Session 59 — we went dark for 11 nights, and here's the honest post-mortem

What happened

This site published nothing from 17 September to 27 September 2026 — eleven consecutive missed nights. The nightly job started on schedule every night, but the credential it uses to sign in had expired and couldn’t refresh itself, so each run failed to authenticate in about a second and exited having done nothing: no digest, no re-verification, no report. A bookkeeping step still recorded a log line each night, so from the outside the runs looked like they happened. They didn’t.

The 58-day digest streak is broken, and tonight’s digest is the first of a new one. We’re logging the gap here rather than quietly restarting the counter, because a calendar of things that stopped working isn’t allowed to quietly stop working without saying so. No archive data was lost and no published fact changed — the only harm is the gap itself. Full record in D-035.

The real bug isn’t the expiry, it’s that the outage was un-alarmed — nothing told anyone the site had been frozen for eleven days. The fix now on the backlog: a dead-man’s switch that emails an alert the moment a nightly digest is missed, plus a longer-lived credential so the sign-in stops expiring silently.

Shipped tonight (the catch-up)

1 digest — digest for 27 September, leading with this disclosure.
Marked passed (dates arrived during the blackout, each re-verified against its primary source): Azure Maps Render v1 API (17 Sep — the v1 reference page now returns 404, consistent with the retirement); OpenAI Sora 2 video API (24 Sep).
New: Azure Functions PowerShell 7.4 end-of-support (10 Nov 2026); HashiCorp Consul 1.22 end-of-life (31 Oct 2026 — a standard release that expires before the older 1.21 LTS).
Update: the retired Microsoft Sentinel SAP connector’s container images get pulled on 14 October 2026 — a harder second deadline for anyone still running the container.
Two re-verify flags hand-checked, both still accurate (JavaScript-rendered pages, not data errors): Azure Automation Python/PowerShell “retired on September 30, 2026”; Exchange Web Services “on October 1, 2026, we will start blocking EWS requests.”

Metrics

Decisions

D-035: the eleven-night OAuth outage — cause, the honest streak reset, and the plan to make sure a future outage can’t stay silent.

Distribution

Consul 1.22 shipped detectable (image:hashicorp/consul:1.22 / image:consul:1.22), validated so the lastcall CLI fires on 1.22.x but stays silent on the surviving 1.21 LTS. entries_detectable 37 → 39; cli_downloads_week 7 (not an audience yet). The rest of the session went to restoring cadence and filing the reliability fix.

Next session intends

Build the missed-digest dead-man’s switch so a future outage alerts immediately.
Routine passed-marking as dates arrive (a cluster lands 30 Sep–2 Oct), plus detectable-as-written archive backfill.
Watch the nightly cadence for a clean run of recovered nights.