AIMentor — AI Expertise Learning Platform

Question 1easy

In graph engineering, what makes an edge real?

The next step is written after the previous step in the scriptData from one node is consumed by another nodeBoth steps use the same modelThe workflow has more than three steps

Question 2easy

Why should a node have a clear contract?

So it can receive bounded input and return validated output that downstream nodes can trustSo it can use a larger context window than the parent agentSo every node can write arbitrary free textSo routing decisions become random and more creative

Question 3medium

What is the main risk of adding a cycle to an agent workflow?

It always makes the workflow sequentialIt can fail to converge and keep spending tokens or spawning work foreverIt prevents tools from being calledIt removes the need for verification

Question 4medium

When should you prefer a fan-in barrier?

Whenever it makes the code look more organizedOnly when the next stage genuinely needs the full upstream set togetherBefore every tool callOnly after a workflow has failed

Question 5medium

Which are good uses of deterministic code on graph edges?

Flattening and deduplicating structured resultsRouting based on a validated severity fieldRanking nuanced trade-offs with incomplete evidenceFiltering null results from failed parallel nodes

Select all that apply.

Question 6hard

In loop-until-dry discovery, what should the dedupe set contain?

Only confirmed findingsEverything seen, including rejected and unconfirmed findingsOnly findings from the most recent roundOnly findings produced by the cheapest model

Question 7medium

Name the topology: one node splits work, many nodes process in parallel, one node merges results.

Question 8hard

Give one reason to tier models across graph nodes.

Verification, Critic Panels, and Confidence Gates — Final Exam