Network Change Roulette
Challenge the change bot across 100 firewall, routing, Zscaler PZEN and PAC tickets. Investigate the pilot network, spend limited checks, then Approve, Hold or Rollback and compare the resulting service and access outcomes.

The builder
- Name
- CodeByDannyV
- First spec project?
- Yes
- Began at
- NAF AC6
How it was built
- Model
- unknown+ ImageGen (UI mockups, exact model ID unknown)
- Harness
- Codex 0.167.0-alpha.4
- Spec method
- GitHub Spec Kit 1.4.3
- Spec artifacts
- constitution, spec, clarify, plan, tasks, analysis
- Tasks
- 7 in 20 loops
- Built
- November 19, 2026
Build environment
- MCP servers
- codex_apps
- Sources & RAG
- Creator's AutoCon 6 handoff and the workshop StarterKit
- Network Automation Forum AC6 workshop lab guide
- Automate Your Network and NetBox Labs (creator-supplied visual references)
- RFC 4632, section 3.1 (CIDR notation)
- NIST SP 800-41 Rev. 1, Guidelines on Firewalls and Firewall Policy
- Natural Earth 1:110m coastline data
- Terraform AWS provider documentation for routes
- Zscaler ZIA Terraform provider documentation for URL filtering rules, URL categories and firewall filtering rules
- MDN Proxy Auto-Configuration (PAC) file reference
- Skills
- cloud-environment-onboarding:setup
- cloud-environment:cloud-environment-runtime
- cloud-environment:cloud-desktop
- game-studio:game-playtest
- Other tools
- Git
- GitHub Spec Kit CLI
- Node.js
- Python
- Playwright
- Chromium
- Pillow
- python-hcl2
- ImageGen
- curl
- ripgrep
- AI assets
- Three design mockups of the retired-subnet screen: amber change ticket, purple roulette wheel, and teal traffic path. These were visual references during development.
Runtime: how to play
- Type
- static-web · HTML, CSS, ES modules
- Start
- python3 -m http.server 8000 --bind 127.0.0.1
- Then open
- http://127.0.0.1:8000/
- Install needed
- No
- Internet needed
- No
- API key needed
- No
- AI at runtime
- No
- Tested on
- Chromium
- Saves
- Change Bot Duel saves in localStorage, with in-memory play if storage is unavailable. The included classic campaign also supports sessionStorage fallback.
- Accessibility
- Keyboard controls, reduced-motion handling and phone layouts are covered by the Playwright persona checks.
Verification
- Tests
- 19 passing · node tests/run.mjs (19 model checks); node tests/duel-model.mjs (1,200 scoring combinations). Arcade review also independently checks all 404 branch outcomes.
- Facts from
- AWS Terraform provider documentation, Zscaler ZIA Terraform provider documentation, MDN PAC file reference, Natural Earth 1:110m coastline
- Known limits
- Service workflows, regions, user counts and recovery estimates are fictional fixtures
- The rehearsal and recovery matrix is an educational rubric
- Multi-service API path tracing is not implemented
- The original ARCADE.md describes the earlier 20-round campaign. The current default is Change Bot Duel: 100 tickets, three calls per shift, and topic-based ticket selection.
- Creator name and publication consent are unfinished in the submission; the public GitHub handle is credited. Exact model, build duration and usage remain unrecorded.
- The 30-minute estimate is creator-supplied for the earlier campaign. Physical-phone, spoken-screen-reader and human-learning checks are unverified.
- License
- Apache-2.0
Who it is for.
Network Change Roulette puts you in the change-review seat. In Change Bot Duel, inspect a firewall, route, Zscaler PZEN or endpoint PAC change, challenge the bot’s recommendation, and decide whether to approve, hold or roll back. Each ticket offers one free investigation; each three-call shift supplies two reserve checks and one double-down chip. A replay compares both decisions with service and access intent. The earlier forecast campaign and its globe remain included.
THE BUILD STORYHow it was made.
We started with the workshop constitution and wrote the specification around a single firewall ticket. Tucson needed to lose access while Phoenix stayed connected. The clarifications defined a fresh HTTPS request and gave “Neither branch” its own choice so a blank answer could never earn credit. Next came the implementation plan with its policy analysis, then the task list. Hand-worked examples gave us expected answers before we wired the evaluator to the roulette wheel.
The creator asked for an explorable globe and a way to assess rollout risk alongside branch impact. The assessment atlas records twenty architect and change-lead review passes. One pass caught a grading bug. Choosing a risk rating cleared “Neither branch.” We restricted the clearing behavior to branch checkboxes, then submitted a zero-impact forecast in Chromium to verify the fix.
The creator later asked to put decisions before supporting evidence and divide the campaign into five-ticket rounds. Routing, forward-proxy and workstation PAC tickets added different ways for a connection to fail. The final solo build kept the change-board decision and skill drills after the creator asked to remove teammate play. The README and design specification document a later Change Bot Duel edition with 100 tickets and a three-call shift budget. During Arcade integration, the creator’s original submission was retained, an actual-game screenshot was captured, and the listing was updated for that default edition. Saved-record validation and a module-load retry screen were added; network evaluation and scoring rules were preserved.
ADVICEFor the next builder.
- Write down at least one correct case, one plausible wrong choice and one edge case before implementing, and turn them into tests.
- Grade the network outcome independently of the player's prediction so explanations stay trustworthy.
- Record the model ID, MCP servers and timings during the build so this file can be completed honestly.
Built a game of your own?
Fill in the ARCADE template with your build agent and send me your GitHub URL.
Add your game →