Goal: keep this repo parsimonious and easy for humans + coding agents to review.
- Mechanism specs + assets:
mechanisms/<mechanism-id>/...- Example:
mechanisms/m010-reputation-signal/ - Keep mechanism docs + schemas + reference-impl + test vectors together.
- Example:
- Cross-mechanism schemas:
schemas/ - Deterministic validation / tooling:
scripts/ - Contributor coordination + maps:
docs/ - Narrative “phase” docs (high-level):
phase-1/,phase-2/,phase-3/- Do not bury mechanism reference code inside phase docs.
One subtree per PR. Prefer PRs that touch only one of:
mechanisms/<id>/...schemas/...scripts/...docs/...phase-*/...
If a change needs multiple subtrees, split it into multiple PRs and link them from a short tracking comment.
Every PR description should include:
- Lands in:
<folder(s)> - Changes:
<one sentence> - Validate:
<one command>(or “docs-only”)
When a mechanism consumes KOI outputs:
- Link to
koi-researchand the relevant path (extractor, ontology, dataset, or script). - Record the expected output shape as:
- a JSON schema (preferred), or
- an example JSON blob in the mechanism README.
In regen-network/koi-research, align changes to the existing top-level folders:
extractors/— parsers/ingestersontologies/— definitions (RDF/TTL, etc.)experiments/— exploratory worktools/andscripts/— ops + deterministic toolingsources/anddata/— source material and derived datasets
Keep PRs there “one folder category” as well.