Open-source decision lab
Clarity before you commit.
See who acts and what it costs. About 15 to 30 minutes to set up a world; a few hours for the full report.
What it runs on and works with
A small cast, big on incentives. Boyun models people with stated goals, cash and information, so a decision can be inspected before anyone makes it.
roles in the bundled NEAR pilot example, sharing one $10,000 budget
possible actions in that example, eight of them a choice to wait
joint-action rules, where two choices together change the outcome
Four kinds of actor.
Meet themFrom question to receipt.
Six steps, the same ones the app walks you through. Setup takes about 15 to 30 minutes; the full report takes a few hours. You review the setup before anything runs, and nothing reaches NEAR without a separate approval.
Four things it does.
Build a world you can review, let a model make typed choices, settle the numbers in code, then keep a receipt if you want one.
Good questions to ask first.
Research notes.
All notesSee it before you commit.
Run the free example on your own machine. It needs no key and makes no model request.
Four kinds of people in a decision.
Every world in Boyun is a cast. We group them as creators, partners, market participants and adversaries. Each kind has different goals, different information and different things it can afford. The examples below come from the bundled NEAR pilot.
Built around one decision.
Four parts that work together: a world you can review, typed choices, numeric settlement and an optional receipt on NEAR testnet.
How every run works.
Review before it runs
You see the world, the actor-call limits and an input-cost estimate, then approve the run. Nothing starts on its own.
No silent retries
There is no automatic retry and no model fallback. Failed moves stay in the record instead of being smoothed over.
Keep what you made
Scenario slides, charts and private Markdown or PDF exports stay on your machine. They are private and not encrypted.
From question to receipt.
Six steps, one decision at a time, the same ones the app walks you through. Producing enough inputs and actors takes about 15 to 30 minutes; a full report takes a few hours, depending on the size of the world and the number of actors and agents in it. Every recorded world can then be replayed exactly.
What Boyun does
- Applies costs and effects in code, not in prose
- Enforces budgets and the bounds on every metric
- Keeps failed moves in the record
- Replays any recorded world exactly
What stays your call
- Approving each run and its estimated cost
- Choosing a model and bringing your own key
- Approving any NEAR commitment, separately
- The decision itself
Named for parting the clouds.
Boyun takes its name from “parting the clouds”: making participants, assumptions and possible consequences easier to see. It is an open-source research project by Cherry Q, previously developed under the working title Scenario.
What guides the work.
Say what’s assumed
Every motive, allocation and link carries a label. You can tell what was assumed from what came from a source, and change it.
Let code do the arithmetic
A model picks from allowed actions. Numeric code applies the costs and effects, so a well-written sentence can’t change a number.
Keep the decision yours
Boyun is a conditional decision tool, not a validated market forecaster or an automated trader. It shows the picture; you make the call.
Where it came from.
Boyun started as an AGPL-3.0 fork of MiroFish. The upstream application it carried is not shipped here, because Boyun never imported or launched it.
Open source
Released under AGPL-3.0. The NOTICE file and the upstream reuse notes say exactly what came from where.
Bring your own key
No paid hosted plan is active. Free examples need no key; real Jev decisions use a key you import for the session.
Written up in full
The whitepaper covers the method, the research it builds on, the experiments and the limitations.
Follow along.
Source code and issues
github.com/Boyun-Protocol/Boyun-core
Build notes and research threads
@qianCherry2102 on X
Licence
AGPL-3.0, with upstream attribution in NOTICE
Research notes.
Short notes on how Boyun models a decision, and what its records do and do not prove. The full method, citations and limits are in the whitepaper.
No notes on that yet.
Try another word or topic. The whitepaper covers the method in full.
The questions the agent asks.
A demonstration of the setup flow with sample answers. Nothing here builds a world or sends anything. The website only demonstrates; the full flow runs on your own machine.
The report time depends on the size of the world and the number of actors and agents inside it.
In this demo
- The real questions, in the app’s own order and wording
- Sample answers you can change and step back through
- What the app does after the last question
Not in this demo
- Building your world or proposing actors
- Running any agent, model or simulation
- Reports, results or NEAR receipts
Questions, answered.
What people usually ask before trying it. If yours isn’t here, open an issue on GitHub.
Download it. Run it yourself.
Boyun runs on your own machine; this website only demonstrates. Three steps are the same for everyone, then choose how you want to run it: in a browser, in a terminal, in Docker, from an AI agent, or with a model on your own computer.
Steps 1 to 3: for everyone.
Run every command from the Boyun folder you download in step 2.
Step 4: choose how to run it.
All five use the same engine and the same recorded results. Start with one; you can try the others later.
If something goes wrong.
This page drifted off.
The link may be old, or the page may have moved. Everything worth seeing is still one click away.