[workshop-sim] Improve: Add self-check recovery branches to the agentic-workflows intro activities

### Problem statement

`05-agentic-workflows-intro.md` combines the second-highest Part 1 dropout rate (20.0% conditional dropout across 37,024 at-risk simulated runs) with the second-lowest curriculum quality score in the entire 30-step corpus (5.43/10), and the lowest `active_learning` score of any step (2.4/10). Semantic review of the page found genuinely active elements (a predict-then-reveal two-file-structure exercise, an agentic-vs-standard classification task, and a draft-then-compare task-brief activity), but every one of them is **self-graded only** — a learner who misclassifies Activity 2 or writes a weak task brief in Activity 3 gets no corrective path back to the underlying distinction before the checkpoint. The dominant simulated failure categories at this step are `agentic-concept-gap` (5,235 failures) and `deployment-capability-gap` (2,172 failures), consistent with a scaffolding/recovery gap rather than an access barrier.

### Proposed change

Add a short "if you got this wrong" recovery branch after each `` reveal in Activities 2 and 3 (mirroring the existing "Still uncertain? Try this before moving on" pattern already used at the end of the page) that restates the distinguishing rule in one sentence and points back to the specific paragraph that explains it. Also make Activity 1 self-sufficient by adding a one-line fallback instruction for learners who do not yet have a compiled `.lock.yml` in their repo to inspect (e.g., a short annotated example block instead of requiring a live file).

### Failure mode classification

**Learning barrier** — the content is conceptually dense and the self-check loop lacks recovery support for a wrong answer, not a tooling/access/auth problem.

### Quantitative guardrail

- Current `overall_score`: 5.43/10
- Weakest rubric dimension(s): `active_learning` (2.4/10), `checkpoint_quality` (0.0/10, likely a rubric-detector gap since the page already has a `## :white_check_mark: Checkpoint` heading that the shared regex may not match — verify before treating this as a real gap)
- Current learning KPI index: `(2.0 × 2.4 + 2.0 × 0.0 + 1.5 × 5.0) / 5.5 = 2.24/10`
- Why the proposed change should raise these scores: adding graded recovery branches directly increases `active_learning` (more guided practice loops) and `scaffolding` (explicit remediation paths) without removing any existing activity or reducing conceptual scope, so `overall_score` and the KPI index should hold flat or improve, never decline.

### Acceptance criteria

- [ ] The shared curriculum quantitative assessment `overall_score` for `05-agentic-workflows-intro.md` stays flat or improves after the change
- [ ] The step's learning KPI index `(2.0 × active_learning + 2.0 × checkpoint_quality + 1.5 × scaffolding) / 5.5` stays flat or improves versus the current 2.24/10 baseline
- [ ] All three existing activities (two-file prediction, classification task, task-brief drafting) and the existing checklist checkpoint are preserved or strengthened, not removed or shortened
- [ ] Each of the three activities has an explicit "if you got this wrong" recovery branch with a one-sentence restatement of the rule and a link back to the source explanation

### Suggested owner

`workshop maintainer`
Related to #3447

> Generated by [🔬 Workshop Student Simulator](https://github.com/githubnext/gh-aw-workshop/actions/runs/35109821545) · copilot · auto · 164.9 AIC · ⌖ 7.67 AIC · ⊞ 15.2K · [◷](https://github.com/search?q=repo%3Agithubnext%2Fgh-aw-workshop+is%3Aissue+%22gh-aw-workflow-call-id%3A+githubnext%2Fgh-aw-workshop%2Fworkshop-student-simulator%22&type=issues)
> - [x] expires on Sep 17, 2026, 2:52 PM UTC

05-agentic-workflows-intro.md is the named entry point; read its three activities and the existing “Still uncertain?” recovery pattern first. Run the shared curriculum quantitative assessment for this step, then verify all three activities retain their exercises and each has a wrong-answer branch linking to the source explanation, with the scores and KPI index no lower than the baseline.

Written by the indexing model from the issue text.