Write your response in your own notes, then compare it with the sample answer. This practice is separate from the scored exam and does not affect your certificate.
Practice 1
Design two records for the fictional Orchard workshop assistant, one for a resumable run and one for a reusable preference. State the fields and source of each, and say which layer chooses the tenant.
Compare with a sample answer
The run is a checkpoint under checkpoint/run-a with schema 1, step 0 and an empty items list. The preference is a fact under fact/preference with the fictional checklist text, source workshop-note-1 and reviewed false. The trusted application, not the caller, chooses tenant orchard. The checkpoint serves one run, while the fact needs its own review and retention decision, and neither record holds a full conversation or a credential.
The sample separates the two record lifecycles and keeps tenant selection in the trusted application, as the model answer does.
Practice 2
A fact is created at time 100 with ttl 10. Predict get at 109 and at 110, the stored deadline after the read at 109, and what remains on disk after the read at 110.
Compare with a sample answer
At 109 get returns the record with expires 110, and the read leaves that deadline unchanged. At 110 get returns None because expiry is exclusive. The row stays in records until forget or purge removes it, so a clock moving backwards could make it visible again. Visibility, row retention and storage-level erasure are separate questions.
The sample applies now plus ttl, exclusive expiry and the rule that reads never extend a deadline.
Practice 3
Two workers read the same live checkpoint revision. Explain why only the first replacement should succeed, what a third worker holding the old revision can do after delete and recreate, and the recovery action after Conflict.
Compare with a sample answer
Worker A presents the current revision, so its write commits and the persistent counter advances. Worker B still presents the old revision and receives Conflict. If the record is then deleted and recreated, the new row takes another fresh counter value, so a third worker holding the deleted row's token cannot replace it. After Conflict a worker should reread and recompute, or stop; copying the new revision onto its stale proposal would bypass the protection.
The sample traces the compare-and-swap, the delete-and-recreate case and the correct reaction to Conflict.
Practice 4
The demonstration forgets a checkpoint and then counts zero rows in records. Write a short deletion receipt that states what was verified, names places that were not inspected and explains why a False result from forget would be ambiguous.
Compare with a sample answer
Verified: the matching checkpoint row was removed and a count query found zero rows in records. Not inspected: database free pages, backups, process memory, and any vector index or other service. A False result from forget would mean either that no matching identity existed or that its revision differed, so a reread under the current policy is needed to tell them apart. This receipt does not describe complete erasure.
The sample keeps the verified claim narrow and lists unexamined copies, as a deletion receipt should.
Practice 5
Trace the demonstration checkpoint's deadlines, explain why a repeated step-2 advance does not extend retention, and describe what remains unresolved if a crash happens after an imagined external send but before put.
Compare with a sample answer
The checkpoint writes at 100, 101 and 102 with ttl 60 set deadlines of 160, 161 and 162. At step 2 advance returns the existing revision before calling put, so 162 is unchanged. If the process crashes after an external send but before put, the checkpoint stays at the earlier step and a restart may send again. Revision checks protect the local record but do not undo or deduplicate the send, so an explicit idempotency or delivery protocol must be added and tested first.
The sample derives the deadlines from now plus ttl and separates local recovery from the external effect.
Practice 6
In a separate copy of the lab, describe a mutation that removes the tenant boundary from purge while keeping valid SQL, the test you expect to fail, and what passing again after the repair does and does not show.
Compare with a sample answer
Change WHERE tenant=? AND expires<=? to WHERE ? IS NOT NULL AND expires<=?, so expired rows for both Orchard and Harbour can be removed. test_purge_tenant_and_boundary should then fail on the removed count or the remaining rows. After restoring the tenant predicate, all ten methods must pass. This shows that the test catches this specific deletion boundary defect, not that every future authorisation path is covered.
The sample names a valid mutation, the failing test and the limit of what the passing suite proves.