[curriculum-eval] corpus: Bloom distribution — core path overweighted toward Apply, higher-order thinking concentrated in side quests #4530

File: Corpus-level (93 workshop pages) — Bloom's Taxonomy distribution
Overall Score: N/A — distributional finding (corpus mean: 6.09 / 10.0)

Flagged Dimensions:

Root Cause (≤ 2 sentences):
The rubric's bloom_gap_analysis confirms the corpus technically clears the minimum thresholds (has_min_understand_steps: true, has_min_analyze_steps: true, 7 Analyse + 5 Evaluate + 8 Create steps total), but the raw distribution (remember 8, understand 15, apply 50, analyze 7, evaluate 5, create 8) shows more than half of all pages classified as "apply" — mostly procedural install/run/configure steps — while higher-order thinking (Analyse/Evaluate/Create) is concentrated almost entirely in a handful of side quests (e.g. 09-agentic-editing.md at Evaluate, 07-your-first-workflow.md and 14b-pr-reviewer-workflow.md at Create) rather than distributed across the core path.

Evidence (quoted from the file):

"bloom_distribution": {"remember": 8, "understand": 15, "apply": 50, "analyze": 7, "evaluate": 5, "create": 8, "unknown": 0}

Learning Science Rationale:
A healthy Bloom's pyramid expects a broad Apply base but a steadily narrowing taper toward Analyse/Evaluate/Create; here the base is disproportionately wide (53.8% Apply) relative to the taper (21.5% combined higher-order), meaning most of the core path (steps 00–14) asks learners to execute instructions correctly rather than to compare, critique, or compose — skills that transfer better to real-world agentic workflow authorship. Anderson & Krathwohl's revised taxonomy treats "apply" as necessary but insufficient for durable expertise; without enough Analyse/Evaluate/Create practice distributed earlier in the sequence (not only in late side quests), learners risk being able to follow the workshop's exact steps but unable to adapt a workflow to a novel task once the workshop ends.

Improvement Prompt (for an agent):

Review the core workshop path (workshop/00-welcome.md through workshop/14-next-steps.md plus
14b-pr-reviewer-workflow.md). Using the per-file bloom_level values in
/tmp/gh-aw/data/rubric-results.json (or by re-running the corpus assessment), identify 2-3 core
steps currently classified as "apply" that could be elevated to "analyze" or "evaluate" with a
small addition, without changing their primary learning goal. For each candidate step, add one
of the following to its existing content (not a new checkpoint item, an actual exercise):

- An "Analyse" prompt: present two example outputs (e.g. two workflow run logs, or two frontmatter
  snippets) and ask the learner to identify which is correct and why, before revealing the answer
  in a <details> block.
- An "Evaluate" prompt: ask the learner to judge whether a given workflow configuration meets a
  stated requirement (e.g. "does this permissions block follow least-privilege?") and justify
  their judgment, before revealing the expected reasoning.

Prioritize steps 08-run-your-workflow.md and 08b-interpret-your-run.md, since they currently sit
at "apply" but their subject matter (reading run output, interpreting results) is naturally suited
to an Analyse-level exercise. Do not remove any existing content; add the new exercise as an
additional subsection before the Checkpoint. Re-run
npx --yes markdownlint-cli2 "workshop/*.md" after editing.

Expected Score After Fix: No single file's overall_score is directly affected (Bloom level is diagnostic, not a weighted rubric dimension), but corpus-level pedagogical balance improves, targeting an Apply share closer to 40–45% with Analyse/Evaluate/Create rising from 21.5% to 28–30%.

Generated by 🔬 Curriculum Quality Evaluator · copilot · auto · 124.3 AIC · ⌖ 8.34 AIC · ⊞ 10.4K · ◷

• expires on Oct 14, 2026, 4:56 AM UTC