Portable detections depend on consistent evidence, but Windows, Linux, cloud platforms, Splunk, Microsoft Sentinel, and Elastic often describe the same activity with different field names, types, and nesting. This SIEM field mapping guide provides a reusable normalization checklist, comparison table, benign validation records, and troubleshooting workflow for Sigma and native detection queries.
Overview
Field mapping is the process of translating source telemetry into a stable schema that detections can use. A Windows process event might expose Image and ParentImage in one pipeline, process.executable and process.parent.executable in another, and nested JSON under a vendor-specific object in a third. If a rule assumes only one representation, it may work in a lab and fail after a connector change, parser upgrade, or SIEM migration.
Normalization does not mean discarding the original event. Keep the raw record for investigation and preserve the source-specific fields needed for troubleshooting. The normalized layer should provide predictable names, types, casing, and semantics for common detection logic. A practical schema should answer questions such as:
- When did the event occur, and in which timezone?
- Which user, host, process, parent process, IP address, or cloud identity was involved?
- Is a value a string, number, Boolean, list, or object?
- Does an absent field mean that the activity did not occur, or that the sensor did not collect it?
- Can the same Sigma or analytic logic be translated without changing its meaning?
Use the normalized fields for portable rules, but document the source, parser, and transformation that produced each value. A field dictionary is most useful when it includes ownership, expected type, example values, and known gaps.
Core mapping reference
| Evidence | Normalized field | Common source or vendor variations | Validation concern |
|---|---|---|---|
| Event time | @timestamp | _time in Splunk; TimeGenerated in Sentinel; @timestamp in Elastic | Confirm event time versus ingestion time and preserve the original timezone. |
| Username | user.name | User, AccountName, SubjectUserName, user | Separate the account name from a full identity, email address, or SID. |
| Host | host.name | Computer, ComputerName, dest, DeviceName | Decide whether the value is a short name, FQDN, asset ID, or device display name. |
| Process name | process.name | Image, NewProcessName, FileName, process.executable | Do not confuse a basename with a full path. |
| Process path | process.executable | Image, CommandLine components, process_path | Normalize separators and document whether symbolic or short paths are retained. |
| Parent process | process.parent.name and process.parent.executable | ParentImage, ParentProcessName, nested parent objects | Parent fields may be absent or populated by a different sensor event. |
| Command line | process.command_line | CommandLine, ProcessCommandLine, process.args | Arguments may be a single string or an ordered array; preserve quoting where possible. |
| Source IP | source.ip | src, src_ip, IpAddress | Validate IPv4, IPv6, proxy, and NAT handling. |
| Destination IP | destination.ip | dest, dest_ip, RemoteAddress | Do not infer a destination from a hostname without documenting the enrichment. |
| Port | source.port or destination.port | src_port, dest_port, RemotePort | Keep ports numeric and do not compare numeric values to quoted strings. |
| File hash | file.hash.sha256, with algorithm recorded | sha256, SHA256, FileHash | Lowercase or uppercase consistently, and distinguish hash algorithms. |
| Cloud identity | cloud.account.id and user.name | UserId, AccountId, PrincipalId, provider-specific identity objects | Separate actor, assumed role, session name, tenant, subscription, and account. |
| Authentication outcome | event.outcome | success, failure, Status, ResultType | Map vendor codes explicitly; do not assume every nonzero code means failure. |
This table is a starting point rather than a universal schema. Confirm field behavior against the actual connector, parser, data model, and event samples in your environment.
Checklist by scenario
Windows process and authentication telemetry
- Record the source event identifier and sensor that produced the event.
- Map the host, user, process, parent process, command line, integrity level, and process identifiers separately.
- Preserve both the process basename and full executable path.
- Test absent parent fields, service accounts, machine accounts, localized usernames, and commands containing spaces or escaped quotes.
- Compare authentication outcome, logon type, source address, target account, and authentication package without collapsing useful distinctions.
For a controlled validation, use a benign process such as a standard system utility launched with a harmless argument, then confirm that the raw event and normalized record retain the expected parent-child relationship. Pair this work with a Sysmon event reference and validation guide and the Windows audit policy baseline.
Linux process, privilege, and network events
- Map executable, command line, effective user, real user, parent process, process ID, and working directory independently.
- Represent privilege changes with explicit outcome and actor fields rather than a single “privileged” flag.
- Normalize source and destination addresses while retaining interface, namespace, and container context when available.
- Check whether audit records arrive as multiple related records that require correlation.
A benign record should include a known executable, a known user, a predictable parent, and a test connection to an approved lab endpoint. The Linux auditd detection lab can help validate process, privilege, file, and network evidence without treating a single parser field as complete coverage.
Cloud identity and access events
- Separate the human or workload actor from the role, service principal, session, or delegated identity used to perform the action.
- Map account, tenant, project, subscription, region, resource, and API operation into distinct fields.
- Normalize success and failure outcomes from provider-specific status codes.
- Retain request identifiers and source network information for investigation.
- Test role changes, access-key use, and administrative actions only in an authorized lab or test account.
Cloud identity records often contain nested objects and arrays. Flatten only the values required by detection logic, and retain the original object so analysts can recover context. See the cloud IAM detection lab for a safe validation pattern.
Porting Sigma and native SIEM rules
Start with a semantic detection statement, such as “a process launched from an unusual parent,” before translating it into Sigma, Splunk SPL, Sentinel KQL, or Elastic Query Language. Map each logical condition to normalized fields, then identify any platform-specific implementation needed for joins, time windows, arrays, or case handling.
For every port, record the source field, normalized field, target query field, transformation, and test event. A conversion that produces syntactically valid queries can still be semantically wrong if a nested array is treated as a string or if a timestamp is interpreted in local time. Use a repeatable regression process described in the Sigma-to-SIEM detection rule testing workflow.
What to double-check
Before approving a mapping, inspect a small set of benign records rather than relying only on a schema screen. Include normal success and failure events, missing optional fields, multiple values, unusual casing, IPv4 and IPv6, long command lines, and events generated near a daylight-saving or timezone boundary.
- Types: Verify that ports, process IDs, status codes, and timestamps have the expected types. A numeric field arriving as text can silently break comparisons.
- Casing: Decide whether usernames, hostnames, hashes, and event codes are case-sensitive. Normalize only where doing so does not destroy forensic value.
- Time: Compare event time, collection time, and ingestion time. Confirm the query language's default timezone and window behavior.
- Nested values: Check whether an identity, command argument, or hash is a scalar, list, or object. Test both one-value and multi-value records.
- Missing data: Distinguish null, empty string, absent key, and “unknown.” These states may behave differently in each SIEM.
- Cardinality: Confirm that repeated fields are not overwritten during parsing. This matters for DNS answers, cloud resources, authentication methods, and process arguments.
- Correlation: Test whether related records share a stable host, process, session, request, or event identifier.
Maintain a small fixture set of benign records with expected normalized output. It becomes a practical contract between the collection pipeline and detection authors.
Common mistakes
- Mapping by label alone: Two fields named “user” may represent the initiating account and the target account. Define the actor and subject explicitly.
- Using vendor aliases as the canonical schema: A field such as
destcan mean a host in one data source and an IP address in another. Prefer unambiguous normalized names. - Dropping raw evidence: Normalization should add a stable layer, not replace the original event.
- Ignoring parser version: The same connector can produce different nesting or types after a parser update. Store pipeline and parser metadata with test results.
- Assuming Sigma conversion is lossless: Some back ends handle wildcards, lists, case sensitivity, and aggregations differently. Review generated queries and test them against known records.
- Testing only positive matches: A detection must also avoid matching ordinary administrative activity, expected service behavior, and incomplete records.
- Over-normalizing identity: Converting every cloud principal into a display name can remove the stable ID needed for correlation.
When to revisit
Review field mappings before a SIEM migration, new connector deployment, parser or agent upgrade, data-model redesign, or major detection-rule port. Also revisit them when analysts report missing context, alert volume changes unexpectedly, a query begins returning no results, or a source adds a new event version.
At each review, rerun the benign fixture set, compare raw and normalized records, and execute representative Sigma and native queries in a controlled validation window. Record changes to field names, types, timezone behavior, nested structures, and known coverage gaps. Recheck mappings during seasonal planning cycles when detection priorities and telemetry sources are reviewed, and whenever workflows or tools change.
A useful final checklist is: source identified, raw event retained, canonical field documented, type verified, timezone confirmed, null behavior tested, benign positive and negative cases checked, query translation reviewed, and owner assigned. Completing those steps makes portable detections easier to maintain and gives SOC teams clearer evidence when a rule fails because of logic, telemetry, or parsing.