Connect the controls to the running system

Wednesday moved a broad set of Steve’s local improvements into a verified Scout release: billing and membership controls, clearer background activity and a more deliberate document-reading workflow. A later monitoring release proved that errors could travel from the running container to an actionable alert. Separate research still kept proposals and measured outcomes apart.

• GrowthScout · ship pause, transfer and organisation ownership controls. The combined release applied five migrations after backup and verified matching public web/API identities, readiness, workers and scheduler. Billing now supports pause → transfer → activate destination billing while retaining website data, integrations and unused credits; transfer requires ownership of both organisations. Membership supports multiple owners, leaving and removal with last-owner protection, while automation survives its original creator leaving. The feature handoff passed 102 focused tests with mocked payment providers; release smoke checks covered billing and narrow-screen members. Real Stripe payment completion and an actual customer transfer were not established.
• GrowthScout · turn PDF reading into a bounded, staged workflow. The reader opens a small preview with bookmarks where available, then reads selected pages and can close the document once enough evidence is collected. The download ceiling rises to 25 MiB, while cumulative extracted PDF text is capped at 32 KiB per run. Cleanup covers close, failure and completion, with scheduled expiry cleanup for crash leftovers; excerpts remain in conversation history. Ninety-four focused and migration tests passed before the reader joined the combined deployment. This replaces the earlier one-shot limit with an explicit document lifecycle.
• GrowthScout and db3 · keep running work separate from unread results. Conversation state now tracks loading and attention independently for simultaneous chats. Finishing one conversation can mark it unread while another keeps running; reading output cannot clear active work or grant permission. WebSocket updates avoid recurring UI polling. A stale local worker and new-chat navigation race were repaired, with completion dots verified on the dashboard and cleared on opening. Focused activity checks passed 51 tests; the initial type blocker was in concurrent membership work. The later integrated release passed types and live saved-conversation smoke checks.
• GrowthScout · make the Articles grid use the available window. The Articles grid stays anchored to its container bottom as the window resizes, with rows scrolling inside it. Toolbar wrapping also accounts for the narrow library header. Twenty-six focused tests, types and conventions passed, followed by actual-app desktop/mobile review. The change joined the integrated release rather than remaining at its initial local-only handoff.
• GrowthScout and db3 · verify error monitoring from the deployed container. A later release configured Sentry monitoring and framework-owned custom logging transports. Two synthetic container errors reached Sentry as one grouped issue with the correct production environment and release, and the email alert arrived. Source maps were uploaded and live services remained healthy. Validation recorded 3,697 tests passing initially; all 23 environment-related failures passed after the disposable database configuration was corrected. Types and conventions passed. Scout hosted CI remained billing-blocked, while framework CI passed on its final revision. Synthetic alert delivery does not prove every real failure path is covered.
• db3 Framework · retain current Stop behaviour while planning faster remote cancellation. Review confirmed that Stop saves cancellation before notification, same-process runs receive prompt signals, and remote workers detect it at persistence checkpoints or a five-second recorder check. Ownership fencing prevents late cancelled output from saving; browser disconnection does not stop execution. Focused streaming, realtime, attention and history checks passed. An outbound worker socket and a controller response API remain proposed, with ordinary awaited Agent.run semantics preserved. Faster SDK cancellation alone would not immediately interrupt tools using their own timeout signals.
• DOM Studio · converge on one designer with an optional Monaco editor. Local source work removes the duplicate Studio experiment and routes its old entry points to the canvas. DomDesigner accepts a code-editor component through a small value, selection and focus contract; Monaco loads through an application-supplied function rather than becoming a package dependency. Either code view retains caret-to-layer selection and layer-to-source highlighting. Assembly documentation and inspector tooltips were updated. The dated source commit establishes implementation, but no package publication, deployment or new independent browser verification was established for this work.
• GrowthScout · distinguish exposure, ranking evidence and context size. The ad review found substantially fewer impressions while click-through rate was often higher than on the first reported day. Pauses interrupted delivery, but history did not isolate the unexplained 7–8 October impression drop or justify blaming new creative. Keyword review likewise concluded that the experimental score describes competitor strength, not a validated link requirement or ranking probability. A proposed context indicator would use the last model input against a verified window limit, keeping cumulative token usage separate. These are analytical conclusions and proposed displays, not newly shipped scoring or compaction features.