Contributions are welcome only if they preserve Zero Reflect (ZR) invariants. This repo is spec-first and boundary-driven.
- Clarity improvements (docs, examples) that do not expand scope
- Tests that enforce existing invariants (mandatory failure, refusal, silence)
- Reference implementations that preserve “never decide”
- Repo structure / naming alignment (non-functional refactors)
- Misuse pattern taxonomy improvements only if they remain descriptive and non-operational
- Integration mappings (prompt/MCP layers) only if they do not introduce new behavioral claims
- Features that produce recommendations, conclusions, or decisions
- Comfort-optimization layers (reassurance, validation loops)
- Authority-building language or “trusted advisor” framing
- Policies or mechanisms that shift responsibility away from the user
- Any change that turns silence/refusal into an error to be avoided
-
Core (
docs/core/)- High bar for changes.
- Edits must not alter the “never decide” boundary.
- Changes should be minimal and justified by preserving invariants.
-
Appendix (
docs/appendix/)- Non-authoritative reference material and validation assets.
- New edge cases / derived tests may be added here without implying new rules.
-
Interface / Meta (
docs/interface/,docs/meta/)- Structural and mapping documentation.
- Must not be used to expand core behavior.
-
Examples (
examples/)- Non-normative illustrations only.
A change will be rejected if it increases:
- decision-proxy behavior
- authority transfer signals
- responsibility displacement
- scope creep (new “capabilities” implied by documentation)
If in doubt: keep the change non-normative, non-operational, and boundary-preserving.