Page 148 - CyberSecurityEssentialsEnglish

TABLE 3-16: A BRIEF RUBRIC FOR THE LAB OUTPUTS

Criterion Score 1 Score 3 Score 5

Substantive errors, or Most of the content is
Correct within scope,
Correctness claims with no correct with correctable
supported, and verifiable.
evidence. errors.

Covers the requirement,
Misses essential Covers the main
Coverage the constraints, and the
elements. requirement.
exceptional cases.

No source, or a Every substantive claim is
A source exists but
Source source that does not tied to the right passage or
attribution is partial.
match. page.

Language, layout, or a Natural language,
Acceptable with limited
Arabic and layout table that obstructs consistent terminology, and
adjustment.
use. formatting ready for use.

Stable results, or variation
Wide variation Variation that can be
Stability that is expected and
between runs. managed.
controlled.

High review and A clear saving of time with
Operating cost Moderate net benefit.
rework. limited review.

● The output of the lab
A one-page card containing: the definition of the task, the tools, the gates, the scores,
the main errors, the cost, the scope of adoption, the limits, the review owner, and the re-
evaluation date.

Chapter Conclusion: The Right Tool Is the One That Fits Your Task
The Mata v. Avianca case brings us back to the distinction we must not forget: the quality of
the wording does not equal the correctness of the source, and the ability to generate an

answer does not equal the ability to retrieve a fact or carry out a decision safely. Choosing a
tool is not a judgment about who writes best. It is a judgment about which system performs

a defined function with evidence of success, with limits, and with review.

We have learned that a product is a system of layers, that probability explains part of the

variation but not all of it, and that retrieval, fine-tuning, tools, and structured outputs
address different problems. We have also learned that a fair comparison needs a local

147