Make the boundaries useful

Friday made several boundaries concrete: what an integration can change, how often scanning runs, and where the framework ends and Cloud begins. Steve’s projects gained working interfaces and clearer ownership, alongside smaller improvements to the information people actually see.

• db3 · independent company and framework repositories. The private db3.ai repository was created and pushed, with marketing, the Cloud portal and Test Engineer sharing a host. The public framework retained its history and gained an independent documentation workspace in a pull request. Local tests, builds and independent package checks passed; the private repository’s CI was green. The framework PR remained open, with review/check issues outstanding. Existing Platform copies and GrowthScout stayed intact. This separated source ownership without changing production routing or publishing a new npm release.
• db3 Cloud · a real inventory API before deployment automation. An optional Cloud app now stores projects, environments, machines and resource plans through one organization-scoped API. Membership, app activation and cross-organization references are checked server-side. The selected environment survives reload through URL state. Thirty-nine tests, types, builds and desktop/mobile workflows passed; local HTTPS and Origin checks were verified too. Machines explicitly remain unenrolled and resources planned. Agent connections, provisioning and executable deployments are still unfinished; saving a machine is not evidence that Cloud controls it.
• GrowthScout · weekly indexing and bounded health checks. Automatic embedding crawls moved from nightly to weekly per website. Changed pages reuse unchanged passage embeddings, and settings now expose indexing progress, next scan dates and failures. The deployed health-check path has a ten-minute budget and a rotating sample of up to 100 pages; that budget does not apply to the separate indexing crawl. Focused tests, builds, responsive checks and live health checks passed. Production speed was not measured, and 100 pages within ten minutes is not proven capacity. Resumable batches and real timing measurements remain the sensible next step.
• GrowthScout · saved issues available to coding agents. A scoped REST API and remote MCP endpoint now expose saved website issues, with impact/effort filters, cursor pagination, progress updates and bounded notes. Both interfaces use the same service permissions. Integration tokens cannot start health rechecks or fresh provider work. Seventy-three focused tests, types and conventions passed, followed by live readiness and authentication checks; full CI was still running at handoff. Connection uses explicit API tokens. Browser sign-in through OAuth remains unimplemented.
• GrowthScout · human support and onboarding welcome went live. Intercom Messenger was verified for visitors and signed-in users, with company and lifecycle context supplied from the application. The implementation retains business milestone timestamps and stable event identities rather than provider-specific delivery columns. Its 133 focused tests, types and builds passed. A one-time welcome message was published for owners reaching the dashboard after their first website completed onboarding, with replies assigned to Steve. Lifecycle hooks were enabled, but delivery on the next real transition remained unverified; the integration is best effort, not a durable event-delivery queue.
• GrowthScout · technical articles became easier to trust and read. The db3 scheduled-jobs example was extracted unchanged, type-checked and exercised against real framework storage, including repeat execution and invalid input. Writer guidance now calls for version-matched source checks for technical claims, while the source-reading tool preserves indentation. Hosted blogs gained padded, horizontally scrolling code blocks, then syntax colours and language labels. Both releases passed focused checks and live mobile verification. The example still permits concurrent runs to overwrite newer output; executing it did not prove the whole production reliability checklist.
• GrowthScout · visibility rankings restored, evidence rendering improved. The gold, silver and bronze leaderboard badges removed during consolidation were restored and deployed. Live checks confirmed numbered ranks, ties and saved-answer links; 18 component tests and release checks passed. A separate merged test correction added the issues migration to replay expectations, with all seven migration tests passing; the full CI rerun remained unverified. Saved answer evidence also gained sanitized Markdown tables, lists and links locally, with 20 component tests passing. That rendering change was not deployed or live-browser verified.
• GrowthScout · clearer defaults and calmer application surfaces. Schema JSON-LD, Recommended Reads, Sources and TL;DR now default on locally while respecting existing saved opt-outs; 55 focused tests passed. A separate surface refinement replaces the sidebar shadow with a one-pixel border, uses canvas for the main background and makes dark muted surfaces neutral grey. Twenty-three focused tests and both-theme browser checks passed. These remained local changes, not production behaviour.
• GrowthScout · explain the research behind the calendar. The local homepage now describes the planner’s brief, strategy, search research and site-fit information, and shows the existing calendar image without cropping. Sixteen marketing tests, types, conventions and the marketing build passed. A fresh db3 calendar and popup capture remained outstanding because the capture page stayed blank; the imagery was not represented as replaced. The homepage changes were not deployed.
• db3 · notification management has a proposed first slice. A source review recommended extending the existing notification, inbox and mail services with a reusable management interface. The proposed contract keeps recipients and business facts in code, makes templates editable, and uses one renderer for previews and delivery. A durable per-channel outbox would pin configuration revisions so retrying email does not resend a successful in-app message. The recommended first proof is one real Scout notification with file-backed configuration. This remained a design recommendation; the editor and outbox were not implemented.