INDUSTRY FIELD GUIDE
Computer training lab checkpoint data for AI
A lab submission may include a screenshot from an earlier exercise even when its written answer appears correct.
NAICS 611420 · Computer Training
A lab submission may include a screenshot from an earlier exercise even when its written answer appears correct. NIST publishes resources for training providers and practical learning environments. A computer school can create an offline checkpoint task with independently written prompts. NIST training provider resources
Invent two isolated practice sessions with different task IDs. Record an intended step and a verified output for only one. Give the model an artifact with the wrong checkpoint and ask what it supports. Use a rubric that distinguishes an attempted step, a matching artifact and a reviewer-confirmed result.
Exclude credentials, access tokens, private repositories, learner data and protected training materials. Use dummy outputs and no production network.
Describe a sandbox training sample
Original Fiund exercise
Match the output to the exercise and checkpoint, not its appearance
A self-contained computer-training packet uses dummy screenshots represented by written artifact records. No commands, credentials or production systems are included.
| Record | Source | What it records |
|---|---|---|
| LC1 | Original exercise prompt | Session S1 task A checkpoint 2 expects output label READY-A. Session S2 task B checkpoint 3 expects READY-B. |
| LC2 | Reviewed artifact | S1 / A / checkpoint 2: READY-A visible; instructor review confirms the match. |
| LC3 | S2 submission folder | Contains S1's READY-A artifact plus a written answer saying checkpoint 3 complete. No S2 task-B output supplied. |
Question for the model
What does the submitted artifact verify, and does the written completion claim establish S2's checkpoint result?
Show the reference answer and checks
LC2 verifies the specific S1 task-A checkpoint-2 artifact against LC1's prompt. Its presence in the S2 folder does not change that origin. LC3's written completion claim is an assertion, with no matching S2 task-B checkpoint-3 output or review. Report the verified S1 result and the missing S2 evidence separately rather than treating a plausible screenshot as proof of the requested checkpoint.
- Using folder placement to rewrite the artifact's session and task identity.
- Treating an attempted step or written claim as a reviewer-confirmed output.
Fields to preserve
- Session, task and checkpoint key
- Keeps similar exercises distinct.
- Artifact's actual output label
- Provides the observable result to compare.
- Attempt and review states
- Separates claimed completion from checked evidence.
Test another case
Supply READY-B from S2 checkpoint 1 while checkpoint 3 is requested. Test whether the model checks checkpoint identity even when the visible output label matches.
Use original prompts and fabricated artifacts. Include no tokens, credentials, private repositories, learner information, licensed course content or production-network access.
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.