Skip to content

[workshop-sim] Workshop Simulation Report — 2026-07-25 (Run #11000, 1000×Monte Carlo) #1998

Description

@github-actions

Overview

  • Date: 2026-07-25
  • Students simulated: 46 × 1,000 Monte Carlo runs (11,000 accumulated runs per student)
  • Workshop steps available: 31/31 main steps (plus side quests)
  • Overall success rate: 17.0% (95% Monte Carlo interval: 16.6%–17.3%)
  • Highest-dropout step: 07-first-workflow — 33.0% conditional dropout among 23,349 at-risk runs (95% CI: 32.4%–33.6%)
  • Lowest curriculum quality step: 07a-your-first-workflow-terminal.md (overall score 7.54/10)
  • Learning KPI index: 7.30/10 (active_learning 4.38 · checkpoint_quality 8.55 · scaffolding 9.52)
  • Model: unknown / 2026-07-assumption-model-v2 (parameter hash unknown)
  • Limitation: synthetic results reflect explicit model assumptions; intervals exclude model and population-assumption uncertainty

Part Summary

Part Files Mean Score Std Dev
Part 1 — core path (lessons 00–14) 19 8.99 / 10.0 ±0.77
Part 2 — advanced (lessons 15+) 12 8.77 / 10.0 ±0.40
Overall corpus 31 8.91 / 10.0 ±0.64

No steps are classified as other.

Critical Findings

  1. Step 07 is the single largest dropout gate (33% conditional, Part 1): workflow-authoring friction and Copilot billing configuration together account for the majority of failures at the most pivotal hands-on step. The terminal file 07a-your-first-workflow-terminal.md scores 7.54/10 overall with the lowest checkpoint_quality (5.0) in the corpus — learners advance without sufficient feedback that their compile passed correctly.
  2. Conceptual overload through Steps 04–05c blocks 37% of at-risk learners before they reach any hands-on step: agentic-concept-gap and concept-overload are the two most common failure categories across all 46,000 runs. active_learning scores for 05-agentic-intro (2.9) and 05b-agentic-security (3.2) are well below the corpus mean.
  3. Learning quality health is mixed: for learners who stay through the workshop, the learning KPI index (7.30/10) is acceptable — scaffolding (9.52) and checkpoint_quality (8.55) are strong — but active_learning (4.38) is consistently the weakest dimension across both Part 1 and Part 2. Most exercises are passive or observational rather than requiring learners to produce artifacts.
  4. All five highest-dropout steps are in Part 1 (lessons 00–14): the core path has the heaviest completion burden. Repairs to Part 1 will have the largest impact on overall workshop completion.

Top Repairs to Prioritize

Note: some student dropout is expected and acceptable. Repairs must maintain or improve the learning KPI index — do not lower the cognitive bar or remove practice to chase headline completion numbers.

  1. Strengthen Step 07a checkpoint and compile feedback — add an intermediate checkpoint after the first compile succeeds and before billing selection, so learners confirm the lock file was generated correctly (completion impact: ↑ · learning KPI impact: ↑)
  2. Add active exercises to Steps 05 and 05b — replace or supplement the passive reveal-answer format with one short produce-an-output task per concept page (e.g., draft one line of frontmatter, classify a scenario) to increase the active_learning dimension from 2.9/3.2 toward the corpus mean (completion impact: ↔ · learning KPI impact: ↑)
  3. Add a test-path section to Step 14b — explicitly instruct learners to open a draft PR to trigger the new PR reviewer workflow and verify a comment appears, with a short troubleshooting checklist for when the trigger does not fire (completion impact: ↑ · learning KPI impact: ↑)
Dropout by step
Step At-risk runs Conditional dropout 95% CI Failure mode Top reason
07-first-workflow 23,349 33.0% 32.4%–33.6% Access + Learning barrier Workflow authoring friction in Copilot Chat and billing path ambiguity
14b-pr-reviewer-workflow 13,794 25.3% 24.6%–26.0% Learning barrier Copilot skill guidance missing for event-driven workflow creation
05-agentic-intro 40,019 19.9% 19.5%–20.3% Learning barrier Conceptual gap between agentic workflow and standard Actions concepts
05c-agentic-practice 32,057 14.2% 13.8%–14.6% Learning barrier Cannot correctly classify agentic workflow components in exercises
05b-agentic-security 27,504 10.6% 10.2%–11.0% Learning barrier Security model concepts (sandboxing, safe-outputs) not fully absorbed
04-actions-intro 44,714 10.5% 10.2%–10.8% Learning barrier Concept overload for beginners with no prior Actions experience
12-test-and-iterate 14,561 5.3% 4.9%–5.6% Learning barrier Test and iterate cycle friction — unclear how to observe and modify
06-install-gh-aw 24,588 5.0% 4.8%–5.3% Access barrier Extension install friction in Codespace environment
17-add-mcp-tools 9,858 4.4% 4.0%–4.8% Access + Learning barrier MCP tooling configuration complexity
09-agentic-editing 15,081 3.4% 3.2%–3.8% Learning barrier Workflow editing friction when revising task briefs
Learning quality KPIs
Step file Overall active_learning checkpoint_quality scaffolding Learning KPI Repair priority
07a-your-first-workflow-terminal.md 7.54 4.6 5.0 10.0 6.22 🔴 High
05-agentic-workflows-intro.md 7.88 2.9 10.0 10.0 7.42 🔴 High
02a-setup-codespace.md 7.90 5.3 10.0 5.0 6.93 🟡 Medium
05b-agentic-workflows-security.md 8.64 3.2 10.0 10.0 7.53 🟡 Medium
14b-pr-reviewer-workflow.md 8.64 4.3 10.0 10.0 7.93 🟡 Medium
08-run-your-workflow.md 8.44 3.0 10.0 10.0 7.45 🟡 Medium
17-add-mcp-tools.md 8.36 3.1 10.0 10.0 7.49 🟡 Medium
15-conditional-logic.md 8.20 3.8 10.0 10.0 7.75 🟡 Medium
20-persistent-memory.md 8.25 3.7 10.0 10.0 7.71 🟡 Medium
07-your-first-workflow.md 9.37 0.0 0.0 10.0 2.73 ⬜ Nav page
06-install-gh-aw.md 10.0 0.0 0.0 10.0 2.73 ⬜ Nav page
00-welcome.md 10.0 0.0 0.0 5.0 1.36 ⬜ Nav page
01-prerequisites.md 10.0 0.0 0.0 5.0 1.36 ⬜ Nav page
Cohort mean 8.91 4.38 8.55 9.52 7.30
Curriculum quality metrics
Step file Overall Lowest rubric dimension Recommended repair focus
07a-your-first-workflow-terminal.md 7.54 checkpoint_quality (5.0) Add intermediate compile-success checkpoint before billing section
05-agentic-workflows-intro.md 7.88 active_learning (2.9) Replace reveal-answer exercises with one produce-an-output task
02a-setup-codespace.md 7.90 scaffolding (5.0) Add step-by-step scaffold for Codespace first-launch verification
15-conditional-logic.md 8.20 active_learning (3.8) Add one branching scenario exercise learner must solve themselves
20-persistent-memory.md 8.25 active_learning (3.7) Add a short write-the-memory-entry task before the reference example
05b-agentic-workflows-security.md 8.64 active_learning (3.2) Add a classify-the-safe-output exercise at end of security concepts
14b-pr-reviewer-workflow.md 8.64 active_learning (4.3) Add trigger-verification test step and troubleshooting checklist
08-run-your-workflow.md 8.44 active_learning (3.0) Add an observe-and-interpret task for first workflow run output
Segment breakdowns

By technical level

Level Students Mean success rate
beginner 11 0.2%
github-basic 19 8.8%
actions-user 11 39.0%
advanced 5 36.3%

By personality

Personality Students Mean success rate
confused 6 19.3%
curious 15 14.6%
impatient 6 18.4%
methodical 12 17.4%
skeptical 7 18.1%

By UI preference

UI preferred Students Mean success rate
Yes 22 5.4%
No 24 27.5%
Notable student journeys (3)

Surprising success — Learner 026 (advanced/confused/devops): Despite a confused personality that often correlates with lower completion, this DevOps engineer with prior Actions experience achieved a 56.8% success rate. Their prior actions-user-adjacent knowledge of YAML triggers and permissions meant the conceptual sections were near-instant skips; their confusion manifested only at billing configuration, which they resolved through trial and error. Advanced technical background more than offset the personality penalty.

Unexpected dropout — Learner 006 (actions-user/impatient/backend-dev, SR=0.0%): Although Actions-familiar, this learner's impatient personality combined with 14b-pr-reviewer-workflow as the most common failure step suggests they consistently skipped the prerequisite model-access step, then hit an event-trigger failure they did not know how to debug. The lack of a guided test path in Step 14b meant they could not recover without re-reading earlier steps — an action impatient learners rarely take.

Content-gap case — Learner 008 (github-basic/enterprise-dev, SR=8.4%): This enterprise developer repeatedly dropped out at 05-agentic-intro despite having GitHub experience. The agentic-concept-gap failure category dominated their run history, pointing to a genuine conceptual gap: the enterprise Actions mental model (deterministic YAML, fixed runners) does not map well to the agentic reasoning loop described in Step 05. Adding a concrete enterprise-scenario analogy (e.g., "replace your hardcoded JIRA sync script with a task brief") to Step 05 would directly address this profile's barrier.

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

  • awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by 🔬 Workshop Student Simulator · 64.5 AIC · ⌖ 5.35 AIC · ⊞ 10.5K ·

  • expires on Jul 26, 2026, 4:07 AM UTC

Metadata

Metadata

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions