[curriculum-eval] side-quest-11-06-anthropic-key.md: cognitive_load — 26 new concepts from redundant troubleshooting bullets in a short page

**File:** `workshop/side-quest-11-06-anthropic-key.md`
**Overall Score:** `5.39 / 10.0` (corpus mean: `6.14`)

**Flagged Dimensions:**

| Dimension | Score | Benchmark | Delta |
|---|---|---|---|
| cognitive_load | 7.2 | 10 (≤800 words, ≤15 concepts) | -2.8 (665 words is fine; 26 new concepts is the driver) |
| active_learning | 3.5 | ≥3 activity density | -0.5 (density 1.05 vs 3.0 ideal) |
| checkpoint_quality | 0.0 | 10 (≥4 checklist items) | -10.0 (scorer defect — see Root Cause) |

**Flagged issue `concept_overload` confirmed:** despite being only 665 words, the page packs 26 new concepts (`ANTHROPIC_API_KEY`, `engine: claude`, `api.anthropic.com`, `network.allowed`, repository secrets flow, key-rotation caveat, whitespace pitfalls, naming pitfalls, `copilot-requests: write` removal, `gh aw compile`, etc.) — nearly double the ideal.

**Root Cause (≤ 2 sentences):**
The page interleaves a step-by-step secret-creation walkthrough, four separate "common mistakes" bullet lists (wrong name, whitespace, closed tab, missing allow-list), and a frontmatter code sample in a short 665-word page, so new-concept density per 100 words is unusually high even though total length is well within budget. `checkpoint_quality` reads 0.0 due to the same corpus-wide `## :white_check_mark: Checkpoint` vs. `## ✅ Checkpoint` scorer mismatch; the actual checkpoint has 5 usable items.

**Evidence (quoted from the file):**
> - **Wrong name**: any variation (`anthropic_api_key`, `ANTHROPIC-API-KEY`, `CLAUDE_API_KEY`) will cause a silent auth failure. The name must be exactly `ANTHROPIC_API_KEY`.
> - **Copied with extra whitespace**: pasting from some tools adds a leading space. Delete and re-create the secret if you are unsure.
> - **Closed the Anthropic tab before saving**: you cannot retrieve the key again. Delete the key at [console.anthropic.com]((console.anthropic.com/redacted) and generate a new one.
> - **Network allow-list missing**: the `claude` engine needs outbound access to `api.anthropic.com`. Make sure it is in your `network.allowed` list (shown in the frontmatter example below).

**Learning Science Rationale:**
Sweller's Cognitive Load Theory notes that troubleshooting/error-avoidance content carries high intrinsic load per item because each "common mistake" requires the learner to hold both the correct procedure and its failure mode in working memory simultaneously; four such paired concepts stacked in one collapsible block, on top of the core setup steps, pushes this short page's concept count well past what a single pass can consolidate, even though raw word count looks modest.

**Improvement Prompt (for an agent):**
```
Open workshop/side-quest-11-06-anthropic-key.md. Consolidate the four "Common mistakes with this secret" bullets into a single compact table (Mistake | Fix) instead of four prose-heavy bullets, and remove the redundant restatement of the whitespace/naming warnings that already appear earlier in the page (in the "Store the key as a repository secret" numbered steps and the tip callout). Keep exactly one troubleshooting reference per failure mode. This should reduce new_concepts from 26 toward 16-18 without cutting any unique guidance. Run `npx --yes markdownlint-cli2 "**/*.md"` after editing and fix any lint errors.
```

**Expected Score After Fix:** `6.6 / 10.0`

> Generated by [🔬 Curriculum Quality Evaluator](https://github.com/githubnext/gh-aw-workshop/actions/runs/35026063971) · copilot · auto · 76.1 AIC · ⌖ 8.13 AIC · ⊞ 9.5K · [◷](https://github.com/search?q=repo%3Agithubnext%2Fgh-aw-workshop+is%3Aissue+%22gh-aw-workflow-call-id%3A+githubnext%2Fgh-aw-workshop%2Fcurriculum-evaluator%22&type=issues)
> - [x] expires on Sep 16, 2026, 9:39 PM UTC

Open workshop/side-quest-11-06-anthropic-key.md and review the numbered secret-storage steps, tip callout, and “Common mistakes with this secret” bullets. Consolidate the four mistake bullets into one Mistake | Fix table, remove redundant warnings while preserving unique guidance, then run npx --yes markdownlint-cli2 "**/*.md". Done means the page retains one troubleshooting reference per failure mode and passes the markdown lint check.

Written by the indexing model from the issue text.