Stream keeps one public repository so the evidence is easy to inspect, but the layers are intentionally separated.
music_pipeline.pymusic_decision_lab.pymusic_theory.pymusic_intelligence.pystreamlens_processor.pyadvanced_metrics.pymedia_liability_lab.py
This is where rows become metrics. Changes here should usually include tests and a note about what claim changed.
app.pystream_backend/build_static.pyrun_analysis.py
This is where API routes, static artifacts, local materialization, event contracts, and adoption commands live. The frontend should read from this layer instead of rebuilding analysis logic in the browser.
Contract surfaces:
GET /api/system/api-contractsGET /api/system/model-registryGET /api/runtime/events/contractGET /api/runtime/invariants/live
loopchii-wasm-core/
This is the Rust lane for small, inspectable metric kernels. It should grow where profiling shows a real need, not where a README wants a stronger word.
index.htmlassets/manifest.webmanifestservice-worker.jsdata/**/*.json
The browser surface is a public explanation layer. It should make the work legible, not invent capability.
tests/benchmarks/stream_backend/cli/doctor.pydocs/PRODUCTION_READINESS.mddocs/MODEL_REGISTRY.md
This is where the repository earns its claims. If a capability matters, it should have a command, a test, or a readiness note.