When designing a game day, what changes if you run it unannounced — no advance notice to the on-call responders — versus announcing the scenario and time in advance, and which would you choose for a team that has never drilled before?
answer
- which half of the timeline you measure
- surprise buys detection and paging
- announcement buys ambition and teaching
- trust is the currency you spend
- earn no-notice, don't start there
basics
~20 sAn announced drill teaches the response and validates the plan with everyone available. Only an unannounced drill tests detection, paging and mobilization, because an expectant team is already watching. A team that has never drilled gets the announced version first.
solid answer
~50 sAnnouncing the drill turns it into a teaching exercise: the right people are present, the scenario can be ambitious, and you validate the plan and the tooling with maximum learning per hour. What you give up is the front of the timeline — if everyone is already staring at the dashboard, you cannot measure whether the alert fires, whether the page reaches a human, or how long mobilization takes. An unannounced drill is the only way to test that chain, which is also the chain that rots most quietly as people leave and rotations change. The cost is real: it burns trust if the team is not ready for it, it can collide with genuine load or a genuine incident, and a team that fails cold will remember the humiliation rather than the lesson. So you earn it. A team drilling for the first time gets an announced scenario; later you announce only the window; eventually you go no-notice, always with a small safety circle who knows.
go deeper
Understand the basic difference: an announced drill tells people it is coming, an unannounced one pages them cold, and only the second can measure how long it takes for a human to notice.
Explain which part of the timeline each format measures — detection and paging versus runbook execution and comms — and name the practical safety controls a surprise drill needs, such as drill-prefixed comms and a stop word.
Show the trust calculus: when a team has earned a no-notice drill, why a first-time team must not get one, and how you handle the drill that collides with a real incident or causes one. Bring a concrete escalation ladder.
Own the program-level policy: who authorizes surprise drills, what windows are frozen, how the organization ensures findings attach to systems rather than people, and how you weigh the reliability value against the on-call load you are adding.
## The two designs An **announced** game day tells participants the scenario, the time, or both. An **unannounced** (no-notice) drill fires the fault without warning: the responders find out the same way they would find out about a real incident, from a page. The choice is not about how serious the team is. It is about **which part of the incident timeline you want to measure**, weighed against how much trust you can afford to spend. ## The timeline argument An incident timeline has a front half and a back half: ``` fault -> first signal -> alert fires -> human acknowledges -> incident declared -> mitigation begins -> recovered ``` An announced drill collapses the front half. Everyone is watching, so detection is instant and mobilization is free. What it still measures honestly is the back half: whether the runbook is correct, whether the mitigation works, whether roles are clear once the team is assembled, whether comms happen. An unannounced drill is the only format that measures the front half truthfully: - Did any alert fire at all, or did the failure produce only a symptom nobody had a rule for? - How long from fault to page? - Did the page reach a phone that was on, and was the person actually available? - If the primary did not acknowledge, did escalation happen, and to whom? - How long from page to a human actually looking at the system? That chain is exactly the part that decays silently between incidents. Nobody notices that the escalation target is a disbanded team until they need it. ## What announcing buys you Announcing is not the lesser option, it is the option with a different goal: - **Ambition.** You can run a scenario that genuinely needs the database owner in the room. No-notice caps you at what the current shift can survive. - **Teaching.** New team members watch experienced ones work. That is a large part of a drill's value for a team building its bench. - **Safety.** You know the room is staffed, so a drill that goes sideways gets contained quickly. - **Fidelity of the scenario itself.** You can prepare state, seed data, and stage a realistic failure rather than being limited to whatever is safe to break unattended. ## What unannounced costs Be honest about the downsides, because an interviewer is listening for whether you have felt them: - **Trust.** A no-notice drill on an unprepared team reads as a test of individuals rather than of the system. If the team perceives it as a trap, you will not get honest findings from anyone again. - **Collision.** It can land during a real traffic peak, a real incident, or the one hour the only person who knows the subsystem is in surgery. - **Compounding.** The injected fault can interact with a genuine latent problem and produce a real outage — which is a legitimate finding, and also a real outage. - **Fatigue.** Adding surprise pages to a rotation that is already noisy makes the on-call experience worse, and the pager load problem is usually the more urgent one. ## The ladder that resolves it Most mature programs do not choose once; they escalate: 1. **Fully announced** — scenario and time known. The team learns the plan. 2. **Window-announced** — "sometime this quarter, during business hours, we will drill the payments dependency." Detection and paging get tested; nobody is ambushed. 3. **No-notice, scoped** — an unannounced page for a pre-agreed class of failure, inside business hours, with a facilitator standing by. 4. **No-notice, broad** — the full surprise, reserved for teams that have earned it and that have asked for it. A team that has never drilled starts at step 1, without exception. The goal of a first drill is to find gaps in a system, and you cannot do that while people are defending themselves. ## Safety rules that apply to every unannounced drill - **A small circle knows.** At minimum a facilitator, a safety owner with the authority to stop it, and whoever monitors the environment. - **Everything is labelled.** Every message in the incident channel is prefixed as a drill, so that a genuine incident arriving in parallel is unmistakable. - **A stop word exists**, is known to responders, and stopping is never treated as failure. - **A real incident preempts the drill immediately**, no debate. - **Business-critical windows are frozen** — peak sales periods, a major launch, a regulatory deadline. - **The people, not the person.** If the drill reveals that one engineer could not do something, the finding is that the system depends on one engineer, never that the engineer was inadequate. ## The interview answer Say which half of the timeline each format measures, say that unannounced is the only test of detection and paging, and then show the judgment: you earn no-notice drills, you never ambush a team that has not drilled before, and you keep a safety circle and a stop word regardless of how surprising the drill is supposed to be.
- What is the compromise between fully announced and fully no-notice, and what does it still test?Announce the window but not the moment: "we will drill the payments dependency sometime this quarter, during business hours." Responders cannot sit watching a dashboard for six weeks, so detection, alert firing, page delivery and mobilization are all measured honestly, while nobody is ambushed and the team can prepare by fixing the gaps they already know about. That preparation is a feature — a drill that motivates fixes before it runs has already paid for itself.
- An unannounced drill triggers a genuine customer-facing outage. What do you do, and how do you handle it afterwards?Stop the drill immediately, drop the fiction, and run it as a real incident with real comms — the drill's labelling exists precisely so this switch is unambiguous. Afterwards, write it up as a real incident postmortem, and treat the fact that the injected fault could cause customer impact as a first-class finding about blast radius and about the safety design of the exercise. Do not bury it to protect the program; a hidden drill-caused outage ends the program's credibility permanently.
- How do you stop an unannounced drill from being perceived as a performance test of individuals?Say out loud, before and after, that findings attach to the system: missing alert, stale runbook, single point of human knowledge, broken escalation target. Never name individuals in the writeup or in metrics. Have the facilitator help rather than score, let anyone stop the exercise without consequence, and run the first few drills with the team's own leads choosing the scenario, so the program is something the team does rather than something done to it.
saying these in an interview costs you the question
- Unannounced drills are the only real ones
- Announce everything, surprise is unprofessional
- A no-notice drill needs nobody informed at all
- The drill measures how good each responder is
- Run the surprise drill during peak traffic for realism