Add analysis requests, action workflows, alerts, and audit history without fake AI results.
Cover local platform flows, viewer write denial, Parse session behavior, pagination, and audit writes.
Make the database-free runtime reusable through a configurable local workspace id.
Reject partial platform migrations through the health check.
Add manual sync retry/cancel, local job events, and stale-worker job recovery.
Protect owner membership and validate action assignees inside the workspace boundary.
Run migrations against a newly provisioned PostgreSQL database.
Smoke-test the local HTTP service under the supported Node 22 runtime.
Add a Parse REST storage driver that preserves the existing frontend/backend API contract.
Add isolated master-key-only Voc* schemas and remote schema reconciliation.
Phase 2 - case-data persistence
Add a batch importer for the normalized Demashi dataset.
Import 2,817 operating products and 9,717 daily metric rows through bounded Parse REST batches.
Import 40 competitor relations with deterministic relation keys.
Add a streaming importer that persists the original Demashi workbook directly through Parse REST.
Create 37 mapped competitor products and backfill all 37 real product details through the company relay.
Verify Parse counts and snapshot totals against demashi-summary.json.
Switch DomesticDatasetService between static case mode and backend API mode.
Run desktop/mobile browser smoke tests against Parse REST API mode.
Phase 3 - real JD source contract
Verify the existing Fmode gateway health endpoint over public TLS (HTTP 200 on 2026-07-23).
Complete quota-controlled live JD product and review requests through the existing Fmode gateway using a process-only credential.
Confirm the catalog JD product-detail path and required itemId parameter.
Confirm the catalog JD review path and optional page parameter.
Confirm the live double-data response envelope and first-page review pagination metadata.
Confirm whether an empty first review page with next-page metadata represents collection-in-progress, then capture one non-empty sanitized review fixture.
Save sanitized response fixtures without credentials or personal data.
Implement JD product and review adapters against company-gateway contract fixtures.
Process queued jobs with partial-failure events and bounded retries.
Implement idempotent product/review UPSERT and truthful job progress updates.
Verify source UPSERT and snapshot totals against the provisioned PostgreSQL database.
Phase 4 - deployable closure
Provision a dedicated PostgreSQL database and least-privilege runtime role.