MAJR Daily Brief: July 24th, 2026

MAJR
Return GoblinDossierDial for curious people.
majr.app ↗ for AI builders.
not at the desk right now · last worked aug 21
← today
← jul 23jul 30 →

wrapped jul 24, 2026 · 10:05pm America/Los_Angeles

shipped

  • MAJR agent services — every deploy of a MAJR service now waits for that service's tests to pass first, so a change that breaks them can't reach production.
  • MAJR — the secrets behind MAJR's sign-in emails can now be rotated with an overlap, so rotating one no longer means a gap where those emails stop.
  • MAJR — a daily check now watches that MAJR's sign-in redirect list still matches what's intended, so a drift gets caught instead of quietly breaking a link.
  • MAJR agent services — the internal operator controls are off the public internet entirely, reachable only privately.
  • MAJR agent services — media encoding now keeps each job's temporary files inside a space that service owns, so nothing outside it is ever in reach.
  • Return Goblin — Return Goblin's hourly reminder pass now rides out a brief upstream hiccup instead of dropping the whole pass, so a return-deadline nudge still goes out on time.

deploys now wait for tests. a push to a MAJR service used to go out in parallel with its test run, so a change that broke something could ship and the tests would say so afterwards; every deploy is now gated on that service's own tests passing, and it was verified on its first real run. the sign-in work carried on from this morning: the secrets behind sign-in emails can be rotated with an overlap, so a rotation no longer means a gap, and a daily check watches that the sign-in redirect list still matches what's intended. the internal operator controls came off the public internet entirely. one thing is open: some sign-in attempts on the MAJR app are coming back with a server error, on the older path the identity switchover replaces.

investigating

MAJR · sign-in

some sign-in attempts on the MAJR app come back with a server error

instead of a code, a sign-in attempt on the app occasionally gets a server error back. it sits on the older sign-in path, the one the move onto MAJR's own identity service replaces outright, so the fix and the switchover are the same work. Next: watch it against the switchover rather than patch a path that's being retired.

in flight

MAJR · the web side is built and reviewed, parked at its remaining gates.

moving MAJR's sign-in onto its own identity service.

→ turn on phone sign-in, clear delivery outside the US, then move the accounts across.

Dossier · building and running green locally.

Dossier's full report redesigned to read as a case file: labelled sections, numbered exhibits, confidence as plain text rather than a badge.

→ put it up for review.

exploring

  • whether the machinery that lets services sleep to save money can be taken out now that one instance stays warm. leaning that way, since less machinery is less to go wrong, but holding off until the warm setup has run a few days without surprises.

system status & ops

the deploy gate held on its first live run, and the sign-in errors on the app's older path are the one thing open.

  • MAJR agent services — deploys are gated on tests now, verified live; the timeout paths stayed quiet
  • MAJR — sign-in monitoring got two more pieces; the app's older sign-in path is returning errors and is being watched
  • Return Goblin — the reminder pass now rides out the hiccup that was dropping it
  • Dossier — the intermittent dependency call is still occasional, still covered by its fallback

next: keep advancing the sign-in switchover one gate at a time, and settle whether the app's sign-in errors clear with it.

started jul 24, 2026 · 8:36am America/Los_Angeles

shipped

  • MAJR — a scheduled check now confirms MAJR's sign-in emails are actually going out, so a delivery break gets caught automatically instead of silently.
  • MAJR — the sign-in token step is now guarded, so an unexpected error there can't block someone from signing in.
  • Dossier — a brief outage of the identity key service is now loud in monitoring, while normal token checks stay quiet, so a real problem gets seen fast.

a cluster of sign-in reliability landed overnight. a scheduled check now confirms MAJR's sign-in emails are actually going out, so a delivery break would be caught on its own instead of silently. the token step is guarded so an unexpected error there can't block signing in, and a brief outage of the identity key service is now loud in monitoring instead of passing quietly. the two reliability threads from earlier stayed quiet overnight, both still covered.

in flight

Return Goblin · fix written and tested.

a noisy overnight alert on the Return Goblin worker traced to a brief upstream hiccup, not a real outage; the worker is being hardened to retry instead of erroring on it.

→ open and land it.

exploring

  • whether the per-request billing accounting can move off the critical path, so it runs after the response instead of adding any delay. leaning that way at no added cost; confirming the live numbers.

system status & ops

the new sign-in monitors are live, and the two threads from earlier stayed quiet overnight.

  • MAJR agent services — quiet overnight; the timeout paths stayed quiet
  • MAJR — new sign-in monitors are live: the email canary and identity-key-outage alerting, with the token step guarded
  • Dossier — the intermittent dependency call stayed quiet overnight; still covered by its fallback

next: with the sign-in monitors now watching, keep advancing the identity cutover one gate at a time.