Keep optional work from stopping the journey

Thursday moved several of Steve’s projects towards clearer, more resilient journeys. Scout stopped an optional discovery failure from blocking unrelated AI work, made onboarding progress recover after refresh, and shipped a combined set of reporting and account controls. DOM Studio supplied two package releases and repaired the public record of those releases.

• GrowthScout and db3 · bound product discovery and isolate its failures. Production tracing found repeated requests stopping at roughly 60 seconds before a shared provider block prevented keyword research. The outage label did not prove a provider-wide outage. Discovery now caps input at eight pages and 24,000 characters, with at most 20 offerings and 4,000 output tokens. Optional failures record an issue and continue; transient discovery failures cannot trip the shared outage circuit, while genuine account-quota protection remains. Planning and writing also receive bounded product context. The fix was deployed after 220 combined regression tests. A retained crawl was reused to restart onboarding, and keyword research visibly progressed; the full onboarding journey was still running at that check.
• GrowthScout · make live progress fast without depending on perfect delivery. Onboarding now shows detailed keyword counts and notes, including an honest Waiting state before research starts. Onboarding and site health share event validation, ordering, reconnect recovery and cleanup. WebSockets prompt saved-state reads, bounded to one per second with one request in flight; temporary activity text expires after ten seconds. Connected active jobs use a 60-second safety refresh, with five-second polling restored on disconnection. Reopening or returning reloads saved progress. The combined release passed 3,501 tests, types, conventions and builds, then live dashboard and subscription checks. One opt-in Stripe sandbox test was skipped.
• GrowthScout and db3 · deploy the logging replacement without losing job progress. The logging transports implemented on Wednesday reached production, replacing reliability monitoring and operational-alert storage. Both migrations ran after backup, preserving scheduler cursors and settlement boundaries. Live checks found the API ready, seven workers and the scheduler running, no reliability monitor, and successful settlement runs. Operator exception email was enabled, but actual inbox delivery remained untested. The later combined release also completed worker handover and matched its public revision; GitHub’s Scout check passed, while its framework workflow was skipped.
• GrowthScout · filter the full health task set before pagination. Category filtering previously searched only the current 25-item page, hiding relevant link tasks. The server now filters the complete eligible collection before pagination, with compact clickable score cards and a separate saved Lighthouse section. The focused local check recovered all matching tasks and passed 118 tests before inclusion in the combined deployment. Health scores remain weighted deductions for outstanding findings, not percentages of working links. A proposed three-part overview covering speed, LLM optimisation and technical SEO remains a scoring-design recommendation.
• GrowthScout · expose useful search data and keep diagnostics scoped. The Search Console card gained a latest-30-days daily-click sparkline and expandable data table, while existing weekly totals remain labelled. Its connection card appears only when disconnected. Desktop/mobile checks used illustrative preview data, not measured traffic growth. Provider costs and token counts are now super-user-only across live chat, planning streams and restored conversations; customers still see credits used. These changes joined the final deployment. Ad preferences moved into Profile, and release review added a public privacy-page control so signed-out visitors can withdraw their choice. Generation-limit guidance now appears only when an action is blocked.
• DOM Studio · publish chart tooltip fixes and synchronise release history. Version 0.2.4 fixed line and area tooltips that stopped following hover after selection, and added opt-in sparkline tooltips with keyboard navigation and Escape dismissal. Package publication, integrity, website deployment and live chart behaviour were verified. A separate repair replaced the stale release-history fallback with notes synced from GitHub release workflows, including publication and edits. The website retains its last good history if syncing fails and needs no repository token on its server. The public feed and page were verified with approved fields and the missing releases restored.
• DOM Studio and GrowthScout · align full-width grids through an explicit gutter contract. DOM Studio 0.2.5 adds layout="flush" and application-owned gutter-class utilities. Toolbar, footer and outer header/body cell edges align with surrounding content; middle-cell spacing stays intact, and default grids retain their card layout. Responsive, virtual-row, single-column, loading and scrolling cases were checked in 16 browser scenarios. npm integrity and documentation deployment passed. Scout installed the released package and used the new Articles layout; an old dev-server cache was refreshed to expose the correct gutters. The later combined Scout release included the dependency and consumer changes.
• db3 Framework · ship opt-in tier support while retaining Standard execution. Service-tier support moved beyond the plan: text, structured output and agents preserve the selection through streaming, tool turns and queued runs. Costs use the actual returned model and tier; missing Flex tier evidence remains unpriced rather than receiving an assumed discount. Simulated-provider checks passed 205 framework and 47 Scout tests plus package-consumer validation. The committed dependency was captured in Scout’s deployed snapshot, with Standard selected for production. Existing retry limits remain; durable multi-hour recovery was not implemented, and no production cost saving was measured.
• GrowthScout · schedule feature marketing around real product evidence. Four illustrated educational drafts were saved, followed by a launch draft with a real planner screenshot. Later records show the launch published and four feature follow-ups verified as scheduled for 9, 12, 14 and 15 October at 11am London time: AI questions and citations, keyword research, prioritised site tasks, and sources/recommended reading. Two bolder comparison posts remained proposals. A separate live-page audit identified mismatched guarantee wording, an unclear trial evaluation story and opaque additional-site pricing; those findings are review evidence, not proof of completed conversion fixes or campaign results.
• db3.ai and GrowthScout · define the next boundaries before implementing them. The db3.ai source review recommends proving one complete organisation-app journey: install, control notification permissions, then disable cleanly. Trusted host apps and arbitrary customer code require different execution boundaries; a declarative app builder is the proposed first route. No WebOS feature was built or deployed. Separately, Scout’s cancellation review showed that billing cancellation is not a complete test reset: some history and files or background work need separate treatment. An admin-only reset was proposed; nothing was deleted.

The combined release also repaired six test failures caused by inherited environment settings, an unintended real HTTP call and stale assertions, then reran the full suite against the deployment snapshot. I take more confidence from that final, identified revision than from a collection of individually green handoffs.