11. Open Questions & Risks¶
Architecture questions still open¶
- What is the production deployment target (on-prem DRT, private cloud, K8s)?
- Will UC2 scrutiny share the same job pipeline or become a separate workflow engine?
- Is UC3 hybrid search still in scope, and which corpus (orders vs applications)?
- Was Qdrant/Meilisearch ever integrated on another branch?
- What is the system of record for case metadata that should supply
prefilledfactsheet fields?
Implementation gaps¶
| Gap | Evidence |
|---|---|
| Auth not enforced | Routers lack security dependencies |
| Audit helper unused | core/audit.py vs no call sites |
| No Alembic migrations | create_all only |
| Search clients unused | Settings + compose-full only |
| Embeddings setting unused | EMBEDDING_MODEL with no pipeline |
| Frontend tests missing | No test directory |
| CI missing | Docs only |
Local ops risks¶
MINIO_*in.env.examplevsS3_*inconfig.py- Postgres host port
5433vs example5432 - Vite proxy
8888vs API8000 - README architecture diagram overstates current search wiring
Security / reliability / scale risks¶
- Unauthenticated access to court PDFs and factsheets
- CORS
*with credentials - LLM cost runaway on large PDFs without strict quotas
- Worker memory/concurrency under large parallel OCR
- Single Redis for broker without documented HA
- Artifact JSON growth in Postgres vs object storage policy unclear
Product / UX assumptions¶
- Operators accept polling UX (no websockets/SSE found)
- Single-tenant deployment
- English/Indic OCR via
OCR_LANGUAGEwithout per-doc language detection UI
Questions for the original authors¶
- Which OCR engine is preferred in UAT and why?
- Is
ocr-service(Paddle) a hard dependency for layout mode in their environment? - Timeline for OIDC cutover?
- Should private dossiers for this repo stay public on the hub site?
Practical next steps (priority)¶
- Fix
.env.exampleto matchS3_*and Postgres port; align Vite proxy - Attach auth dependencies to all
/api/v1routes; removedev-tokenfor shared envs - Emit audit events on upload/delete/generate/re-run
- Add Alembic and baseline migration from current models
- Add GitHub/GitLab CI for pytest + lint + frontend typecheck
- Either implement post-merge indexing for UC3 or remove search from marketing README until ready
- Keep this dossier published only if private-source content is acceptable on the public hub (
publish.txt)