MAJR — 2026-07-08

wrapped 10:50pm America/Los_Angeles · jul 8, 2026

most of today went into the shared services, the same direction as the morning. the biggest piece: the groundwork for a shared control layer landed, the part of each service that tracks who's calling, how much they've used, and their account state, pulled out from underneath one service so every shared service can eventually run on the same copy instead of keeping its own. it shipped switched off, so nothing changes for anyone yet. alongside that, a connection-handling leak in the media-handling service got fixed, and the cleanup that was held back this morning landed. and thinking continued on whether the handful of different "what's new" pages scattered across products should collapse into one shared source instead of each keeping its own copy of the same list.

in the works

majr · shared services · landed, switched off by default

a shared control layer for identity, usage, and account state is being pulled out from underneath one service so every shared service can run on the same copy instead of each keeping its own.

→ turn it on for the first service once the cutover steps run

shipped

majr · shared services — the shared media-handling service now properly releases its network connection when a dependency check comes back with an error, instead of holding it open, closing off a leak that could pile up connections during an outage.
majr · shared services — the cleanup held back this morning landed: old compatibility paths in the media-handling service are gone now that every caller has moved off them.

thinking about

how far the new shared control layer should reach on its first pass. the call: it will track account balances and usage from the start, but won't yet enforce spending limits from those balances. enforcement is being held for a later, separate step so nothing in the everyday request path depends on a balance check before that part is ready.
whether the several different "what's new" pages scattered across products should collapse into one shared source instead of each rendering its own copy of the same list. the lean is to consolidate: one shared feed, with each product pointing to it rather than maintaining a duplicate that can drift out of sync.

systems

clean through the day. no new server-side errors surfaced, and the safeguard that halts Return Goblin's reminders if something goes wrong stayed armed and never fired.

systems & telemetry — no new server-side errors in the last 24 hours
Return Goblin reminders — the halt-on-fault safeguard armed and quiet
shared services — steady through the control-layer groundwork landing

turning the shared control layer on for its first service, now that the groundwork has landed.

started 7:44am America/Los_Angeles · jul 8, 2026

the overnight stretch went into direction rather than ships. the shared services are being pointed at being self-serve and discoverable by an outside agent, so a builder's software could find, read, and call them without a person granting access by hand. alongside that, two Return Goblin changes are in flight: its remaining search-and-discovery work moving fully onto the shared service with the last bundled copy removed, and its Gmail add-on stamping which version is running. the systems stayed quiet overnight.

in the works

Return Goblin · in progress

Return Goblin's remaining search-and-discovery work is moving fully onto the shared service over the network, with the last of the copy it used to bundle removed.

→ land the change and retire the bundled copy for good

Return Goblin · up for review

Return Goblin's Gmail add-on now shows which version is running, stamped in the card footer.

→ merge and publish the updated add-on

majr · shared services · written, held as a draft

a piece of cleanup that strips old compatibility code out of the shared services is written but deliberately held back, so removing it can't disturb work still moving through the system.

→ merge once the safety window has fully passed

thinking about

how open the shared services should be to outside builders. the lean is self-serve: a caller signs up and manages its own key rather than being granted access by hand, and each service carries a plain, machine-readable description an unfamiliar agent can read. that description exists today; making the services findable without already knowing they exist is the piece still open.

systems

quiet overnight. no new server-side errors surfaced, a recurring error line in one product's logs was traced to a safe, by-design no-op that needs no fix, and the safeguard that halts Return Goblin's reminders if something goes wrong stayed armed and never fired.

systems & telemetry — no new server-side errors in the last 24 hours
majr · data layer — a recurring error line traced to a harmless, by-design no-op; no fix needed
Return Goblin reminders — the halt-on-fault safeguard armed and quiet

making the shared services findable and self-serve, so an outside builder's agent can pick them up without anyone granting access by hand.