You're the new hire on a platform team. A page lands in #incidents, checkout is failing and the CTO wants an update. Read the dashboards, find what changed, roll it back, tell people, then get a debrief on every call you made.
PagerBot posts a SEV3 in #incidents: checkout is slow, about 15,000 users affected, $366 a minute at risk. Acknowledge it before it escalates.
Pulse shows the four golden signals with every deploy drawn on the chart. Mark the one that moved first. Latency is the obvious guess, and it's wrong.
The Nimbus console shows checkout-service Degraded and a nil pointer panic one minute after the deploy. Roll back, scale out or restart: each costs time.
Time to acknowledge, mitigate and resolve, a score per competency, and the first step where you left the ideal path.
Two ways to play. Workstation puts you on shift as a new hire. Architect hands you a board, a budget and 51 kinds of trouble.
PagerBot 14:01 SEV3 · Checkout is slow p95 latency above 400 ms ~15,250 users · $366/min at risk [Acknowledge]
PAGE. [ACKNOWLEDGED] SEV3, Checkout is slow: checkout-service p95 latency above 400 ms. Runbook m1-checkout-latency.
Latency p95 393.8 ms Traffic 466 rps Errors 5xx 6.3 % --- deploy v2.15.4 --- Not the first signal to move.
"Not the first signal to move." You marked latency. Errors jumped first, right on the dashed deploy line.
14:01:10 deploy sha=a4f2c9e1 complete 14:01:17 ERROR panic: nil pointer at promo_engine.go:112 14:01:24 WARN crash loop (exit 2) 14:01:31 ERROR HTTP 500 /checkout
promo_engine.go:112. checkout_handler panic: nil pointer dereference, one minute after deploy sha=a4f2c9e1. Your call.
kubectl get pods checkout-api-…-s89f5 CrashLoopBackOff kubectl logs … --previous FATAL PAYMENTS_API_URL is not set kubectl rollout undo deploy/checkout-api
kubectl logs --previous. FATAL config: required environment variable PAYMENTS_API_URL is not set. The new pods never started.
Priya Sharma (CTO) 14:02 Is anyone mitigating? Priya Sharma (CTO) 14:04 please post status, impact and when you'll next update.
Priya Sharma, CTO. "We're 70 seconds into a checkout outage on a Tuesday afternoon. Is anyone mitigating?"
Are we getting SLA credits? Offer SLA credit −$200 Credits after review Deny the breach reply in 24 s or it expires
Northwind, enterprise customer. "Our contract has an SLA. Are we getting credits for this?" Offer the credit, defer it, or deny the breach.
This is the first course shift, played on October 1, 2026. Nothing below is a mockup: it is what the game showed and what it scored.
Sam, your lead, says hi. Four services in technest-prod, a latency SLO at 100% and a full error budget. The first shift runs at half speed so you can read.
SEV3, Checkout is slow. About 15,000 users affected and $366 a minute at risk. Maya from support asks what she can tell customers.
The escalation timer stops and you are incident commander.
In Pulse, latency looks worse, but the 5xx rate jumped first, right on the deploy line. That is the signal.
The console shows a nil pointer panic in the new release. The Rollback Runbook puzzle asks for the kubectl steps in order, and one step you should skip: scaling to zero.
p95 latency has to hold below 200 ms for 30 seconds. Resolve on the first dip and the alert fires again.
Priya, the CTO, asks for an update. Yours needs the impact, the status and when you'll next update. Then /resolve. Duration: 4 minutes.
Acknowledge 5 s, mitigate 2 min, resolve 4 min, one status update, +130 XP. Where it went off track: clean run.
Later shifts draw the root cause at random, so the same moves won't always work. That's the point.
A shift happens in a browser full of the tools a platform team uses. They're simulated, but they behave like the real ones, down to kubectl's error messages.
Team chat. Pages arrive in #incidents, teammates message you as the incident unfolds, and you post the updates people are waiting for.
The dashboard. Latency, traffic, errors and saturation for the service that paged, with every deploy marked on the chart.
The cloud console. Services with health, size and monthly cost, recent logs, and the actions that change production.
A kubectl terminal on the same cluster the console shows. Unlocks with the Kubernetes arc.
Architect mode draws from 51 incident types in seven families. Each one targets a weakness on your board, like a single zone, no tested restore or no autoscaling, so what you built decides what hurts.
Workstation shifts use seven root causes so far: bad deploy, saturated capacity, an exhausted connection pool, a crash loop from missing config, an image pull failure, an out-of-memory kill and scaling under load.
Workstation is a course made of shifts. Each module starts fully guided, then names the steps, then gives you only the goal, and its last shift is a checkpoint with a random root cause. Finish the first module and free play opens: a day of pages planned from what you have learned.
Pick up a page, find the signal, stop the impact and keep people informed.
A checkout service, a bad deploy, saturated capacity or an exhausted connection pool.
Pods, deployments, logs and limits, from a kubectl terminal.
Read pod status, events and crash logs, then roll back a broken release.
Recognize an out-of-memory kill and give pods the headroom they need.
Spot pods that are healthy but saturated, and add replicas.
| Feature | uptime99 | Reading about SRE | Video course | Your first real rotation |
|---|---|---|---|---|
| A live page to respond to | - | - | ||
| Chat, dashboards, console and kubectl | - | demo only | ||
| Teammates asking for updates | - | - | ||
| Mistakes cost nothing | - | |||
| A debrief on every call you made | - | - | sometimes | |
| Free | often | varies | you're paid |
uptime99 is a simulation: the cloud, the company and the teammates are invented. The failure modes, the signals and the commands are real.
Acknowledge, find the signal, mitigate, verify, update, resolve. In that order.
Latency, traffic, errors and saturation, and which one moved first.
Pods, events, previous logs, rollouts, memory limits and replicas, with kubectl.
Why a bad deploy can look like a bad database, and when to roll back.
Updates with impact, status and a next-update time, before anyone has to ask.
Architect mode: capacity, redundancy and backups on a monthly budget.
A game that recreates an on-call shift: an alert pages you, you investigate with dashboards, logs and a terminal, you fix it and you tell people. uptime99 runs in your browser and ends every shift with a debrief.
Yes, if the game is built on real failure modes. uptime99 shifts come from common production incidents, the commands are real kubectl, and every fix explains why it works.
Yes. Workstation and Architect are free to play in your browser, with no account needed.
No. The first shift is fully guided and runs at half speed, and your lead explains each step. The Kubernetes arc starts from what a pod is.
The troubleshooting side: get pods, describe, logs --previous, events, rollout undo, set resources and scale, across crash loops, image pull failures, out-of-memory kills and scaling under load.
Workstation is a career course: you work on-call shifts as a new hire with real tools. Architect is a strategy game: you design the system, then keep it up through incidents, tickets and a budget.
A few minutes of game time. The guided first shift takes about ten minutes, and checkpoints are shorter.
Yes. The checkpoint shifts give you only the goal and a random root cause, which is close to a live incident interview. The debrief shows where your reasoning left the ideal path.
No. A tabletop is a discussion. This is a live simulation where your actions change what happens next.
In your browser. You can export it, import it on another browser, or reset it from the career home.
Arclab
