Understand

What it is: Imagine a course team’s inbox. One message asks for a new password, another wants the price, and a third challenges a charge. Before writing a reply, the team needs to decide who should receive each request. This choice has few known alternatives and can be studied separately from the response writing.

Why learn: Choose between a rule, a bounded decision, and a generative response.

Key concepts: context, criteria, and evidence applied to lesson 1.1.

What it is: A decision depends on three elements. The context contains the material to evaluate. The question declares exactly the intended judgment. The options bound the responses the system knows how to interpret. If any of them is poorly defined, the output may be valid in format and useless for operation.

Why learn: Choose between a rule, a bounded decision, and a generative response.

Key concepts: context, criteria, and evidence applied to lesson 1.2.

What it is: An ad may show a large speed difference between two systems. To understand the number, we need to know which task was used, how much context was included, which model was compared, what configuration it had, and what counts as a correct answer. An observed maximum doesn’t become a fixed advantage in every use.

Why learn: Choose between a rule, a bounded decision, and a generative response.

Key concepts: context, criteria, and evidence applied to lesson 1.3.

What it is: Choice is used to select between alternatives described in advance. In triage, the options may represent teams. Names must be stable for code, and criteria must be understandable for people labeling examples. It’s not enough to write a list of vague words: “financial” and “administrative” can overlap if the boundaries aren’t explained.

Why learn: Distinguish Choice, Noul, and Score based on the type of response needed.

Key concepts: context, criteria, and evidence applied to lesson 2.1.

What it is: Noul answers a binary question with a number between zero and one. The value represents the probability of “yes” according to the model. Close to one points to yes; close to zero points to no. The intermediate value doesn’t mean an intermediate intensity of the topic: it indicates the alternatives are not well separated.

Why learn: Distinguish Choice, Noul, and Score based on the type of response needed.

Key concepts: context, criteria, and evidence applied to lesson 2.2.

What it is: Score is suitable when there is a scale with descriptions. “What’s the priority?” works well only if each level has an operational meaning. Low may mean doubt without blocking; medium, difficulty with a workaround alternative; high, work blocked without an alternative. The goal is to make the rubric reviewable by people.

Why learn: Distinguish Choice, Noul, and Score based on the type of response needed.

Key concepts: context, criteria, and evidence applied to lesson 2.3.

What it is: In Choice, the distribution provides a value for each alternative. The confidence field summarizes a property of that distribution according to the provider’s implementation. We don’t assume it equals the highest probability. When logging a result, we store both fields with distinct names.

Why learn: Use uncertainty without turning it into automatic authorization.

Key concepts: context, criteria, and evidence applied to lesson 3.1.

What it is: Create a fictional case: the person writes “I don’t want to cancel; I just need to change the date.” The model chooses cancellation with high probability. The output is structurally valid, but the interpretation failed on the negation. This example doesn’t need to be a real benchmark to reveal an architecture flaw: executing an irreversible action just because the number is high.

Why learn: Use uncertainty without turning it into automatic authorization.

Key concepts: context, criteria, and evidence applied to lesson 3.2.

What it is: The next step depends both on uncertainty and on the cost of being wrong. Showing a suggested queue and deleting an account have different consequences. There is no universal threshold that makes the two actions equivalent. The project should define which actions are reversible, which require confirmation, and which are only allowed for authorized people.

Why learn: Use uncertainty without turning it into automatic authorization.

Key concepts: context, criteria, and evidence applied to lesson 3.3.

What it is: A per-million-tokens fee needs to be converted before multiplying by the volume of calls. In the price reference consulted on 18/09/2026, Jev’s entry cost is US$ 0,042 per million. For 10,000 total input tokens, we divide 10,000 by one million and multiply by 0,042. The result is US$ 0,00042.

Why learn: Calculate feasibility including the work that happens after the model.

Key concepts: context, criteria, and evidence applied to lesson 4.1.

What it is: A cheap call doesn’t guarantee a cheap process. After triage, there may be a generative model, human review, text extraction, storage, and rework. If an error routes a request to the wrong team, the real cost includes the time to fix the forwarding.

Why learn: Calculate feasibility including the work that happens after the model.

Key concepts: context, criteria, and evidence applied to lesson 4.2.

What it is: A first pilot should answer a small, verifiable business question. “Automate the whole company” doesn’t provide a completion criterion. “Suggest the ticket queue in Portuguese without increasing rework” makes it possible to gather examples, review the results, and decide whether to move forward.

Why learn: Calculate feasibility including the work that happens after the model.

Key concepts: context, criteria, and evidence applied to lesson 4.3.