Teaching component gallery
A reference fixture, not a lesson. Nothing here is course content: the questions restate definitions this resource has already fixed, the citation is a placeholder, and the page is absent from the sidebar and from search. It exists so that every teaching component and every variant is rendered somewhere, and a regression shows up here before it reaches a chapter.
The strip above the first paragraph is <SectionMeta />, reading this section’s
own minutes and bloom frontmatter. This fixture claims all six Bloom
levels so that the full set renders; a real page claims the one or two it
actually works at.
1. Section metadata
Section titled “1. Section metadata”<SectionMeta /> takes no props. It reads minutes and bloom from the
frontmatter of the page it is on, so an author never restates what the
schema already holds. On a page that declares neither field it renders
nothing at all, not an empty rule, not a placeholder, which is why there is
no “empty” fixture to show below.
2. Knowledge check · single choice
Section titled “2. Knowledge check · single choice”The default type. One correct option, and feedback on every option including the correct one, because the reader who picked it still needs to know why it holds. The explanation is the component’s body.
3. Knowledge check · multiple answers
Section titled “3. Knowledge check · multiple answers”Every correct option is required, and the hint says so before the reader commits. Selecting some but not all of them reports “partly right” rather than “wrong”, and the feedback for a correct option that was missed is shown alongside the ones that were selected.
Knowledge check
Correct.
Partly right.
Not quite.
The environment the agent experiences changes as its teammates learn.
Yes. Teammate policies were folded into the environment, and those policies are being updated, so the environment moves.
Experience collected earlier may no longer describe what happens now.
Yes. Earlier transitions were generated under teammate policies that have since changed, so they describe a setting that no longer exists.
The agent's own set of available actions grows as its teammates learn.
No. What changes is the behaviour of the teammates. The actions available to this agent are fixed by the environment, not by how well anyone has learned.
Explanation
Non-stationarity is a consequence of the modelling choice rather than an extra assumption on top of it: the moment teammates are folded into the environment, their learning becomes environment drift.
4. Knowledge check · prediction, no correct answer
Section titled “4. Knowledge check · prediction, no correct answer”A predict never grades. It records the commitment, shows the feedback for
the option the reader chose, and reveals the explanation. Nothing is ever
called incorrect, and no option here is marked correct, the value is
having committed before reading on.
Predict first
Prediction recorded.
Not knowing which agent's action earned the reward.
You are predicting that credit assignment binds first. Watch what happens when the team reward is high and one agent contributed nothing to it.
Teammates changing while each agent is still learning.
You are predicting that non-stationarity binds first. Watch what happens to experience an agent collected before its teammates moved.
Neither dominates; it depends on the setting.
A defensible position. Hold onto it and notice which of the two the sections that follow actually spend their effort on.
Explanation
There is no scored answer here, and the component does not pretend otherwise. The value is that the prediction was committed before the reading, so it can be compared against afterwards.
5. Knowledge check · prediction, free text
Section titled “5. Knowledge check · prediction, free text”A predict with no options at all becomes a free-text commitment. The answer
is stored as the string branch of CheckState, echoed back on submit, and
followed by the explanation.
Predict first
Prediction recorded.
You predicted
Explanation
Compare what you wrote against the definition this resource uses: a decentralized policy maps an agent’s own observation to an action, so anything outside that observation cannot influence the choice, however relevant it is.
6. Knowledge check · no explanation
Section titled “6. Knowledge check · no explanation”A check may end at its per-option feedback. Here there is no explanation
prop and no body, so the response is the verdict and the feedback rows and
nothing else. A check with neither feedback nor an explanation reveals
nothing and is warned about at build time, as is any incorrect option with no
feedback of its own.
Knowledge check
Correct.
Not quite.
the actions of all agents at one time step, taken together
Yes: one entry per agent, at a single time step.
the actions one agent takes over several time steps
No. That is a trajectory of a single agent, which is indexed by time rather than by agent.
the action a team agrees on before anyone acts
No. Nothing in the definition requires agreement or any communication at all. The joint action is simply what everyone did, collected into one object.
7. Research connection
Section titled “7. Research connection”The minimal form: an identifier but no canonical url, so the title is not
a link, and no claim block. A connection given neither url nor arxiv
renders the same way minus the identifier line, and is warned about at build
time on the grounds that a reader who cannot reach the paper cannot check
what it says.