Skip to content

INDUSTRY FIELD GUIDE

Engineering requirements and verification records for AI

A verification matrix connects requirements to the evidence used to check them.

By Fiund · Updated 2026-11-05

NAICS 541330 · Engineering Services

A verification matrix connects requirements to the evidence used to check them. NASA’s systems engineering handbook includes requirements and verification matrices. A consulting engineer can create an original low-risk fixture example to test whether an AI system can locate the right evidence and keep exceptions visible. NASA Systems Engineering Handbook appendices

Invent a tolerance for an inert bracket and a second revision with a different test condition. Give the reviewer two results, only one taken under the current condition. Ask which requirement has matching evidence and how the older result should be treated. Score the requirement ID, condition and supported result separately.

Use practice hardware, original documentation and dummy readings. Keep actual client designs, proprietary methods, regulated files and product safety determinations outside the exercise.

Describe an engineering verification sample

Original Fiund exercise

A verification result must match the current requirement condition

An inert practice bracket and arbitrary readings support an original requirements matrix. The exercise makes no actual engineering, regulated-product or safety determination.

Fictional records supplied to the evaluator
RecordSourceWhat it records
EV1Practice requirement R1 revision AMeasure Q under condition C1; allowed practice range 4–6 units.
EV2Authorized revision BR1 now requires condition C2; numerical range unchanged. Result T1=5 under C1 predates B.
EV3Current result T2R1/Q=5 under C2, reviewer links to B. Separate R2 has no result supplied.

Question for the model

Which result matches R1's current condition, and does the older in-range T1 or R1's result establish R2 verification?

Show the reference answer and checks

EV3's T2 matches R1 revision B's C2 condition and reports a value within the fictional 4–6 range. T1 is also numerically in range but uses superseded C1, so it is not matching evidence for the current condition without an equivalence record. Neither result names R2. Preserve requirement, revision, condition and result link rather than accepting a bare number or transferring verification across requirements.

  • Accepting an in-range result under the wrong test condition.
  • Using one requirement's result to mark an unrelated matrix row verified.

Fields to preserve

Requirement and revision
Defines what must be checked.
Condition and measure
Makes the result's scope explicit.
Evidence link and reviewer
Connects a matrix status to its actual source.

Test another case

Change the range while retaining C2. Require the model to check both condition and numerical criterion rather than only the latest condition label.

Use original practice hardware and dummy readings. Exclude client designs, proprietary methods, regulated files and product-safety judgments.

Download this fictional example and answer key (JSON)

Discuss a contribution

Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.

Primary sources and further reading