Separate local parsing from model indexing, bind review decisions to immutable manifests, persist vectors behind active profiles, and expose retrieval, chat, evaluation, and document workflows through the React workbench.
Constraint: Live Bailian authentication currently fails for all three configured capabilities
Rejected: Direct upload-to-embedding flow | bypasses local review and manifest binding
Confidence: high
Scope-risk: broad
Directive: Keep private-data deployment blocked until authentication, RBAC, and separate database roles land
Tested: make verify; fresh and replay Docker document smoke; worker recovery smoke; frozen synthetic evaluation; migration 0003-0004 roundtrip
Not-tested: Successful live Bailian calls, OCR, real multi-user authorization
Expose the verified synthetic retrieval path through a typed React client and a non-root Nginx edge while keeping database credentials and data-network reachability out of the browser tier. A fixed-origin gateway preserves request boundaries and now closes upstream streams even when downstream disconnects before body iteration. The deployment ADR and runbooks record the four-network topology and its accepted Web edge-egress risk.
Constraint: The previously exposed Bailian key must be revoked and no live provider credential may enter Git, images, logs, or the browser.
Rejected: Connect Web directly to the data network | expands lateral reach to PostgreSQL.
Rejected: Publish the database-aware API on the edge network | gives a credential-bearing process a public default route.
Rejected: Buffer streaming responses in either proxy | prevents incremental chat delivery in the future.
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Do not mark Stage 1 or Stage 2 complete until the rotated-key live smoke and remaining stage gates pass.
Tested: make verify; 65 backend tests; 14 frontend tests; Docker image build; container health and isolation probes; real HTTP demo/status/search/docs/OpenAPI checks.
Not-tested: Live Bailian models, real document ingestion, business chat SSE generation, and final browser screenshot automation because the browser skill runtime was unavailable.
The project begins with architecture, data governance, reproducible evaluation, deployment boundaries, and secret-handling contracts so later code has measurable acceptance criteria.
Constraint: Real provider credentials, workspace identities, and restricted geological data must never enter Git
Rejected: Add placeholder runnable services in the design commit | would imply unverified implementation readiness
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Run make verify before every commit and update ADRs when architecture boundaries change
Tested: Secret scan, Markdown links, YAML parse, shell syntax, and staged diff validation
Not-tested: Application runtime is intentionally deferred to the implementation stage
