Glossary
The terms the game uses without stopping to define them.
Last updated:
Most of these are real operations vocabulary, used the way an operations team would use it. Where the game narrows a term, that is noted.
Metrics and state
Uptime — a composite 0–100 measure of service quality, averaged across the whole shift. Not a per-machine reading and not a current value. See resources.
Latency — average response time in milliseconds. When it is above target the log says so, and uptime follows it down.
Bandwidth — incoming requests per second measured against fleet capacity.
Only web and lb machines contribute capacity.
Saturation — a metric sitting at or near its ceiling. CPU saturation is the most common phrasing in the logs.
OK / WARN / CRIT / DOWN — server status, derived from whichever
metric is worst, except DOWN which is set explicitly and is sticky.
RSS — resident set size, the memory a process actually holds. A growing RSS with no plateau is the signature of a leak.
OOM killer — the kernel component that starts terminating processes when a machine runs out of memory. It chooses; you do not.
Thermal shutdown — a machine powering itself off to avoid damage, at 95 °C. It stays down for a full minute.
Incidents and timing
Signature — the log pattern that identifies an incident. The thing the game is actually teaching you to read.
Impact curve — what an incident does over time if nothing intervenes.
Grace window — the period after an incident appears during which it costs nothing extra. 60 seconds on night one, 22 on night ten.
SLA debt — uptime drained directly by an incident left open past its grace window: 0.1 points per step, capped at 45. See escalation.
Escalation — an incident passing its grace window and beginning to accrue SLA debt. It also stops occupying a concurrency slot.
Concurrency limit — how many incidents can be active at once, from 1 on early nights to 4 from night nine.
Mercy window — a 45-second hold on new injections when uptime is below 70%. 20 seconds during the finale.
Compound incident — two or more incidents running together, from night six, often where one caused the other.
Tick — one simulation step: 250 ms of real time, 12 seconds of game time. A shift is 2400 of them.
Networking and infrastructure
CIDR — address block notation, as in 203.0.113.0/24. The game accepts
blocks wherever it accepts addresses.
RFC 5737 — the standard reserving 192.0.2.0/24, 198.51.100.0/24 and
203.0.113.0/24 for documentation. Every address in the game comes from these
blocks, so none of them is a real host.
Rate limiting — capping request rate at the edge. It cannot distinguish attackers from customers, which is why it trims 10% of legitimate traffic too.
Draining — moving traffic off a machine without stopping it. The load goes somewhere, and that somewhere now carries it.
Lock wait — a database query blocked waiting for another to release a lock.
Visible only through db_stat.
Seed — the number the night's pseudo-random generator starts from. Same seed, same night, every device. See determinism.