LogisticsRnD · branch chief/38c18fdf1d · live on logistics.loresis.ai as 24d7510 since 2026-10-26 07:24 UTC · merged to main as 88e44fe

LogisticsRnD reports its failures to Chief, and checks the features you rely on

Every warning and error from any user now reaches Chief as a labelled event. A check every 30 minutes also finds features that broke while nobody was using them: email connections, meeting transcripts, Browser Use and the services behind them. It is live in production. The other forks and Loresis itself are next, through one Chief logcover run each.

Livelogistics.loresis.ai runs it, shipping to Chief

14relied-on features checked every 30 min

0broken in the first production checks

372/372tests passing

Is a feature broken? The scheduled check

Each check reads only; none calls a Claude model. When a feature breaks, Chief gets an event at once, a reminder every 6 hours while it stays broken, and a note when it recovers. A feature that isn't configured or connected counts as "off" and sends nothing.

Feature	How it's checked	Breaks as
Gmail, Outlook, GitHub connections	Composio account status, then a live profile read through the connection	warn expired, revoked or refused
Google Calendar connection	Composio account status	warn
Mail and calendar sync	whether recent sync runs finished	warn failing or stale · error hung
Meeting transcripts (notetaker)	calendar sync freshness, calls that failed or got stuck, undelivered Recall webhooks	error stalled or stuck · warn a meeting without a transcript
Browser Use	the service itself, and each workspace's saved cloud browser	error credentials or credits · warn limit, outage, missing profile
Claude subscription	the credential file only, no model call	error missing or expired · warn under 1 h left
OpenAI, Telegram bot, notetaker setup, logistics pipeline, storage	a free API read, the poller's state, settings, the pipeline files, a write probe	error (Telegram polling: warn)

Before this, an expired Gmail or Outlook connection simply looked "not connected" in the app; now it's reported with its status. Single meetings also report on their own: a transcript Recall failed, or a notetaker that left without a recording.

First production result (24d7510, 07:27 UTC)

9 workspaces, nothing broken. ok: Claude subscription, storage, logistics pipeline, Browser Use service, OpenAI, notetaker setup, and Gmail plus its mail sync in 1 workspace. No workspace currently has Outlook, a notetaker calendar or a saved Browser Use browser connected, so those checks are off for now. They start automatically once something is connected. Telegram isn't configured on this server.

Failures during use

API failures: only 5xx were printed, as raw stack traces → every failure is a labelled event named after its route, e.g. api.logistics.desk.requests.deliver.
Failed analyses, drafts, B/L requests, assistant runs, dashboard builds and syncs used to be visible only as a status on screen. Now each also sends an event with its ids; failed analyses include the stage, the Python step and the traceback tail.
Browser crashes: never reached the server → relayed to Chief.
Log text: raw messages that included addresses and ids → scrubbed templates. All 38 existing error/warn logs were converted, and 13 that were the system working were downgraded or marked expected.

Forks and Loresis itself

Your next step: on each project's Health tab in Chief, click cover with logs for LoresisLinkedin, LoresisGithub, RestaurantRnD and LiveCEOEmailCalendar. Chief gives each project its own ingest token (Health tab → Coverage card), and each run goes through that project's own test and merge gate.

The runs can follow the porting guide in docs/implementation/chief-logging.md on this branch. The logger and the health checks copy over unchanged; only the one-file token, the route lists and each fork's own features differ. Doing upstream first lets the forks pick the shared files up when they merge upstream.

Verification

npm test: 372/372 passed, including 16 new tests (logger and health checks). eslint and tsc are clean; next build shows only the 9 warnings that were already there.
Production: two server-side deploys, ad2f57b then 24d7510. Each passed the in-image test gate 21/21, took a verified backup and started healthy. /healthz returns 200; the startup event shows shipping on, with the release stamped; there are no delivery errors.
Before each deploy: origin/main (9d16dff, the image production was running) was merged in, and nothing was in progress across the 45 analysis, checklist, request and draft status files.
Earlier: Chief accepted a live info test event (HTTP 202). In a real browser, thrown errors arrived scrubbed.

Not done or not verified

The other three forks and LiveCEOEmailCalendar aren't instrumented yet: they need their own Chief runs and tokens (see above).
No live breakage has been seen yet, because nothing is broken. The broken, reminder and recovery paths are covered by tests, not by a real outage.
Google Calendar is checked by status only, and the notetaker through its own sync and calls; there's no direct Recall ping.
Proxy refusals, the voice tool's input refusals and the local-only test runner aren't logged. A multi-word name written without quotes inside an error sentence can get past the scrubber.

Branch and commits

chief/38c18fdf1d (private repository)
b056b39 logger and entry points · 78041d1 feature instrumentation · 797988e docs · 5cee1b3 per-project token · 3be729a feature health · ad2f57b merge of main · 4f7d772 porting guide · 24d7510 per-feature summary