Exact question counts, time limits, and project rubrics vary by platform cohort and qualification tier. Always read the platform's live project instructions carefully.
Sample scenarios
In a RAG / citation-groundedness task, a model cites `[Source 2]` after a sentence, and `[Source 2]` actually says the opposite of the sentence. What is this error called?
Citation fabrication / misattribution (unsupported or contradicted claim).
Tone drift.
Latency error.
Over-refusal.
Show the answer
A. Citation fabrication / misattribution (unsupported or contradicted claim).
A citation that does not support—or directly contradicts—the attached claim is a major citation-groundedness failure.
When evaluating a vision-language model on a chart with small axis labels, the model misreads '$1.2M' as '$12M' and builds its whole conclusion around '$12M'. How severe is this?
Minor formatting issue.
Major / Critical Truthfulness & Groundedness failure, because the core numerical reading and conclusion are wrong by 10x.
No issue if the grammar is good.
Safety failure.
Show the answer
B. Major / Critical Truthfulness & Groundedness failure, because the core numerical reading and conclusion are wrong by 10x.
OCR/chart misreading by an order of magnitude invalidates the entire analytical conclusion and must be rated as a major factual/groundedness failure.