Nordlicht Reisen GmbH (R-01) has received the folder from lesson 6.1. Before anyone releases money there, a second person reads it critically and asks: Why this model, why this service, what happens if it is not right? This lesson practices defending, shows openly what the course has not taught, and turns the "open" places of the lessons into a checklist. It is paper: No command runs against Azure, and there is no learning check.
Prior knowledge. Lesson 6.1 (folder, conflicts K-1 to K-6, Bicep sketch) and the decision notes of lessons 1.1 to 5.5. The form of the note (context, decision, consequences, alternatives) is taught in "K4, lesson 4.1.3". The map of the exam skills is in lesson 0.1; here it is read against what the course has really delivered.
The answer in three parts
A question is not answered with a preference. The pattern has three parts: the requirement (which row of the sheet), the decision (what was chosen, with the lesson) and the limit that flips it (at which measured value or which fact the answer would be different). If the limit is missing, it is an opinion. If the requirement is missing, something was built that nobody ordered.
| Area | Question | Answer in the pattern |
|---|---|---|
| 1 | Why Data Zone Standard and not Global Standard? | A-02 (data in the EU) → Data Zone Standard, policy against GlobalStandard (1.1, 1.4) → flips only for services without personal data that do not exist otherwise: bild, video (K-2 in 6.1) |
| 1 | How do you know that a new agent version is not worse? | A-04 → test set locally and as an evaluation against exactly this version (1.2, 2.6) → flips if the evaluation is restricted in the region (2.6: open); then the local run plus a manual step remains |
| 1 | When would you buy reserved throughput? | A-06 and A-05 → distribution to chat-klein, smoothing in the client (1.3) → flips if the measurement shows a permanently high base load |
| 2 | Why a prompt agent and not a hosted agent? | E-01, A-01 → rules are in the application (2.5) → flips as soon as code has to run in the agent |
| 2 | How often may the assistant revise its draft? | A-05 → at most one revision after criticism from the pruefer (2.7) → flips if the test set shows that a second one improves things and the cap allows it |
| 3 | Why do images run in the code and not through an agent? | Approval and counter must be in the code (3.1) → flips if the assistant is to draw a sketch for customers |
| 3 | How reliable is an alt text from chat? |
The editorial team requires drafts (3.3) → house rules in the code, spot checks → flips if the model behind chat does not understand images (3.3: open) |
| 4 | Why a language model for the analysis and not Language? | 4.1 requirements (tone, summary) → model with schema, PII through Language → flips with thousands of similar documents with fixed types |
| 4 | Why does a recording become text first? | R-11: audio is foreign content (4.4) → text stays checkable → flips if the test set shows a clear advantage of direct audio processing |
| 5 | Why an indexer and not file search? | R-03: 120 PDFs with changes (5.1, 1.1) → indexer with schedule → flips with few files if nobody wants to operate an indexer |
| 5 | Why Content Understanding for receipts? | 5.4-01, 5.4-02 → fields with confidence, threshold 0.8 (assumption) → flips with Danish handwriting that the language list does not carry (1.1) |