DECEPTION CHECK / SKYKICK / PORTABLE HUNT LOGIC / 2026-10-19

Purpose: carry the investigation intent into another SIEM or EDR without assuming
that field names, event categories, or query syntax transfer unchanged. The PDF
prints every supplied query. This file describes the base predicates and mappings.
These are hunt starters and proposed triage decisions, not installed alerts.

COMMON GATES
1. Confirm customer tenant/workspace/integration; endpoint queries also need device.
2. Confirm the necessary source and category with a known source event.
3. Select an explicit UTC window and record the end time and actual retention.
   Default incident start is 2026-10-13T00:00:00Z; broaden to October 1 if retained.
   KQL/endpoint code uses a half-open [start, end) range. Fusion/Data Lake use their
   UI selectors; verify actual selector behavior and boundaries on that platform.
   SPL supplies earliest/latest; record the exclusive end. CQL also needs the UI/API
   window and its lower bound. Verify record versus activity and ingestion times.
4. Retain source event ID, tenant, event time, ingest time, raw result, and provenance.
5. Preserve missing context. Do not convert unavailable telemetry into a clean hunt.

AVAILABILITY
KQL 01A/01B: for each tenant and UTC day, count sign-in/audit records and retain first
and last timestamps. They are separate queries, so an absent table is not hidden.
Fusion C00: counts per schema per selected day, not combined incident counts.
Data Lake L00: counts by query_name workload partition, not sign-in coverage.
Endpoint E01: counts by UTC day plus first/last IP-journal record on each endpoint.
None of these queries fills missing days with zero or proves uninterrupted collection.

AUTHENTICATION / INDICATOR PATHS
Let ListedApps be the seven public application client IDs printed in KQL 02 / L01.
Let ReportedIPs be the three public addresses printed in KQL 02 / L01.

KQL 02 base selection:
  in_scope(event)
  AND (case_insensitive_exact(event.AppId, ListedApps)
       OR exact(event.IPAddress, ReportedIPs))

Do not change OR to AND. Do not remove IPv6 or filter to successes prematurely.
Keep raw results; a success still does not prove data access or exfiltration.

SPL SP01 and CrowdStrike CQL CS01 implement the same app OR address intent over
the declared dc_* contract. Read skykick-cross-platform-mapping.txt and apply its
verified source mapping first. These are templates, not native connector fields.

Fusion C01: reported source_address within auth/cloudaudit sources.
Fusion C02: any listed GUID appears as a substring in original event content.
These are separate searches; C02 is not equivalent to an exact AppId-field match.
Resolve whether the matching GUID is actor, target, resource, or a property value.

Data Lake L01: query_name = AzureActiveDirectory AND (lower(application_id) in the
listed apps OR client_ip in ReportedIPs OR actor_ip_address in ReportedIPs).
This is Microsoft 365 directory audit, not the dedicated sign-in table. Exact IP
strings can miss IP:port or alternate forms. Do not silently strip data without
documenting and testing the normalization.

ACCESS CHANGES
KQL 03: selected broad directory-audit categories OR credential/certificate/consent/
owner/role operation-name fallbacks, across all apps and actors. Retain outcomes.
Expand all target objects while preserving empty-target parent events. Keep nested
modifiedProperties with the correct target. Expanded targets are not new incidents.

Fusion C03: raw-content keyword matches for credentials, roles, owners, consent, and
app/service-principal changes. Data Lake L02: operation and modified_properties text
keywords in AzureActiveDirectory audit. These two keyword implementations do not
provide the same category coverage or structured parsing as KQL 03.

Sigma S01/S02: narrower application-actor event predicates, detailed in
skykick-base-logic.txt. They supplement the broad queries; they are not conversions.
SPL SP02 and CQL CS02 retain the broad category/operation selection across all
applications and actors. Entire nested actor/target JSON remains on the parent row;
unlike KQL 03, these adapters do not expand targets. Inspect every target explicitly.

VERIFIED CREDENTIAL / IDENTITY PIVOT
KQL 04: exact case-insensitive match of verified tenant ID, application client ID,
and credential key ID in the sign-in table. Replace all three placeholders.
Fusion C04: substring match for one verified nonempty identity/key ID in raw events.
Do not treat it as an automatic join; establish identity type, tenant, and field.
Never use a secret value. Missing key metadata or token reuse can leave a pivot empty.
Compare successful/failed outcomes and the tenant's recorded containment timestamp.
SPL SP03 and CQL CS03 use the verified tenant/app/key tuple in mapped sign-in records.
Their named query files retain all outcomes and need no secret values. Check UI and
export limits before treating a projected result table as the complete timeline.

NETWORK + PROCESS CONTEXT
Endpoint E02: source OR destination equals a reported IP in the IP journal window.
No inner join to process data, so a missing process record does not erase a hit.
Endpoint E03: same endpoint AND verified sophos_pid, with explicit retained-history
time bounds. This is not a numeric Windows PID. Narrow deliberately because the
process identifier is not indexed; inspect execution errors and truncation.
Fusion N01: reported address in logical @ip within netflow, with parser-dependent
source/destination/NAT mappings. Inspect direction and outcome on the original record.
These paths cannot see traffic that never crossed a monitored device or network.

TRIAGE / PROPOSED DECISIONS, NOT EXECUTED CORRELATION RULES
A. Successful, unapproved authentication from a reported address merits escalation.
   Keep attempts separate. Review unlisted addresses for implicated identities too.
B. Successful, unapproved credential/role/owner/consent changes merit escalation.
   Check the target object and approval evidence; an app actor does not prove abuse.
C. Confirmed unauthorized credential use or related resource activity after that
   tenant's recorded containment time merits escalation. Distinguish event time from
   late ingestion. Successful authentication is not itself proof of resource access.

For recurring detection, implement tenant-scoped deduplication by source event ID,
appropriate late-arrival handling, a tested result mapping, change context, and an
assigned response owner. Validate against positive, negative, absent-field, and
missing-source cases in the destination environment. The package does not implement
these scheduling, correlation, or response controls.

Continue into collected Graph and Microsoft 365 resource records. The supplied
queries do not cover every mailbox, file, token, or policy action. An EDR installation
alone does not imply collection of cloud-identity and workload logs.

Public indicators: https://www.connectwise.com/company/trust/advisories
Source schemas, validation boundaries, and language-specific instructions are in
the companion PDF. Do not contact indicator infrastructure during validation.
