Phase 4 · A Game / Simulation

A Game or Simulation

Use Game if your story is a tough decision. The player is the figure, choosing — the player feels the weight of the choice your character once faced. You brief the AI in plain English, never look at the code, and iterate by playing.

Duration · ~2h 30minSteps · 4Output · Playable web game + Creator's Note + AI Audit
1. Set up your game

Use Game if your story is a tough decision. The player is the figure, choosing — the player feels the weight of the choice your character once faced.

AOpen your project in your AI workspace

  • Start a new thread. Give it a clear name: Untold Stories Game — [your character's name].
  • Paste in: your character's name, age, role, time, place, the museum object.
  • Add a short description (3–4 sentences) sharing your character, object and story angle.
  • Upload your first character image from Phase 1. This is your starting point.

BFind the decisions in your story

Re-read your storyboard. Where does your character face a tough choice? Pick 3 moments where the player will choose on the character's behalf.

Each decision should:

  • have at least two real options (no obviously-right answers)
  • fit the period and the character's knowledge
  • have a consequence the player will feel later in the game

CWrite your intention

Use the template below. Replace the bracketed slots. Start in PLAN mode.

Game intention template
I want to build a simple 2D indie game about [CHARACTER NAME], a [age] [role] from [TIME / PLACE]. The story arc: [1-3 sentences. What happens? What does the player learn?] In the game, the player faces these decisions: - [decision 1, e.g. "do you write Nanni's anger in cleanly, or soften the venom to keep his goodwill for next time?"] - [decision 2, e.g. "do you charge the full clay-tablet fee, or accept less to keep a regular customer?"] - [decision 3, e.g. "do you tell your master what you wrote today, or stay silent?"] Each decision has consequences the player feels later in the game. It should feel: [mood - e.g. "slow, contemplative, period-warm tones, hand-drawn pixel art"]

DAsk the LLM to coach you

Continue the conversation and prompt the LLM:

Coach prompt
Be a coach and ask key questions to help me design this gaming or simulation experience.
CHECK
  • Character named in the intention?
  • Story arc described in plain language?
  • 3 real decisions listed (each with at least two options, no obviously-right answer)?
  • Mood / look described in 1–2 lines?
  • If unsure: "We are not sure if our intention is specific enough because…"
2. Build Version 1

Brief the AI and get the first playable version. Don't aim for perfect — aim for something you can play. Refining comes next.

ASend your intention

Paste your intention to the AI and add:

Build instruction
Build this as a webpage. Use HTML5 canvas + JavaScript. Make it self-contained — all code, styles, and assets in one HTML file. Publish it as a webpage when done and give me the URL. I'll iterate by playing it and telling you what to change.

BOpen the URL and play

Open the URL in a new tab. Play your game. Don't read the code. Your job is the experience.

CIf something fails immediately

Tell the AI specifically what's wrong:

  • "the game won't load"
  • "I don't see my character anywhere"
  • "there are no decision moments yet"

Ask it to fix and re-publish to the same artifact. Refresh your tab and try again.

When 2 things fail the same way — fix the brief

This is one of the most important moments in the whole pathway. Most groups will hit it once — and that's exactly the point.

If the AI keeps making the same mistake (modern building in the background, wrong period clothing, controls that don't fit your character), stop iterating on individual fixes. Go back to your intention from Step 1 and add a "DO NOT" line.

Then send:

Re-build with updated brief
Re-build using this updated brief. Keep the existing functionality. Fix the [specific issue].

You've just discovered the most useful AI workflow lesson there is: the brief matters more than the individual fix.

CHECK
  • The AI published a working URL?
  • The basic shape of what you asked for shows up?
  • At least one thing you can interact with?
  • You can see roughly how the decision moments will appear?
3. Play, iterate, audit

Now you refine. Play it, see what works and what doesn't, describe changes in specific language. The AI re-publishes to the same URL.

APlay and find issues

Ask yourself:

  • Does my character do what I asked?
  • Are the decisions clear? Do they actually feel like choices?
  • Does the story arc come through?
  • Is anything confusing?

BDescribe changes specifically

Good
"When I press space, my character should drop the clay tablet, but right now nothing happens. The apprentice in the corner is wearing a t-shirt — change to a simple linen wrap. The dialog box at the bottom is cut off."
Vague
"Make it better. The game feels off."

The AI re-publishes to the same URL. Refresh your tab. Play again. Describe again.

CRun the AUDIT every few iterations

The audit matters more for history than for fun.

  • Perspective — from your character's authentic point of view?
  • Evidence by ID — every claim cites a specific evidence ID (F1, SG2) from your Phase 2 list. No ID = doesn't go in.
  • No modern slips — in art, dialogue, mechanics, controls, fonts.
  • Real decisions — does each choice have a real cost? Or is one option obviously right?
  • Dignity — character feels real, not a stereotype.
  • Visitor clarity — a stranger could pick this up without instructions.
CHECKRun after each iteration
  • Same character throughout?
  • Audit checklist still passes?
  • Decisions feel real, not fake?
  • Does it feel finished, or is something still missing?
4. Polish & publish

All the decisions work. Now you wrap the game with a title, a closing screen, the Creator's Note + AI Audit. Without these, you have a webpage. With them, you have an Untold Stories deliverable.

AAdd a title screen and closing screen

Add title and closing screens
Same artifact. Add: 1. A title screen: "[Story Title] — by [your group's name]" with a "Begin" button. 2. A closing screen with the Creator's Note + AI Audit plus a clear line: "This was made with AI assistance."

BFinal test

  • Play the whole game one more time.
  • Anything still breaks? Fix it.
  • Copy the share link. This is what you hand to your audience.

When you finish early

  • Add more decision moments — deepen the story with one more meaningful choice.
  • Show consequences mid-game — a master's mood shifts based on earlier choices, dialogue changes.
  • Stats screen — show how the player's choices compare to other groups.
  • Multilingual — character voice in their plausible historical language.
  • Project gallery — one webpage hosting every group's finished game.
  • Save your decision-design as a skill — so other groups can build a game with the same shape, swapping in their character.
CHECKFinal review
  • Title screen + closing screen visible?
  • Creator's Note + AI Audit in the artifact?
  • Marked as AI-generated?
  • Share link works in a fresh tab (not logged in)?
  • One clear story, one clear perspective, honest uncertainty?