Map, Premise, Project Plan: Alex (Three Seconds)

Premise

In this story, a broke co-founder wants to save the startup their friends have bet everything on, but can’t do it because they’ve found out the app lets anyone copy a stranger’s voice easily.

It’s 1:40 a.m. In a few hours, you and three friends vote on whether to launch Echo, an app that clones a voice from three seconds of audio so that people who’ve lost theirs can talk again. Two nights ago you got past its consent check using words cut from your own old voicemails, and you haven’t told anyone. The vote has to be unanimous, and you go last.

I made the danger people misusing a real tool instead of an AI going rogue, because a skeptical Stanford student can wave off a killer robot but not a scam call that already happens. Nobody in the room is a bad person: Maya has been covering your rent, Jordan is observing that you only have seven to nine weeks of cash before you run out, and Dev has already heard his uncle read a birthday card in the voice ALS took from him. Good people with real reasons can still ship something they know is broken, because the incentives (a Friday investor deadline, rent that’s eleven days late) push everyone the same way. Overall, I want players to feel how reasonable yes is before they decide whether to be the one who breaks the unanimity.

  • Player role: you’re one of the four co-founders, and the only one who knows the consent check can be fooled.
  • Player goals: keep the company (and your friends) afloat without shipping a tool that lets strangers talk in anyone’s voice.
  • Player conflict: speaking up might lose you funding, and Maya, the friend who carried you, has the most to lose.
  • Player choices: speak up or wait; push back when Dev says cloning already exists, or let it go; then vote yes, no, or wait two weeks.
  • Player actions: look at three objects in your bedroom (each one opens a memory), talk in the meeting, and cast the last vote.
  • Resources: what you know; Maya’s backing, which you only get if you spoke up; time and money; and an outside cheap test report.
  • Game events: three memories (a friend’s kindness, being bullied with your own cut-up voice, a stranger’s kindness), then Dev’s challenge, Maya’s case for shipping, the vote, and one of three endings.

For scope: two rooms, three remembered places, three friends, three or four real decisions per playthrough, and three endings, built in Ink and aimed at 10-15 minutes.

Map 1: where the story happens

The story starts in your room, reaches back into three memories, pulls everyone into one shared room for the vote; the decision happens between four friends, and people none of them will meet face costs.

Map 2: how the choices branch

Speaking up in the meeting is what makes Maya back you, and only her backing turns wait two weeks into a real option at the vote.

Project plan

The design and a rough playable draft were done on 10/21, so most of what’s left is making every word mine and testing the game on people who haven’t heard me explain it.

  • 10/21 (done): premise decided; playtest draft v0.1-v0.3 built in Ink; playtest 1 with tester A, with the three must-fix items in v0.3.
  • 10/24 (done): both maps and this post.
  • Sun 10/25: rewrite the opening, bedroom and meeting in my own words (v0.4).
  • Mon 10/26: refined prototype in class, with an in-class playtest.
  • Tue 10/27: fix what the playtest found; playtest 2 with someone new.
  • Wed 10/28: playtest 3, where I don’t explain anything while they play, recorded with timestamps.
  • Thu 10/29: fix what the playtest found
  • Fri 10/30, 11:59 PM: final post with a playable link and playtest quotes.
  • Sat 10/31, 11:59 PM: reflection.

Challenges

I only have one tester so far, and I explained the premise while they played, so the opening probably looks clearer than it is. The first tester never mentioned Larch, the outside firm that only tested what the company paid for, even though that’s the idea I care about most; v0.3 puts its report on screen, but that’s untested.

Only Maya felt like a person, so Jordan and Dev each need one personal moment or it’s mostly Maya. Wait might be the easy way out and if most players pick it, the vote isn’t too hard. Overall, most of these come back to one fix: more players who haven’t heard me explain the game.

AI use disclosure

  • Claude Opus 5.5 (Anthropic), in Claude Code: wrote the game’s code (Ink with inkjs), the build script and a test that walks every path; kept my design log; and ran research reports on interactive fiction design. For the first playtests (v0.1-v0.3, 10/21), it drafted all the passage text and choice labels at my request.
  • Rules that started as AI suggestions, which I accepted: voice cloning as the technology; launch needs every founder’s yes; Maya backs you only if you spoke up; the rent at stake is Maya’s; no timer and no shown vote percentages; three friends instead of four; and the fixes after playtest 1.
  • GPT 6 Astra (OpenAI): critiqued playtest draft v0.1, and I applied its changes in v0.2 (stated costs for each vote option, the outside tester, and Maya making her own case).
  • Claude Opus 5.5, in claude.ai: drew both maps from the game file and wrote this post, from my design log, playtest notes and the game file.
  • Mine: the premise (friends at a broke startup vote on a launch while I privately doubt it), misuse instead of rogue AI, survival money instead of a big lab, being the first to ship, keeping the bullying memory, having an ally and the rent at stake, showing visuals up front, and every final call in the design log.

About the author

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.