| title: CC Experiment Report API | |
| emoji: 📊 | |
| colorFrom: indigo | |
| colorTo: blue | |
| sdk: docker | |
| app_port: 7860 | |
| pinned: false | |
| # CC Experiment Report API | |
| FastAPI + DuckDB read-only API over the processed experiment Parquet. | |
| See `docs/superpowers/specs/2026-06-25-report-frontend-backend-split-design.md`. | |
| ## Build & runtime notes | |
| - **Data is NOT in this (public) image.** This Space is public so the Vercel frontend can reach the | |
| token-gated API; committing the parquet here would make it directly downloadable and bypass the gate. | |
| Instead the app pulls the parquet at startup from a **private HF Dataset**. | |
| - **Required Space secrets:** `API_TOKEN` (the shared bearer the frontend sends), `HF_DATASET_REPO` | |
| (e.g. `your-namespace/prism-cc-data`), `HF_ACCESS_TOKEN` (read token for that dataset), and | |
| `ALLOWED_ORIGINS` (the Vercel origin). On boot the app downloads the dataset into `DATA_DIR`. | |
| - **Publishing data:** `make deploy-space` first runs `scripts/push_data.py` to upload | |
| `analysis/data/processed/*` to the private dataset (run `make analyze` first to regenerate it), | |
| then syncs only the code into the Space. | |
| - **CORS is read at startup.** `ALLOWED_ORIGINS` is captured when the app process starts, so | |
| changing the Space secret requires a container restart to take effect. (`API_TOKEN` is | |
| re-read per request and does not.) | |