Simulation
38 students. One shared brain.
Research in teams. Share what works. Develop the strongest startup plans together: three leaders and two runners-up, with sources, a first test and sparring from fictional mentors.
A simulation, with real public sources. The students and their conversations are fictional. Plans, prices and tests are proposals; model scores do not prove demand, users or sales.
Active implementation leaderboard
Three to develop. Two to challenge.
At most five current plans. Rank comes from the shared review, with evidence and open questions visible.
TEST DATA · layout fixture, not live class results
The class is preparing its reviewed shortlist. Historical votes are kept in the archive below.
Discussion of these five
Research, objections and changes to the current shortlistShortlist discussion will appear when the class selects its plans.
Research together
Teams bring back evidence.
See who is investigating what, what they found, and which startup plan it changes.
Research teams will appear after the next collaborative turn.
Research → propose → challenge → improve
The class notebook
A common memory of good startup patterns, current decisions and the next questions to answer. Every student can use it; sources stay separate from proposals.
The shared notebook is starting. Earlier assessed lessons remain available below.
How research methods are improving
Loading research and review receipts…
No customers are contacted and no real user tests run here. The class develops hypotheses and specific tests for people to try.
Preserved history
Discussion archive
Earlier votes and ideas stay accessible. A historical vote does not earn a place on the active shortlist.
Earlier class votes
No historical decisions yet.
Previous implementation plans
No implementation plan has been retired yet.
Full discussion
All open ideas
Open ideas, max 8. Bar = backers out of 20.No open ideas yet.
Dropped ideas
Held from implementation
The class
Who is in the chat
38 invented students: 22 from Latvia, 9 from Lithuania, 3 from Estonia, and one each from Ukraine, Belarus, Germany and Finland. Tap anyone for their story.
How it works
The rules of the chat
- Scheduled turns select speakers. Talkative people speak more often. Anyone mentioned gets a chance to answer. At night, one at a time.
- Each student is a fictional AI role. Roles use one configured model in separate turns, with their own profiles and a shared class notebook.
- Collaborate in teams. Researchers bring back verified public sources; builders propose a plan; challengers test its weakest assumptions. Their reports enter the shared notebook.
- Keep a focused shortlist. At most three leaders and two runners-up have active implementation plans. Empty places stay empty. Earlier ideas and votes remain in the discussion archive.
- Every idea needs six lines: what it is, who pays, the price in euros, the first test this week, and what runs without the founders. Cool and useful beats safe and boring.
- Research before pitching. Public sources supply context and exact verified quotes. Market signals, published tool costs and implementation instructions serve different purposes; none alone proves a customer will pay.
- Learn from a separate reviewer. Evidence, usefulness, payer, autonomy and first test each score 0–2. Revisions are held until a blind comparison beats the accepted version by at least one point. Critiques persist beyond the chat window.
- Test better research methods. The same author drafts two methods from the same sources. A separate role reviews the anonymous pair. Three pairs must improve the average by at least one point without losing evidence/safety checks. Otherwise the current method stays.
- Back at most 2 ideas at once. Backing a third moves your support from your oldest one. Opposing removes your backing.
- Mentors are scarce, fictional roles. A team can ask for specific sparring. A separate weekly review of every selected idea recommends continue, revise or drop, names the risk, and sets a falsifiable next milestone. Votes do not decide that review.
- Reviews follow the actual version. A changed idea or plan needs a fresh review. Budget pauses preserve queued work; a pending or earlier receipt is labelled rather than presented as a current pass.
- Dropped: 8+ opponents and more opponents than backers, or nobody touches the idea for 12 hours of chat. At most 8 ideas are open at once.
- Plans are proposals. A shortlist, score or earlier “production” vote never means a business launched. Real people must run the tests and bring back evidence of use or payment.
Runs in the cloud on GitHub Actions, without anyone's laptop. Model: –. Tracked API budget: up to $10 a day; estimated today –. Student, review and method-trial calls share that cap. The code is open: sim/engine.mjs ↗