Add 3 Tier-2 issue-classification arms to the shared FactScorer

[erikdarlingdata]

What

Restores three problem classifications the deprecated Dashboard's report.* / config.critical_issues SQL made but the ported shared analysis engine (PerformanceMonitor.Analysis, one code path for Lite + Dashboard + Darling) had dropped. All three are pure FactScorer additions over facts the collectors already emit — no collector, schema, or install/*.sql change — so they light up in all three apps' recommendations at once. Every threshold cites its source Dashboard SQL line in a code comment.

Arm 1 — Query Store disabled (INFO)

ScoreDatabaseConfigFact: query_store_on_count < database_count (user DBs only; system DBs excluded from both counts by the collectors) → low 0.3 base folded into the existing DB_CONFIG finding, which roots at any positive severity (a ConfigAdvisoryRootKey). INFO band (< 0.75) matches the Dashboard.
Source: install/50_configuration_issues_analyzer.sql:83 (severity = N'INFO').

Arm 2 — Security-cache (TokenAndPermUserStore) growth (WARNING)

ScoreMemoryFact: reads the USERSTORE_TOKENPERM clerk size (MB) from the otherwise context-only MEMORY_CLERKS fact (the fact carries each top-clerk's MB keyed by clerk_type, and MemoryClerksCollector stores clerk_type = sys.dm_os_memory_clerks.type). Flat WARNING (0.9) at >= 1 GB, matching the Dashboard's single WARNING tier with no size escalation.
Source: install/50_configuration_issues_analyzer.sql:562 (WARNING), :583 (pages_kb / 1024 / 1024 >= 1.0).

Arm 3 — Runnable-queue / worker pressure (STANDALONE)

ScoreCpuFact: a standalone arm rooting the collected cpu_scheduler_stats snapshot on total_runnable_tasks_count, tiered > 50 → 1.0, > 20 → 0.75, > 10 → 0.5, plus the collector's runnable_tasks_warning (SUM(runnable) >= cpu_count) small-box HIGH fallback. Additive to and distinct from the #1494 THREADPOOL runnable-queue amplifier, which is unchanged and still fires independently. Base maxes at 1.0 (WARNING) like every base fact — the CRITICAL band is reached only via amplifier corroboration (the runnable-queue → THREADPOOL escalation).
Source: install/47_create_reporting_views.sql:1839-1844 (report.cpu_scheduler_pressure pressure_level CASE).

Advice wiring

So the new/expanded root keys don't render contentless cards, advice is wired (sourced from the Dashboard's own message / recommendation SQL): new static MEMORY_CLERKS + RUNNABLE_TASKS blocks in FactAdvice, and Query Store folded into ComposeDbConfig (lists "N with Query Store off" + enable-QS remediation).

Collector changes

None. All three arms read facts/metadata the three collector families already emit identically (DB_CONFIG.query_store_on_count/database_count, MEMORY_CLERKS[USERSTORE_TOKENPERM], RUNNABLE_TASKS.total_runnable_tasks_count/runnable_tasks_warning). No fact-key-inventory pin is affected (no new fact keys).

Tests

• New FactScorer tests in both Lite.Tests and Dashboard.Tests (mirrored, Recs Stage B (severity): cap tuning-class CXPACKET, escalate meltdowns through THREADPOOL #1494 style): every tier fires at the right band, stays silent below the bar, and a co-present worse DB_CONFIG issue is left unchanged.
• The Lite clean-server seeder (TestDataSeeder.SeedDatabaseConfigAsync) now defaults Query Store ON (the healthy value, like the rcsi-on / CHECKSUM defaults beside it) so the QS-off arm stays dormant in fixtures not exercising it; the arm itself is covered directly with hand-built facts.
• Green: Lite.Tests FactScorer + Inference + Scenario + Collector + AnalysisService = 173 passed; Dashboard.Tests FactScorer + FactAdvice + Inference = 170 passed. Darling analysis project builds. No Installer/live-server tests run.

🤖 Generated with Claude Code