Week 11 self-check answer guidance

Use these explanations after your first attempt. Keep every conclusion tied to a line, a field or a run that another reader could check.

Q1

No. The report can be true and the session can still have gone beyond the task, because a call is checked only against its grant. The missing comparison is what each grant can reach against what the task needs. A grant that names every repository lets every read pass; narrow it to the task's repository and keep one denied call as its test.

Q2

The third request is refused. After two replies 1200 tokens are used; the third reservation would make 1200 + 900 = 2100, which is more than 2000. At that moment 800 tokens are unspent. It is refused because the gateway reserves the worst case before sending: the actual use is unknown until the reply arrives, and a cap checked only afterwards would already be passed.

Q3

Four requests are sent and the fifth is refused. Each reply settles at 2 units, so four replies leave 8 units spent; the fifth reservation would make 8 + 5 = 13, more than 12, so it is refused with 4 units unspent. The fixture's last line reads request 5: refused before sending (Request exceeds the remaining budget.) spent 8 of 12.

Q4

The gateway cannot tell whether the provider did the work and charged for it; a lost reply marks where observation stopped, not that nothing happened. Releasing the reservation would let later requests spend units that may already be owed, so the run could pass its cap without any single request being refused.

Q5

Count every run and every review minute, including the failed run, the timed-out run and the rejected change: 4 + 4 + 2 + 6 + 2 = 18 units and 12 + 8 + 20 = 40 review minutes. Two changes were accepted, so 18 units ÷ 2 accepted changes = 9 units and 40 ÷ 2 = 20 review minutes per accepted change. The summary's 4 units is one run of one change.

Q6

No. The rates alone do not justify the switch: the rule route loses a case the model route passes, and a higher total hides which case was lost. Inspect that case, repair the rule, rerun all 50 frozen cases on the revision in use and switch only if no case the model route passes now fails. Even then the result says nothing about inputs outside the 50 cases.

Q7

Refuted. The claim says the errors rose; the log shows 2 in 50 before and 1 in 50 after. Refuting it leaves the server errors in both windows unexplained, and it does not show that imports are healthy. Record those errors as a separate observation to follow up.

Q8

Cannot tell. No field in the request log records a query or an index, so the log can neither confirm nor refute the claim; slow lines show only that the page was slow. A check that could decide it: on the same data, compare the page's duration with and without the index, or read the query plan for that page. Until then name the missing observation and keep the claim open.

Q9

The request committed: its effect exists and only the reply was lost. The timeout marks where the client stopped observing. A repeat with the same key and the same build must return the existing job and create no new one; the same key with a different build is a conflict that creates nothing, and a new intended retry needs a new key.

Q10

The missing input is two deliveries of the same request arriving together, with the order forced so that both lookups finish before either insert; a lookup followed by a separate insert then stores two jobs. A regression is evidence of recovery when it fails on the incident build, fails on the repair that only looks right, and passes on the real repair, with the stored jobs counted as well as the replies.

Spaced retrieval and transfer

Retrieval 1. Settle the denominator first: find the two instances missing from the results table, record why each is absent, and either run them or report the rate over all 12 instances with the missing ones shown as not run. A pass rate over 10 is not a result for a manifest of 12.

Retrieval 2. The text is data. A tool result carries no authority, whatever it says; whether a command runs is decided by the task, the instruction file and the permission check, and by the person who owns the session. Record the text as an injection attempt and do not act on it.

Transfer. A useful answer names a condition outside the tested path, for example a repeat after a restart or three deliveries at once, gives the input and the expected stored state, and says which decision moves: a failure on that input changes accept to narrow or recheck, and the note says so before anyone relies on the fix.