Skip to main content

Council-fix query validation — 2 November 2026

Result: passed (targeted query validation, not a campaign; rules disabled). A short live check of the PR #169 council round 1 content in a real Sentinel workspace (run bower-ws2d-20261102-065107, australiaeast, resource group rg-bower-ws2d-20261102-065107, table and DCR from deploy/bicep/main.bicep). Rows were uploaded directly to the DCR stream (method: direct); no collector, no AWS. Rule queries were run through the Log Analytics query API with now() replaced by the stated instant and maxDelay set to 120 minutes. No scheduled rule ran, so this is not evidence of scheduled alerting, the catch-up path in a real run, or the monitor.

CheckResult
Default deployment (maxIngestionDelayMinutes=120, PT205M look-back)succeeded: both functions and all three rules, so Sentinel accepted every rule query (deployment)
Second deployment beside it, detectionKeyScope=secondsucceeded: six rules side by side, the second set with [second] display names (output)
Pre-aggregated rules vs full ranking, both rules, 6 instants eachidentical output at every instant (results)
Fixed grid, crossing first ingested 06:59:17now() 07:05:00, 07:07:30 and 07:09:59 all give owned-window (window 06:55–07:00); 07:10:00 gives nothing (slot 1); 07:15:00 and 07:17:31 give catch-up-2 (results)
Below-threshold bucketnever returned by either rule (results)
BowerSecurityEventsOfType vs BowerSecurityEventsauthentication 17 of 17 and denials 43 of 43 identities with the same first copy and copy count; one identity differs, see below (result)
Event id reused with a changed type (ws2-q-below-deny-0000-b48f00b6f16e: denial first, authentication later)head typed function: 0 authentication rows and 2 denial copies, as the canonical function; the previous filter-first version: 1 authentication row and 1 denial copy (result, rows)

The one differing identity, ws2-q-cross-deny-0000-087d2939cde4, has two rows of different types with the same ingestion_time() (one upload batch), so arg_min has no single first copy and the two functions picked different rows. That is the exact-tie limit in the guide's known limits, not a collapse difference. The two typed rows differing from canonical counts in typed-reuse.json (18 and 44) come from a flawed check (a full outer join comparing nulls) and are superseded by typed-vs-canonical.json.

The first typed-function draft read ingestion_time() after its semi-join, where it is null; this run found it, and the shipped function reads it before the join (covered by a content test).

Cleanup​

cleanup.json: 9/9 absent (resource group, tagged resources, force-deleted workspace, no soft-deleted copy, application, service principal, both purged from deleted items, role assignments), each by an empty listing or an exact not-found code.