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
Sketch three fixed questions for the fictional policy. For each, write an exact evidence string and explain what a hit would and would not establish.
Check your answer against these points:
- Use fixed queries
- Label exact evidence strings
- State what the hit measure cannot show
Compare with a sample answer
A useful set includes duration, return location and a boundary-sensitive fact. Each expected string is labelled. A hit only means one chunk contains it; ranked retrieval and grounded answers need separate checks.
Ungraded paper reflection: Build a labelled fixture set. Use the cited source passage; this activity does not measure competence.
Practice 2
Explain what changes between settings 8/0, 8/2 and 10/4 in the supplied table. Recommend what to inspect next before choosing for a real collection.
Check your answer against these points:
- Report the three measured rows
- Mention overlap repetition cost
- Avoid declaring a production winner
Compare with a sample answer
The table reports 2/2/16, 3/2/20 and 2/3/20 for chunks/facts/occurrences. The larger setting covers all three tiny facts, but a varied labelled and held-out set plus ranked results is still needed.
Ungraded paper reflection: Compare measured settings. Use the cited source passage; this activity does not measure competence.
Practice 3
Propose a held-out fixture that could reveal a weakness hidden by the three workshop facts. State the expected evidence boundary and how you would inspect results.
Check your answer against these points:
- Describe a varied case
- Name the expected evidence
- Separate retrieval from answer review
Compare with a sample answer
A case can place a needed sentence across a heading or include a long table row. Label the complete evidence and inspect both ranked passages and any grounded answer; do not infer performance from the toy hit count.
Ungraded paper reflection: Design one held-out case. Use the cited source passage; this activity does not measure competence.
Practice 4
Write a compact record for one fictional ingestion run, using identifiers and counts rather than sensitive content.
Check your answer against these points:
- Include identity and both hashes
- Record policy, parameters and runtime
- Include counts and test outcome
Compare with a sample answer
Include source ID, raw and cleaned hashes, policy and implementation revision, exact window parameters, runtime, chunk count, total word occurrences, test outcome and the source revision reference.
Ungraded paper reflection: Draft a reproducible handover. Use the cited source passage; this activity does not measure competence.
Practice 5
Describe how a team could replace an index generation and what must happen if validation or switching fails. Mark what this lab actually demonstrates.
Check your answer against these points:
- Prepare and validate before switching
- Define obsolete-data handling
- Include a rollback route and lab limitation
Compare with a sample answer
Prepare and validate the replacement generation before switching readers. Define obsolete-chunk removal and rollback. The lab makes no index writes or deletion and proves none of those lifecycle properties.
Ungraded paper reflection: Plan a replacement generation. Use the cited source passage; this activity does not measure competence.
Practice 6
For a cleanly extracted but withdrawn fictional policy, list separate freshness and permission checks. Add one rule for retrieved instructions.
Check your answer against these points:
- Check source revision
- Check access at request time
- Treat retrieved text as evidence data
Compare with a sample answer
Confirm the current source revision and served-request authorisation independently of text quality. Keep derived data in the real lifecycle plan. Treat retrieved text as evidence, not executable authority; chunking itself does not solve prompt injection.
Ungraded paper reflection: Review source freshness and access. Use the cited source passage; this activity does not measure competence.