Show what changed, and what still needs proof

Friday improved how Steve’s projects explain permissions, background work and commercial value. Scout shipped specific account and chart fixes; voice interaction and website offers reached more limited stages. The testing work supplied an equally useful result: a longer run with timeouts is not evidence of an optimisation.

• GrowthScout · repair member removal and require joined members before ownership. The deployed member-removal fix handles bodyless requests correctly. Public release identities, health and the member confirmation screen were checked after 31 tests, types and conventions passed; no member was removed during verification. A later release restricts invitations to Admin or Member, with existing owners promoting someone only after acceptance and confirmation. The combined owner/chart gate passed 125 focused tests and live desktop/mobile checks. Broader failures were corrected and retested, but the full suite was not rerun. Hosted CI remained billing-blocked.
• GrowthScout · make AI-visibility history readable without hiding changes. The deployed chart repair labels saved checks, explains model-version gaps and opens the recorded check when a point is selected. Tooltips work with hover and keyboard input on desktop/mobile. A later local overview connects results across model changes using markers, while prompt changes produce labelled gaps; platform-specific views retain relevant version boundaries. That increment passed 37 focused tests, types and conventions, but remained uncommitted and undeployed. A Compare platforms switch is still proposed, rather than part of the shipped chart.
• GrowthScout · preserve useful access while a website is paused. Local paused-dashboard work adds a Resume website link for owners/admins leading to Billing, and permits free Domain Rating lookups with daily caching. Sixty-eight checks, types, conventions and desktop/mobile keyboard review passed. Separately, a backed-up production billing row had exactly three stale provider fields cleared, restoring its billing summary without changing the payment provider’s customer or subscription. That was a targeted data repair, not proof of a general billing fix or a new paid conversion.
• GrowthScout · build an opt-in voice and page-guidance experiment. A separate branch now has a persistent voice panel, typed fallback, compressed page context, highlighting and navigation. It delegates background work through existing Scout routes and displays progress/results. The feature is disabled by default and limited to superusers. Twenty-one tests, types, conventions and the enabled production build passed before the branch commit. Signed-in desktop/mobile review, microphone timing and a real voice-to-background-task round trip remain unverified because testing stopped at sign-in. It is neither merged nor deployed; billing integration remains unfinished.
• GrowthScout · stop an optimisation attempt without claiming a speed gain. The latest full test run took 16 minutes 29 seconds and ended with four timeouts. Their relationship to the changes or the previous baseline was not established. Source review found no concrete loss of the requested real database, queue, checkpoint or migration assertions, but that review ran no workloads and cannot validate performance. The task stopped with local uncommitted changes, including the latest untested health and screenshot edits. A clean passing baseline is needed before further optimisation.
• db3 Framework and db3.ai · document tools and separate hosting from execution semantics. Dock gained three local guides covering setup, projects/terminals and servers/routes, with four expandable demo screenshots. The docs build, 76 documentation tests, runtime check and desktop/mobile review passed; deployment was not established. Separate architecture research retained the durable queue, continuously running workers and optional live communication as distinct responsibilities. Cloud error reporting was proposed as a default project capability, with app/environment/release identity and separately controlled AI investigation. No new hosting backend or automatic Cloud monitoring service was implemented.

  

  

  

  

• Personal site · make the software-building offer reviewable. A local Build your software page and menu/homepage links explain the savings-share offer, ownership and future automations, with an interactive calculator. Browser navigation checks caught offer cards losing their link addresses after loading; that was repaired and rechecked. Build and browser checks passed across the site’s three modes and mobile. The page links to AI Offers without putting it in the main menu. These changes remain local, uncommitted and unpublished, with no accepted offer or realised saving claimed.