First DR call

The Walkthrough

A narrated overview of the response, ending with a live demonstration on alphanet. Not a drill: the scenario is announced, nothing is timed, nobody is assessed. Questions at every step, not held to the end.

~75 minuteseverything posted carries [DRILL]parked questions go to the task list

Run of show

Segment	Min	What happens
Open	5	What this call is and is not. Cold open on the 2026-08-01 incident: everything here exists because something like it already happened.
The seats	10	Go around the named seats; each person confirms which page is theirs. Expected question: "what if the named person is asleep" - the acknowledgement clocks answer it.
Walk the board	20	The nine board steps, top to bottom. Per row: what happens, who moves, which channel. Pause at steps 2, 8, and 9 - the operator-facing ones.
Demonstration	20	Crash alphanet, watch the real signal path fire. The beats below.
The asks	10	Contacts, consent to be paged, drill readiness - put a name and a deadline on collecting the answers.
Open floor	10	Parked questions. Set the first tabletop date while everyone is present.

The demonstration - who is alerted, and where

Denis narrates every beat with the same sentence shape: who is alerted, and which channel it goes to. Nothing here is a test of anyone.

Show alphanet healthy
Ledgers advancing on the dashboard, all validators proposing. Alphanet is our test network: real validators, nothing at stake.
Fire the fault
Enough validators stop to lose quorum, and ledgers stop advancing. This is a page-tier condition: no validated ledger for 90 seconds, and it is correlated, so it pages rather than warns.
The alert lands in the Slack channel
Who is alerted: a correlated pattern pages every seat at once - Denis, Brett, Ayo, Vijay, the org on-call engineer, Vet. One validator's alert would route to its own operator and nobody else. True today: the pager's enrolled population is one, so the page rings one phone - closing that is one of the asks.

PAUSE FOR QUESTIONS

Declare
Denis posts the declaration, prefixed [DRILL], in #unl-validators, and walks its fields: condition off the seven-word list, severity, Cause: not yet established. Unknown is a complete condition and the normal one at minute zero. The naming post lands in #dunl-operators-only; Vet mirrors to #disaster-recovery-news.
Open the bridge
Everyone on the call clicks in once, so every person has used the link before the night it matters.
Give the two orders
One message: Vijay builds the damage report, due in 15 minutes; Ayo finds the root cause. The trap, said out loud: the damage track is the one the response is timed off.
Pin the instruction
Brett pins a [DRILL] restart instruction in #dunl-operators-only; Denis countersigns in the thread - two names, or nobody acts. Anyone running an alphanet node posts done in the thread.

PAUSE FOR QUESTIONS

Recover
The validators come back, ledgers advance on the dashboard, Brett posts the count. Narrate the ending: unwind at de-escalation, the 24-hour clean window, the four stand-down conditions, Vet's stand-down notice on #disaster-recovery-news.

PAUSE FOR QUESTIONS

After the call

Findings route the same way a tabletop's do: anything the walkthrough exposed as unclear becomes a task or a fix to the page that failed, the same day. Each person's follow-up read is their one role page, nothing else.

The full run sheet, including the pre-call staging checklist: drills/walkthrough-call.md