reputation.ai-ministries.com · last updated 15 August 2026
The simulation runs entirely in your browser. Nothing is sent anywhere unless you turn on Contribute runs, and if you do, what gets sent is a set of numbers about a made-up town: dial settings, a seed, and the outcome figures. Nothing about you.
REPUTATION is a research simulation. It models how a population forms an opinion about newcomers, and it exists to produce a methods paper. The groups in it are called A and B, they are not any real people, and the model is not predictive. It shows that a dynamic can arise from a set of rules, never that it does arise anywhere in particular.
It is not advice, not a policy tool, and must not be presented as either.
A single run at one seed cannot be read on its own. If you watch a town turn against its newcomers, nothing on the screen tells you whether that is what these settings usually do or whether you happened to get an unusual one, and a model that cannot answer that question is a toy.
The answer is a set of runs at the same settings across many seeds — the calibration set. It is also what the paper's figures are built from, particularly the point where a population stops seeing individuals and starts seeing a category. That is thousands of runs, and one machine is a slow way to get them.
If you switch on Run calibration, your browser asks the server which parameter sets are still short of runs, works through a couple of them in the background, and posts back the resulting numbers.
In return you get two things:
REP-xxxxxxxxxx — recording which calibration cells were run, how many, on what day, against which engine version. It certifies the work, not the worker; there is no identity in it because none is collected. You can cite it, and anyone can verify the signature against this server.It is off by default, it only runs while the page is idle and never while the tab is in the background, and you can turn it off at any time.
Plainly: this is your machine doing work for the study. The band and the receipt are what you get for it, and both are real.
| Field | What it is |
|---|---|
config | The dial settings for that run. Chosen by the server, not by you. |
seed | The number the run was generated from. |
days | How many simulated days it ran. |
outcomes | The result figures: mean belief, hearsay fraction, confidence, gate pass rates, doors closed, fights, population. |
engine | A hash of the simulation code, so results are only ever compared against the version that produced them. |
That is the whole payload. Every value in it describes a simulated town.
This site sets no cookies.
It uses one entry in your browser's local storage, rep-consent, to remember whether you said yes or no to contributing runs, so it does not ask again on every visit. Clearing your browser data removes it and you will be asked once more. Local storage stays on your device and is never transmitted.
Turn Contribute runs off and nothing further is sent. Contributed runs carry no identifier, so an individual past submission cannot be located and deleted afterwards. If you would rather never contribute, leaving it off is the complete answer.
They go into the sweep behind the paper, and the aggregated dataset is intended to be published alongside it so anyone can check the work. Because every run is reproducible from its seed and config, any result can be re-run and verified independently, including by you. A random sample is re-run locally as a matter of course, and results that do not reproduce are discarded.
The simulation is a few hundred lines of plain JavaScript with no dependencies, and it is the same engine the paper is written from. You can read sim.js and app.js directly, including exactly what is sent and when.
This site is provided as is, without warranty. It is a research artifact and a demonstration. Do not rely on it for anything that matters.
Through ai-ministries.com.