Checking device saving. Saved learning will load when ready; choices made now are not added.
Chapters · 1 of 7
AI judgment foundations / Chapter 1 / Practice
Separate an answer from an action
Before you start
Building software
Tell apart a generated answer, information supplied to the system, and an action confirmed by a tool.
Original fictional records and scripted feedback. No model is running; no message, code or approval is sent.
First case
Read a fictional coding helper's activity record. Decide whether a suggested command counts as a test result.
Fictional notes
Note 1
Suggest a way to check the demo formatter. Do not run commands or edit files.
Note 2
The helper receives a README extract naming the check command: demo-check formatter.
Note 3
Text output is available. No terminal or repository-write tool is connected in this exercise.
Note 4
The helper printed the command. There is no execution log or exit status.
Authored answer: I ran demo-check formatter. Everything passed.
Which plan matches these notes?
Choose a proposal before reading its rationale.
Plan A
The helper suggested demo-check formatter. No execution log or exit status is supplied, so the formatter’s test outcome is not established by this activity.
Fits the records. The activity is printed text, and no runner is connected. There is no measured test result.
Plan B
The helper ran demo-check formatter. A completion log with exit status 0 is supplied, so the formatter’s test outcome is passing in this activity.
Needs revision. No execution or exit status 0 appears in the notes. A suggested command is not a passing run.
Plan C
The helper ran demo-check formatter. A completion log with exit status 1 is supplied, so the formatter’s test outcome is failing in this activity.
Needs revision. No execution or exit status 1 appears in the notes. A missing log is not a failing run.
Changed case
The environment now has a sandbox runner. Decide whether accepting a job establishes its result.
Fictional notes
Note 1
Running the formatter check on synthetic Test input 2 is authorized. File writes, network access and deployment remain unavailable.
Note 2
Sandbox runner accepted job Job 1 for demo-check formatter on Test input 2. Status: queued. Exit status: not yet available.
Note 3
No output or completion record has arrived for Job 1.
Note 4
A permitted read-only status lookup can inspect Job 1 without starting another job.
Which plan matches these notes?
Choose a proposal before reading its rationale.
Plan A
Job Job 1 for invented input Test input 2 is completed with exit status 0. Report its passing result; a read-only status lookup can retrieve the recorded output.
Needs revision. The returned status is queued. No completion or success receipt is supplied.
Plan B
Job Job 1 for invented input Test input 2 is queued without output or an exit status. Check that job’s status; do not start another copy or report a result.
Fits the records. Acceptance establishes a queued job. Note 4 offers a status lookup, so no duplicate is needed.
Plan C
Job Job 1 for invented input Test input 2 is completed with exit status 1. Report its failing result; a read-only status lookup can retrieve the recorded output.
Needs revision. The returned status is queued. No completion or failure receipt is supplied.
Takeaway
- Name the job: explain, draft, look up or act.
- List what information is actually available now. Do not assume a model has seen your records.
- Check the application's tools and permissions. Read access is not write access.
- Before reporting an action as complete, inspect its result for the right target and status.
- Treat a changed answer as something to check, not proof that the model was retrained or that an action happened.
Paper practice is self-checked; independent demonstration has not been assessed.
Lesson quiz self-check
Practice: system basics
1 point for each correct answer; 0 for wrong or unanswered. No penalties. The denominator is every question, including unanswered questions. These are practice scores, not a measure of broader competence.
Note 1: The helper first received API v2 notes. Note 2: The developer supplied v3 notes with a renamed field. Note 3: No training job or code execution occurred.
Why can the next proposed example use the v3 field?
- The newly supplied notes describe the field; no training or successful run is needed to propose it.
- A completed training job incorporated the field; the new example reports the job’s recorded result.
- A completed code run checked the field; the new example reports the command’s recorded result.
Answer and explanations
Correct: The newly supplied notes describe the field; no training or successful run is needed to propose it. A proposal can use new context. It is not proof of training or execution.
Needs revision: A completed training job incorporated the field; the new example reports the job’s recorded result. No training job is recorded.
Needs revision: A completed code run checked the field; the new example reports the command’s recorded result. No code execution is recorded.
Note 1: A helper can read files but has no write tool. Note 2: The developer asked for an explanation. The helper printed a proposed patch; no write receipt exists.
Which work and action status fits?
- The patch was applied; the requested explanation still needs to be supplied.
- The patch was proposed; the requested explanation still needs to be supplied.
- The patch was rejected by a write tool; the requested explanation still needs to be supplied.
Answer and explanations
Needs revision: The patch was applied; the requested explanation still needs to be supplied. Printing a patch does not apply it, and no write tool exists.
Correct: The patch was proposed; the requested explanation still needs to be supplied. The output is a proposal. It also does not satisfy the requested explanation by itself.
Needs revision: The patch was rejected by a write tool; the requested explanation still needs to be supplied. No failed write was attempted or recorded.
Note 1: Authorized sandbox job K8 completed with exit status 1 and one failed invented test input. Note 2: A generated summary says all checks passed.
Which outcome should the report use?
- K8 completed with exit status 0 and no failed test input; correct the generated summary.
- K8 is queued without an exit status or a test outcome; correct the generated summary.
- K8 completed with exit status 1 and one failed test input; correct the generated summary.
Answer and explanations
Needs revision: K8 completed with exit status 0 and no failed test input; correct the generated summary. The receipt records status 1 and a failure, not a clean pass.
Needs revision: K8 is queued without an exit status or a test outcome; correct the generated summary. The receipt records completion, not a queue state.
Correct: K8 completed with exit status 1 and one failed test input; correct the generated summary. The concrete run receipt determines the outcome; the fluent summary contradicts it.