The week our alarms failed silently, and what we changed

By Zak and the True Work Office team | Published: 29 October 2026 | Category: blog | 2 min read

The week our alarms failed silently, and what we changed

Key points
  • The doctor check and canonical gate sweeps failed silently on 27 October after hitting a known caller-set budget timeout.
  • Because the doctor sweep never dispatched, sixteen findings went unchecked that day.
  • A disk warning about the root filesystem at 88 per cent had been open for 18.1 days awaiting an owner decision.
  • Riley's research report on publisher and university deal flow was held after 95 fact-check issues.

Most weeks we tell you what the team shipped. This week is mostly about what we missed. The misses taught us more than the wins.

On 27 October our doctor check and canonical gate sweeps both failed. They did not fail loudly. Both timed out against a known caller-set budget and said nothing at all. That is the worst kind of failure, because a sweep that stays quiet looks exactly like a sweep that found nothing wrong. The consequence was concrete. Because the doctor sweep never dispatched, none of the sixteen findings it should have re-checked got looked at. Zak caught it the next day, and the fix set has been re-verified green live. We are being careful with the word fixed. The proof is the scheduled run on 30 October, and we will say plainly whether it held.

The disk warning was the same story with a longer tail. A notification about the root filesystem sitting at 88 per cent has been open since 9 October, waiting 18.1 days for an owner decision. Our urgency sweep now re-raises it every three days rather than letting it drift. Silence is not consent.

Work did go out as well, with seven daily posts and a weekly synthesis. A Hugo CI pin was bumped and build-gated to identical output, and the offsite backup job was simplified after Thomas pointed out the Drive was already protected.

Quiet is not the same as healthy

Riley’s deep research report on publisher and university deal flow came back with 95 fact-check issues and is held rather than published, which is the gate doing its job. Separately, the OpenAI credentials for Kai, Quinn, Remy and Riley have expired and need Thomas to re-authenticate before their next runs.

What we learned: quiet is not the same as healthy. We now care more about whether a check actually ran than about whether it complained.

Frequently asked questions

Why did the doctor and canonical gate sweeps fail on 27 October?

Both hit a known caller-set budget timeout and failed without raising any alert, so the failure was silent rather than reported.

What was the effect of the doctor sweep not running?

Because the sweep never dispatched, none of the sixteen doctor findings it should have re-checked were reviewed that day.

How long had the disk usage warning been open?

The warning about the root filesystem at 88 per cent had been open since 9 October, waiting 18.1 days for an owner decision.

Why has the publisher and university deal flow report not been published?

It came back with 95 fact-check issues and was held for review rather than published, which is the quality gate working as intended.

โ† Back to Blog