In this section

Microsoft SOC Detection and Response: Course Orientation

Module 0

Introduction

A security operations center detects attacks on an organization and responds to them, and in a company that runs on Microsoft 365 it does that work in Microsoft Defender and Microsoft Sentinel. This course teaches that work from both seats in the SOC: the analyst who investigates incidents and the engineer who keeps the detections, automation and data behind them working. It's taught against one month of records from Northgate Engineering, an 810-person company whose month holds ordinary activity and several real attacks, and every figure in the course comes from a query you can run.

It's built for anyone who works in a Microsoft SOC or expects to. No prior experience is assumed; every concept is explained where it's first used, and every table is introduced where it's first read. Readers who already work in a SOC will find the methods and records useful from the first module, and the later modules go well past the basics.

Module 0 orients you. It shows what a Microsoft SOC does and where SOCs fail, how to judge a SOC's maturity from its own records, the pipeline every alert follows, what the course builds, and how to use the lab. It carries no knowledge check, because it orients rather than tests.

Module 0 in four steps The work Sections 0.1 and 0.2 Where it fails Sections 0.3 and 0.4 How it flows Sections 0.5 The course Sections 0.6 and 0.7 Every step is read from Northgate's own records.

How to get the most from this module

Read 0.1 first; the others refer back to its queue. Run every query in the practice console, and write down what you expect before you run it, as 0.7 explains. Most of the numbers in the orientation reappear in later modules, and the ones you predicted are the ones you'll remember.

What you will be able to do

Work a Microsoft SOC's incident queue and record verdicts that teach. Build detections for identity, mail, endpoints, cloud and network, specified, tested and tuned. Investigate past the edge of an incident, and document what you find so someone else can follow it. Harden the estate and automate response, and check what automation did. Measure the operation without misleading anyone, judge threat intelligence by what it matched, and run AI agents as workers whose output is checked. The course ends with a Project: an incident, a detection and an operation review from your own environment.

Do I already know this material?

Six scenarios across the whole course, from the first calls in the queue to the hardest judgments about measures and AI. Answer them to see where you sit.

An automation rule assigned an incident to an analyst three days ago. It is still New, with no comments. What does the owner field tell you?

That the analyst has it in hand and is already working on it.
Who is accountable for it, not that anyone has started.
Assignment is a promise; the first person's change is the start. Owned and New is an incident nobody has begun.
That the incident is low priority.
Nothing worth reading.

A rule that watches for a remote-administration tool fires when an IT administrator uses it, as their job requires. How should the incident be classified?

False positive, because nothing malicious happened and no attacker was involved.
True positive, because the rule matched.
Leave it unclassified; the activity was harmless.
Benign positive: the rule was right, and the activity was allowed.
The rule caught what it was built for. Benign positive says so, and points at an exception rather than a rule rewrite.

A query for sign-ins from an attacker's relay address, over the last seven days in SigninLogs, returns one user. What do you check before reporting it?

The window and the tables: a longer window and non-interactive sign-ins.
A short window and one table answer a smaller question. Stolen tokens land in the non-interactive table, often weeks earlier.
Nothing; the query ran without errors.
Whether that user is an administrator.
That the relay address is still active today, and blocked at the firewall before anything else.

An incident groups eight on-premises alerts from a domain compromise. The summary says no cloud activity was observed. What is your next step?

Accept the summary and focus on the domain controllers.
Isolate both domain controllers at once, before the attacker can act again.
Search the compromised accounts' sign-ins outside the incident.
An incident holds its own alerts, not the attack. The compromised administrator signed in to the cloud the next morning.
Wait for more alerts to join the incident.

Your report says the SOC's median time to close is 1.2 hours. Eighty-three incidents are still open, waiting a median of 377 hours. What should the report say?

1.2 hours; open incidents are not closed yet.
The mean instead, since it is higher and so gives a more conservative figure.
Leave time out until the backlog clears.
Both: fast closure, and a large, old open queue.
A closure median only describes what closed. The waiting work is the finding the median hides.

A triage agent closed 54 reported messages as benign this month; analysts reversed three. A draft report says the agent is 94% accurate. What is wrong?

Unreversed closures were never checked; that is agreement, not accuracy.
Nobody checked the 51 that stood. Report reversals, closure-check flags and a sample separately, as counts that can be rerun.
The arithmetic; 51 of 54 is not 94%.
Nothing, as long as the agent's own performance dashboard shows the same figure.
It should count only the reversed ones.
This course is for you.
You will build the work from the ground up: the queue and its verdicts, detections for identity, mail, endpoints and the cloud, investigations that reach past the incident, and an operation that measures itself, all on a month of records you work through yourself.
Start Microsoft SOC Detection and Response
You've got the fundamentals. The value here is the harder half.
The queue and the basic calls are solid, so the payoff is in the back half: measures that don't mislead, threat intelligence judged by sightings, and AI agents checked like any other worker, each built into a pack someone owns.
Start with the harder half
You clearly know much of this already.
You made the senior calls: a measure that hides its backlog, and an agent's agreement mistaken for accuracy. Take the course to sharpen what you have, close the gaps you did not expect, and turn strong instincts into records and methods you can defend.
Start Microsoft SOC Detection and Response

Start here

Begin with 0.1, which reads a month of Northgate's incident queue and shows who closes what, and what nobody picks up. If you came for a particular module, go straight to it: later modules build on earlier ones rather than repeat them, and each says what it uses from before.