CompTIA CS0-003: Detection Engineering from Rule to Signal
Detection engineering is the discipline of turning an adversary hypothesis into a reliable operational signal. A rule is only one implementation artifact. Mature detection engineering starts with the behavior to detect, identifies the telemetry that can prove it, writes analytics that survive normal variation, tests them against benign and malicious examples, tunes the resulting alerts, and measures whether the signal helps analysts make a better decision.
On this page
MITRE ATT&CK’s current framework has moved toward detection strategies and platform-specific analytics rather than treating high-level data-source labels as the endpoint. ATT&CK v18 deprecated the old data-source catalog while retaining data components and adding detection strategies that organize multiple analytics into cohesive ways to detect adversary techniques. That evolution matches the operational lesson: useful detection is more than a one-line query.
Effective security operations measure whether detections reveal actionable behavior. Rule counts alone do not show coverage, and a low alert volume means little unless analysts can connect each signal to a threat hypothesis and an investigation path.
Exam edition context: Detection engineering remains relevant across exam editions. CompTIA launched CySA+ V4 (CS0-004) in June 2026; readers using this older CS0-003-titled page for certification study should check the current exam objectives rather than infer scope from the URL.
Begin with a threat hypothesis
State what an adversary is expected to do and why the organization cares. Detection lifecycle works best when engineers can explain the behavior before they choose SIEM syntax. A hypothesis such as “a newly compromised cloud identity rapidly enumerates services before modifying resources” is more useful than “alert on List APIs.” Detection logic should also have an explicit scope statement. A Windows process analytic, a cloud API analytic, and a SaaS identity analytic may all map to the same ATT&CK technique while relying on different evidence. Platform-specific scope prevents analysts from assuming a rule covers environments it never observes.
Map the telemetry that can prove it
Identify the exact event sources, fields, timestamps, identities, process metadata, network records, or cloud APIs needed to observe the behavior. Cloud detection requires cloud context because the same API can be routine automation for one role and suspicious discovery for another. Do not deploy a rule until the required telemetry is actually enabled and retained long enough to support investigation. Telemetry health belongs in detection engineering. A rule that never fires may indicate low adversary activity, but it may also mean an endpoint sensor stopped reporting or a cloud audit trail was disabled. Critical detections should have companion checks for required log sources so silence does not create false confidence.
Detection coverage maps should show both techniques and data dependencies. A technique may appear “covered” because one rule exists, while the required telemetry is missing from half the environment. Track platform, log source, deployment percentage, and analytic health so coverage reflects reality rather than documentation.
Use ATT&CK as a behavior map
ATT&CK detection strategies and analytics can help defenders see which techniques require multi-event correlation or platform-specific evidence. Current ATT&CK analytics often include mutable elements such as time windows, allowlists, thresholds, or environment-specific monitored paths. Those mutable elements are where local engineering begins; a copied analytic is a starting point, not a finished production rule. Detection engineering should also integrate with threat intelligence and incident response. A newly observed adversary technique can create a hypothesis; a confirmed incident can produce new test cases; repeated false positives can reveal telemetry weaknesses. The lifecycle is healthiest when those inputs flow back into the analytic rather than living in separate SOC processes.
MITRE ATT&CK deprecated its older Data Source objects with ATT&CK v18 in October 2025 in favor of Detection Strategies and related log-source descriptions. When mapping a threat technique to required telemetry, use the current ATT&CK detection guidance rather than assuming older data-source identifiers are maintained indefinitely.
Build the rule around discriminating context
High-signal detections combine the suspicious behavior with context that separates it from ordinary administration. Useful context can include new identity, unusual geography, process ancestry, resource criticality, timing, frequency, device posture, or follow-on actions. One broad indicator with no context often creates a queue of technically correct but operationally useless alerts. Detection engineering should feed prevention. If the same behavior is repeatedly detected and confirmed, the organization may be able to block the vulnerable path, narrow privilege, or redesign the system so the SOC no longer has to catch the same problem forever.
Mutable elements should be reviewed as configuration rather than buried constants. Time windows, thresholds, allowlists, monitored paths, and known service accounts can vary across business units and change over time. Store those values where they can be reviewed and updated without rewriting the detection’s core hypothesis. Correlation should be used where single events are too common. A service-enumeration API call may be benign, while rapid enumeration by a new identity followed by snapshot creation and role changes is much more suspicious. Behavioral chains usually create better signal than increasingly complicated single-event pattern matching.
Test with positive and negative cases
Generate or replay known malicious behavior where safe and include representative benign workflows. The detection should fire on the intended technique and remain quiet on expected administrative behavior. Keep test data versioned so a future query refactor can prove it preserved the original behavior. Detection validation should be repeated after major platform upgrades and logging changes. A query that worked against one endpoint agent, cloud schema, or SIEM parser can silently lose fields after migration. Maintaining test fixtures and telemetry-health checks keeps the detection estate aligned with the environment it is supposed to protect.
Worked example (hypothetical): to detect an unfamiliar service identity granting itself a privileged cloud role, test one authorized deployment, one malicious role-grant sequence, and one benign identity administration event. Confirm that the analytic captures actor, target, old and new privileges, source IP, and surrounding activity; an alert with only a rule name gives the responder no decision-quality context.
Tune without erasing the threat
False positives should lead to better logic, not a growing list of broad exclusions. SIEM tuning should identify which field or context separates expected activity from attack behavior. If an allowlist removes an entire administrator group, the detection may become blind exactly where credential theft creates the greatest impact. When a detection becomes noisy, inspect the business workflow before adding exclusions. New automation, deployment tools, or cloud-management services can create legitimate behavior that resembles attack techniques. Updating the analytic to recognize expected context is safer than suppressing all activity from the affected team.
Analysts should be able to suppress a known incident without changing the global detection logic. Case-level suppression, temporary maintenance windows, and scoped exceptions are safer than editing the rule every time a legitimate operation creates noise. Separate operational exception from permanent tuning.
Design the analyst output
An alert should include the entities, timeline, evidence, severity rationale, and next investigative question an analyst needs. Detection engineering is partly user-experience design for the SOC. A perfect query with an opaque alert description still wastes analyst time because the responder must reverse-engineer the logic under pressure. Ultimately, signal quality is a product outcome for analysts. The rule should deliver enough evidence, context, and confidence that the responder can decide what to investigate next. A detection that technically maps to ATT&CK but does not change an analyst decision is not finished engineering.
Detections should have an expected investigation path. If a rule fires but no one knows which logs to inspect, which entity to isolate, or which owner to contact, the alert is incomplete. Embed the next question and evidence links into the alert metadata where the platform allows it.
Measure signal quality
Useful metrics include true-positive rate, precision, coverage, time to triage, recurrence, alert volume, and whether the detection discovered activity other controls missed. Do not optimize only for lower alert count; a rule can become quiet because it was tuned into uselessness. Measure whether the signal changes an investigation or containment decision. Over time, good detection engineering reduces rule count in some areas. Correlated strategies, better telemetry, and stronger prevention can replace several brittle signatures. Maturity is not the number of analytics deployed; it is how reliably the organization turns adversary behavior into timely, actionable signal.
Operate detections as code
Store query logic, tests, metadata, owner, severity, ATT&CK mapping, required telemetry, and tuning notes in version control where practical. Security operations becomes more repeatable when rule changes are reviewed and promoted instead of edited directly in production without history. For CompTIA CySA+ and CompTIA SecurityX practitioners, the durable workflow is hypothesis → telemetry → analytic → test → deploy → triage → tune → measure → retire. The rule is just the compiled form of that lifecycle. Detection retirement matters too. When a platform changes, telemetry disappears, or another control supersedes a rule, remove or redesign the detection rather than leaving a silent broken analytic in the catalog.
Detection ownership should survive staff turnover. Every production analytic should have a team, review date, data dependencies, escalation path, and retirement criteria. Ownerless rules accumulate exclusions and syntax changes until nobody knows whether they still represent the original threat hypothesis. Keep detection documentation concise but complete: threat hypothesis, data requirements, query, severity, owner, test cases, expected false positives, response guidance, and tuning history. That metadata turns a query into an operational security control another engineer can maintain without relying on the original author’s memory.