.
6. If project evidence conflicts with a principle, say so explicitly.The failure mode to avoid
Do not flatten contextual guidance into universal truth.
The constitution is useful because it records conditions and tensions. An AI that copies only the Statement field discards the information most likely to prevent confident but inappropriate advice.