This page summarizes where the system writes artifacts and what each directory contains.
data/
├── audience_research/ # Personalized learner/entity profiles (JSON/MD)
├── domain_research/ # Professional domain analyses (JSON/MD)
data/config/entities.yaml<display-name>_research_<YYYYMMDD>.json (e.g. Barry Bonds_research_20250903.json); stable identifiers are recorded in metadata and manifestsdata/config/domains.yaml<display-name>_research_<YYYYMMDD>.{json,md} (e.g. Coffee Roasting_research_20250903_073951.json); stable identifiers are recorded in metadata and manifestsdata/
├── written_curriculums/ # Generated curricula (per domain/entity)
├── translated_curriculums/ # Multilingual versions by language
.md files and a summary .json.data/written_curriculums/Coffee Roasting/spanish/, french/, chinese/).<entity-id>_curriculum_<language-id>_<timestamp>.md with source hashes and parity metadata.data/visualizations/
├── charts/curriculum_metrics.png
├── diagrams/ # Stable-ID Mermaid flow diagrams
├── metrics/curriculum_metrics.{csv,json}
└── visualization_manifest.json # Input/output hashes and evidence boundary
Run from the repository root:
uv run start-curriculum --non-interactive --stages visualizations --json
The render stage returns only artifacts created by that invocation and does
not treat unrelated pre-existing files in the output directory as part of the new
bundle. Validate the resulting manifest with start-validate-outputs.
data/prompts/
├── research_domain_analysis.md
├── research_domain_curriculum.md
├── research_entity.md
└── translation.md
These drive LLM prompt engineering for research, curriculum, personalization, and translation.
data/config/
├── entities.yaml
├── domains.yaml
└── languages.yaml
Use these to control which entities/domains/languages are processed.
Audit is read-only. Curation requires a human-reviewed JSON allow-list and is
plan-first; mutation requires an explicit archive destination or --remove.
--check fails on unreadable or unsafe artifacts and on unprovenanced
artifacts not recorded in --baseline. CI passes
--baseline data/artifact-manifests/current.json, so the recorded debt stays
visible without failing every pull request; a new unprovenanced artifact still
fails. The release gate uses --fail-on-unprovenanced, which fails on any
unprovenanced artifact.
uv run start-audit-artifacts --root . --write data/artifact-manifests/current.json
uv run start-curate-artifacts --manifest data/artifact-manifests/current.json \
--keep /path/to/reviewed-keep.json --only-duplicates --json
uv run start-curate-artifacts --manifest data/artifact-manifests/current.json \
--keep /path/to/reviewed-keep.json --archive-dir /path/to/archive --apply --json