feat(ms-ai-architect): C3.3 — course_products (live-enumerert) + makeCourseClassifier (TDD) [skip-docs]

Produkt-slug->kategori-mapping for kurs-deteksjon (C3.3 av GODKJENT C3-spec).

- lib/taxonomy.mjs: makeCourseClassifier(tax) — speiler makeClassifier; forste
  in-domain produkt vinner; skill AVLEDES fra category_skill (lagres ALDRI i
  course_products -> kan ikke divergere); ukjent/tom/non-array -> null.
- data/domain-taxonomy.json: ny topp-niva course_products (slug->category) +
  provenance. 15 in-domain live slugs.
- test-taxonomy.test.mjs (+6): atferd (in-domain->{skill,category}, first-wins,
  out-of-domain/tom/non-array->null, manglende map) + REELL data-invariant
  (hver slug-kategori resolver til skill, derivasjon matcher) + azure-openai-live.

ENUMERERING (gotcha #2, live mot /api/v1/modules?products=<slug> server-side):
- DOD (0 treff): azure-ai-foundry, ai-services, azure-ai-services, foundry ->
  Foundry/AI-services-kurs dukker opp under 'azure-openai'.
- Slug-feller bekreftet live: 'fabric' (ikke microsoft-fabric), 'entra' (ikke
  microsoft-entra-id), 'azure-cosmos-db' (ikke cosmos-db), 'azure-cognitive-search'
  (ikke azure-ai-search).
- 'azure-kubernetes-service' live men EKSKLUDERT (ingen AI-kategori -> stoy).
- Inkluderte (15): azure-openai, azure-machine-learning, azure-cognitive-search,
  azure-cosmos-db, fabric, azure-databricks, microsoft-copilot-studio,
  power-automate, power-apps, power-platform, ai-builder, microsoft-365-copilot,
  microsoft-purview, entra, defender-for-cloud.

Avvik fra spec 4.4-eksempel: course_products lagrer slug->category (ikke
slug->{skill,category}); skill avledes — anti-divergens, encapsulated av
makeCourseClassifier (downstream-kontrakt {skill,category} uendret).

Gate 7 C3.3 mott: test gronn, eval --json IDENTISK (refTall/K-counts uendret).
kb-update 263->269, validate PASSED (239/0), null regresjon.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

scripts/kb-update/data/domain-taxonomy.json

@@ -4,7 +4,8 @@
   "provenance": {
     "category_skill": "PHYSICAL DISK is canonical (skills/<skill>/references/<category>/). category-skill-map.json drifted on 4 entries (copilot-extensibility, monitoring-observability, performance-scalability, prompt-engineering) and is no longer consumed by code.",
     "sitemap_prefixes": "poll-sitemaps superset (18). discover-new-urls previously had only 12 — now reads this list for poll parity.",
-    "notes": "dotnet_en-us_ deliberately excluded: 75 sitemaps, only ~12 matches — not worth weekly polling."
+    "notes": "dotnet_en-us_ deliberately excluded: 75 sitemaps, only ~12 matches — not worth weekly polling.",
+    "course_products": "Product slugs enumerated LIVE from the Learn Platform API (C3.3, server-side /api/v1/modules?products=<slug> probe). Doc slug 'azure-ai-foundry' is DEAD (0 hits) → Foundry/AI-services courses surface under 'azure-openai'. Slug gotchas confirmed live: 'fabric' (not 'microsoft-fabric'), 'entra' (not 'microsoft-entra-id'), 'azure-cosmos-db' (not 'cosmos-db'), 'azure-cognitive-search' (not 'azure-ai-search'). 'azure-kubernetes-service' is live but excluded (no AI category → noise). Maps slug->category; the owning skill is DERIVED from category_skill via makeCourseClassifier (never stored, cannot diverge)."
   },
   "sitemap_prefixes": [
     "azure_en-us_",
@@ -49,6 +50,23 @@
     "responsible-ai": "ms-ai-governance",
     "security-scoring": "ms-ai-security"
   },
+  "course_products": {
+    "azure-openai": "azure-ai-services",
+    "azure-machine-learning": "mlops-genaiops",
+    "azure-cognitive-search": "rag-architecture",
+    "azure-cosmos-db": "data-engineering",
+    "fabric": "data-engineering",
+    "azure-databricks": "data-engineering",
+    "microsoft-copilot-studio": "platforms",
+    "power-automate": "platforms",
+    "power-apps": "platforms",
+    "power-platform": "platforms",
+    "ai-builder": "platforms",
+    "microsoft-365-copilot": "copilot-extensibility",
+    "microsoft-purview": "responsible-ai",
+    "entra": "ai-security-engineering",
+    "defender-for-cloud": "ai-security-engineering"
+  },
   "relevance": {
     "include": [
       { "pattern": "/azure/ai-foundry/", "category": "azure-ai-services" },

scripts/kb-update/lib/taxonomy.mjs

@@ -120,6 +120,31 @@ export function makeClassifier(tax) {
   };
 }
 
+/**
+ * Build a course classifier for the Learn Platform API course-detection spore
+ * (C3). course_products maps an in-domain product slug → category; the owning
+ * skill is DERIVED from category_skill (never stored redundantly), exactly as
+ * makeClassifier derives skill for URL classification — so the slug→skill path
+ * cannot diverge from the canonical category ownership. The FIRST in-domain
+ * product in the list wins; a course whose products are all out-of-domain (no
+ * slug in course_products) returns null and is therefore not a lead.
+ * @param {object} tax
+ * @returns {(products: string[]) => ({skill: string|null, category: string}|null)}
+ */
+export function makeCourseClassifier(tax) {
+  const map = tax.course_products ?? {};
+  return function classifyCourse(products) {
+    if (!Array.isArray(products)) return null;
+    for (const slug of products) {
+      const category = map[slug];
+      if (category) {
+        return { skill: tax.category_skill[category] ?? null, category };
+      }
+    }
+    return null;
+  };
+}
+
 /**
  * Build a file-priority function mirroring report-changes getFilePriority:
  * ordered regex rules over the lowercased path; first match wins; else default.