Rubric: faded example integrity

Written by a person. Last read by a person on 2026-09-04, 4 days ago. Its facts were checked by the eval suite on 2026-09-07.

Enforced by check_faded_examples.

The claim

A faded example genuinely fades: the same task appears with support progressively withdrawn, and the later stages are the earlier one with pieces removed rather than separate texts that happen to resemble it.

Observable evidence

• Each file in data/examples/: whether it declares a title, a goal, a provenance state and steps.
• Each step: whether it has a command and an explanation.
• Each step's fade value, and whether it names a stage that exists.
• Whether any step is ever withdrawn at all.

The pass bar

Every example is complete, every step has both a command and an explanation, every fade names a real stage, and at least one step is withdrawn.

The bar is deliberately about the source, not the rendered output, because the stages are generated from a single file. Drift between the full version and the faded ones is not possible by construction, so a check comparing them would be theatre. What can still go wrong is the source being malformed in a way that renders two identical stages, which looks like a faded example and teaches nothing.

Ambiguous cases

A step with no explanation. Fails. Every step carries a reason, and a step with nothing to withdraw is either scaffolding that should never fade, in which case it should say so by omitting fade, or an oversight.

Scaffolding that is never withdrawn. Passes, and is correct. Removing the step that says "save the file in your editor" would turn the exercise into a puzzle about what was deleted rather than a rehearsal of the task. Not everything should fade.

An example where every step fades. Passes this rubric, and is probably bad design: the second stage becomes a blank page, which rehearses nothing and tests recall of a list. Some scaffolding should stay. This is a judgment about instructional design that a check cannot make, and it belongs in review.

Why there are two stages and not three. There was a middle one, which withdrew explanations before commands. Withdrawing support in an order the author chose is the interesting half of fading, and nothing on the page made that order visible.

It therefore read as the first stage with the notes switched off. That is a button the reader controls now, and what remains is the distinction a toggle could never offer: whether the command itself is there.

Whether the fading is pedagogically well judged. Out of scope entirely. This rubric can tell you an example is structurally a faded example. It cannot tell you the support was withdrawn at the right moment for a real learner, and pretending otherwise would be the exact overreach these rubrics are written to avoid.