docs(architect): weekly KB update — 106 files refreshed (2026-04)

Updates across all 5 skills: ms-ai-advisor, ms-ai-engineering,
ms-ai-governance, ms-ai-security, ms-ai-infrastructure.

Key changes:
- Language Services (Custom Text Classification, Text Analytics, QnA):
  retirement warning 2029-03-31, migration guides to Foundry/GPT-4o
- Agentic Retrieval: 50M free reasoning tokens/month (Public Preview)
- Computer Use: Claude Sonnet 4.5 (preview) + OpenAI CUA models
- Agent Registry: Risks column (M365 E7), user-shared/org-published types
- Declarative agents: schema v1.5 → v1.6, Store validation requirements
- MLflow 3: 13 built-in LLM judges, production monitoring, Genie Code
- AG-UI HITL: ApprovalRequiredAIFunction (C#) + @tool(approval_mode) (Python)
- Entra ID Ignite 2025: Agent ID Admin/Developer RBAC roles, Conditional Access
- Security Copilot: 400 SCU/month per 1000 M365 E5 licenses, auto-provisioned
- Fast Transcription API: phrase lists, 14-language multi-lingual transcription
- Azure Monitor Workbooks: Bicep support, RBAC specifics
- Power Platform Copilot: data residency (Norway/Europe → EU DB, Bing → USA)
- RAG security-rbac: 4-approach table (GA + 3 preview access control methods)
- IaC MLOps: Well-Architected OE:05 principles, Bicep/Terraform patterns
- Translator: image file batch translation Preview (JPEG/PNG/BMP/WebP)

All 106 files: Last updated 2026-04 | Verified: MCP 2026-04

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

skills/ms-ai-security/references/performance-scalability/model-distillation-performance.md

@@ -1,6 +1,6 @@
 # Model Distillation for Performance
 
-**Last updated:** 2026-02
+**Last updated:** 2026-04 | Verified: MCP 2026-04
 **Status:** GA
 **Category:** Performance & Scalability
 
@@ -352,6 +352,80 @@ print(f"ROI: {savings['roi_months']} måneder")
 | Hyppig endring i oppgave | Unngå distillation | Re-training overhead |
 | Latens-kritisk (<500ms) | Distiller til nano + PTU | Lavest mulig responstid |
 
+
+## Modellvalg og routing-strategi (oppdatert 2026-04)
+
+Microsoft dokumenterer nå **10 seleksjonskriterier** ved valg av AI-modell for distillasjon:
+
+| Kriterium | Relevans for distillasjon |
+|-----------|--------------------------|
+| Task fit | Velg teacher og student basert på oppgavens art |
+| **Routing strategy** | Definer routing FØR distillasjon — påvirker teacher-modellvalg |
+| Cost | Studentmodellens kostnad er primær motivasjon |
+| Context window | Student må håndtere samme kontekst som teacher |
+| Security | Studentmodell arver ikke teachers sikkerhetstiltak — re-evaluer |
+| Region | Student deployes i samme region som teacher for dataresidency |
+| Deployment | PTU vs Standard — student er oftest Standard til start |
+| Domain | Domene-spesifikk teacher gir bedre student |
+| Performance | Latens- og throughput-krav til student (se modellmatrise) |
+| **Tunability** | Studentmodellen MÅ støtte fine-tuning (f.eks. GPT-4o-mini, GPT-4.1-nano) |
+
+### Modell-routing som distillasjonsstrategi
+
+```python
+# Model routing strategy i distillasjonskontekst
+# Teacher: GPT-4.1 (høyeste kvalitet)
+# Router: Klassifiser oppgavekompleksitet → velg modell dynamisk
+# Student: GPT-4.1-mini eller GPT-4.1-nano (basert på klassifisering)
+
+from openai import AzureOpenAI
+import json
+
+client = AzureOpenAI(
+    azure_endpoint="https://my-foundry.openai.azure.com",
+    api_key="...",
+    api_version="2024-10-21"
+)
+
+def classify_task_complexity(user_input: str) -> str:
+    """Klassifiser oppgavekompleksitet for routing."""
+    response = client.chat.completions.create(
+        model="gpt-4.1-nano",  # Rask og billig til routing
+        messages=[{
+            "role": "system",
+            "content": "Klassifiser denne brukerforespørselen: 'simple' (fakta, svar, klassifisering) eller 'complex' (resonnering, kreativt, multi-steg). Svar med ett ord."
+        }, {"role": "user", "content": user_input}]
+    )
+    return response.choices[0].message.content.strip().lower()
+
+def route_to_model(user_input: str) -> str:
+    """Route til riktig modell basert på kompleksitet."""
+    complexity = classify_task_complexity(user_input)
+    
+    if complexity == "simple":
+        model = "ft:gpt-4.1-nano:distilled-v1"  # Distillert nano for enkle oppgaver
+    else:
+        model = "gpt-4.1"  # Teacher for komplekse oppgaver
+    
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": user_input}]
+    )
+    return response.choices[0].message.content
+
+# Routing strategy gir: lavere kostnad for enkle oppgaver + høy kvalitet for komplekse
+```
+
+### Oppdatert modellmatrise for distillasjon
+
+| Modell | Tunability | TPM (PTU, input) | Anbefalt student-rolle |
+|--------|-----------|-----------------|----------------------|
+| GPT-4.1-nano | Ja | 59,400 | Enkle oppgaver, latens-kritisk |
+| GPT-4o-mini | Ja | 37,000 | Generelle oppgaver, kostnadsoptimal |
+| GPT-4.1-mini | Ja | 14,900 | Moderate oppgaver, god balanse |
+| GPT-4.1 | Nei (direkte) | 3,000 | Teacher (ikke student) |
+| GPT-4o | Nei (direkte) | 2,500 | Teacher (ikke student) |
+
 ## Referanser
 
 - [Azure OpenAI stored completions & distillation](https://learn.microsoft.com/azure/ai-foundry/openai/how-to/stored-completions) — Distillation workflow