Check your reasoning

Answer without an agent, then compare with the public answer guidance. The questions are formative. Most use situations outside the lecture's examples, so you apply the method rather than recall a run; what one run can decide is the table to keep open.

1. A demonstration shows a pipeline take one issue to a merged change with nobody touching it. The presenter concludes that the pipeline runs itself. What is the reach of that conclusion, what does the one run support, and what single observation would count against it?
2. A pipeline's record for a change reads: scope pass, gate stop with one failed case, release not-run, exit status 1. A colleague runs it again with an acceptance name and reports the change as approved. What does the record of the second run say for each stage, and why can a name not clear it?
3. You have two gate records for the same change digest. The first shows 9 cases and one cases digest; the second shows 7 cases and another digest. Both say pass. What changed between the runs, which record can an acceptance rest on, and what do you ask before you decide anything?
4. A change passes every frozen case, and its note says "no other behaviour changes". What does the green gate support, which part of the note did nothing test, and how would you test it?
5. A change makes the gate report a pass by printing the report line itself and ending the run. Which property of the gate allowed that, who is the last chance to catch it, and what would have to change in the gate to close the gap?
6. A team says its test stage has been green for two hundred runs and is ready to run with no person watching. Name the four things a stage must show first, and say which of them two hundred green runs supply.
7. A review says: "The upload endpoint accepts any file type." Your requirement allows only .csv and .json. You send report.exe and the endpoint refuses it. Which disposition does that one run support, and what scope sentence keeps the conclusion within what you checked?
8. A review says: "Some search query makes the results page list another user's reports." You try three ordinary queries and each lists only your reports. One classmate wants to mark the claim false, another wants to mark it as impossible to decide. What is wrong with each, and what do you do next?
9. A review says the deployment settings store a database password in plain text. Those settings are not in your snapshot. A classmate runs a secret scan on the tagged tree, sees no secrets and wants to close the claim. What is wrong with that evidence, and what does the row say instead?
10. In week 10 you ended your audit of an evaluation instance with "accept" for its scorer, citing a results table. Since then the scorer gained two cases and the model version behind your agent changed. Which decision word describes that old line today, and what do you do before you rely on it in your final project?

Keep one revised answer beside the command output that supports it.

Transfer

Pick one stage of your own project. Write the claim "this stage can run with nobody watching" as a bounded sentence: which changes, behind which check, with which stop. Then name the run you could show someone in which that check failed.

Spaced retrieval: evaluation and recovery

Answer these two short prompts after the session. They revisit weeks 10 and 11 so that the evidence, not a memorised label, is carried forward.

R1. Retrieval 1. A scorer accepts a scalar answer but the contract changes to JSON. Which parser and malformed cases must be added?

R2. Retrieval 2. A timeout leaves no persisted-state observation. Name two possible histories and the next discriminating check.