# Validation and execution limits

Validation date: October 29, 2026. The identities of the exact public files are in MANIFEST.json. These checks establish useful source-scoped hunt starters, not exploit-specific or production-proven detection coverage.

| Check actually performed | Result and scope |
| --- | --- |
| Final Sigma YAML | Five files loaded by PyYAML 6.0.3, compared with their SHA-256-bound fixture models, validated with JSON Schema draft 2020-12 against the Sigma rule specification V2.0.0 schema, and parsed by pySigma 1.5.1 |
| Real backend conversion | All five rules converted with pySigma Splunk backend 2.1.0 and Kusto backend 1.0.1; the Microsoft XDR Linux pipeline uses transform_parent_image=False. Emitted image, immediate-parent, command, and account fields were inspected and locally fixture-evaluated |
| CrowdStrike conversion | Backend 3.0.0 was exercised. The default pipeline rejected these rules because of existence or full-parent-path restrictions. Generic parent-basename conversion can also retain a Linux path slash. No default-converted Falcon query is shipped; hand-native CQL uses actual basenames and numeric UID |
| Event fixtures | 30 inert Linux cases per rule checked the exact final YAML-bound model and extracted predicates from all four event-query routes, including positives, negatives, known legitimate matches, missing fields, empty Image, source exclusions, case sensitivity, direct parents, and XML-escaped command text |
| ESXi fixtures | Five complete shell-prefix records checked each of D1 and D2-ESXi, including a positive, a nonmatching command, echoed command text, and a rejected unrelated layout. Splunk/Sumo extraction and imported VMware command predicates were evaluated locally |
| Sophos snapshot SQL | Five exact distributed SQL files each executed against 30 synthetic processes-table cases using SQLite. D6's known combined-short-option miss is explicit; this is not a Sophos agent execution |
| Catalog/source checks | Catalog paths exist; D2 distinguishes Linux and ESXi routes; D4 uses a 15-minute review window; Sumo XML fields are extracted once; relative documentation links resolve |
| Native vendor parsers, engines, tenant sources, and endpoint agents | NOT_RUN. Python re, ElementTree, and a bounded extraction evaluator are not Kusto, CQL, SPL, Sumo RE2, osquery, or vendor ingestion runtimes |
| Exploit, production, deployment, and monitor execution | NOT_RUN. Fixture command strings were treated as inert data |

Run the portable fixture checks from this directory with `python tests/run_tests.py` using Python 3. The runner requires only the standard library and verifies the canonical YAML hashes; changing a YAML rule requires updating its parsed model through a real YAML/schema review. The fixture model is test support, not a required customer event schema. The public runner extracts regex predicates from the shipped native files and uses actual SQLite SQL execution; it does not claim full native-language parsing or exact regex-engine equivalence.

The independent technical review uses separately authored cases and source documentation. Its evidence and reviewer records stay outside the download. Final release review, archive inspection, and publication authority belong to the enclosing release packet; this document does not claim they have happened.

Before enabling a route, follow the source guide, verify required fields on sanitized actual records, run an authorized bounded local query, and review legitimate matches against the application's baseline. Missing or truncated telemetry, different XML layouts, wrong identity semantics, or absent Linux inventory can reduce coverage. D5 observes a root-account child and cannot prove a privilege transition; Sophos uses current effective UID and can include namespace root. No tenant-specific verification or operational tuning is claimed here.
