# Validation scope

| Check | Status |
|---|---|
| Local Python policy comparison and CSV handling | PASS on the supplied synthetic cases; author fixture run |
| Native-query regex/equality/path primitives against synthetic fixtures | PASS in Python only; 14 auth fixture records and 9 process-image cases |
| Sigma text/source sanity | Included in the local tests; not a Sigma parser or backend conversion |
| Sigma parser and backend conversion | NOT_RUN |
| Microsoft Kusto parser or runtime | NOT_RUN |
| Elasticsearch ES\|QL runtime or Elastic Security rule creation | NOT_RUN |
| Splunk, Sumo, and Falcon vendor parser/runtime | NOT_RUN |
| Real ingestion, sensor coverage, tenant isolation, alerts, and operational precision | NOT_RUN |

`tests/test_companion.py` is safe to run locally with Python 3.10 or newer and its standard library. It reads the supplied queries and synthetic fixtures, compares regex/equality/path matching primitives with explicit expected positives/negatives, and exercises event-time policy/identity/gap behavior. It does not implement the complete vendor query languages. Successful local tests cannot establish vendor syntax acceptance, ingestion correctness, query scheduling, or effectiveness on customer traffic.

The recorded author run completed all 23 unittest methods successfully. The example CLI comparison also returned exactly one policy conflict, one password-allowed outcome, and one missing-policy gap. These are author checks, not a claim of independent review or production readiness.

The tests include ordinary password acceptance, publickey/failure negatives, IPv6, absent fields, wrong program, unsupported raw-prefix shape, invalid ports, and quoted examples. Policy cases include host/account/scope mismatch, case sensitivity, interval boundaries, stale/future state, UTC offsets, unknown/missing/overlapping/invalid policy, empty exports, malformed CSV, and duplicate event identity. The separate process filter includes temporary paths and nonmatches such as `/tmp2`, `/usr/bin/curl`, and `/bin/sh`.

## Before relying on a hunt

Use an isolated authorized source or existing safe historical records. Record the exact query hash, product/engine/integration/sensor versions, source filters, organization/host/time scope, and expected source events. Run positive, negative, and missing-field cases through the real ingestion and query path. Check the original event against extracted account/method/host/time values, including relay behavior and timezone parsing.

For the policy comparison, an asset owner must validate the scoped asset/account identity and historical policy evidence. Test authorized changes, expired and overlapping intervals, clock skew, renamed/recreated accounts, and repeated exports. Assign unresolved gaps an owner. No event rows or matches do not establish collection health; compare expected hosts/source delivery with actual ingestion separately.

Check query/export caps, deduplication, source clock quality, retained windows, search performance, access controls, representative benign logins, and temporary-program false positives. The candidate selectors are investigative starting points; routine password acceptance and temporary execution are not stand-alone maliciousness criteria. Alert scheduling, severity, suppression, and response routing need separate receiving-environment decisions. No automation or alert is installed by these files.
