Skip to content
Open Choice

Open Choice

A plugin-based workbench for choice-modeling research — design, simulate, estimate, and compare models in one place.
Install Your first plugin

One file. Four steps. Fully reproducible.

Every analysis is an .oce file you commit alongside the data. Share the file, rerun it a year later, or hand it to an AI assistant and get an exact answer back.

1. Design

Generate an experimental design — MaxDiff, CBC, conjoint — with the parameters you actually want, not the defaults the GUI gives you.

design-maxdiff::design

[["design-maxdiff::design"]]
n_items = 10
items_per_task = 3
tasks_per_version = 8
versions = 4
run_dir = "./design"

2. Wrangle

Turn raw respondent data into a clean, model-ready long file. The same wrangling runs on the pilot and the full field.

choice-wrangling::pipeline-maxdiff

[["choice-wrangling::pipeline-maxdiff"]]
design_csv = "design.csv"
respondent_csv = "respondents.csv"
run_dir = "./workspace"
run_name = "pilot"

3. Estimate

Fit a hierarchical Bayesian model via Stan and finalize diagnostics + part-worths in one call. Swap the plugin to switch estimator — the rest of the pipeline stays the same.

model-hb-stan::fit-finalize

[["model-hb-stan::fit-finalize"]]
input = "./data/hb_data.csv"
run_dir = "./runs"
run_name = "pilot"
num_samples = 1000
num_chains = 4

4. Report

Score scenarios, rank subsets, compute reach — then commit the output next to the .oce that produced it. A year from now you’ll still know exactly how the number was made.

maxdiff-turf::score-scenarios

[["maxdiff-turf::score-scenarios"]]
utilities = "./utilities.csv"
scenarios = "./scenarios.csv"
out_dir = "./runs/score-scenarios"
reach_method = "first-choice"

Built for scripting in the AI era

`.oce` files — your pipeline, your file, your repo

One TOML file captures a full analysis. Commit it to git. Diff last quarter’s run vs today’s. Hand it to a coworker and they get the exact same result. A GUI’s “save state” button can’t touch this.

First-class help + IntelliSense, built for LLMs

Every endpoint has structured, machine-readable docs. IntelliSense autocompletes params as you type. Ask an AI to write your .oce — it has real schemas to anchor against. Scripting without the friction.

Timeline — every edit snapshotted, diffable

Regret a parameter tweak? Roll back. Want to see what changed between Monday and Friday? Scrub. Every .oce edit is captured automatically.

Job queue — parallel, sequential, or scheduled

Queue a batch of runs and let them work through in order, or fan several out at once on separate lanes. Need to kick off a 4-hour HB fit at midnight? Schedule it and close the lid.

Open-source, native-speed plugins

Compiled, open-source executables — estimations that crawl in R or Python run in seconds in native Rust. No vendor black-boxes. No cloud round-trips.

Plugin SDK in the language that fits

Rust, R, Python, Julia — plugins speak ocp-json/1. The fast solver in Rust, the Stan model in R, the data prep in Python — all interchangeable from one workbench.

Registry updates — always the latest method

New techniques ship as signed plugin updates on registry.openchoice.app. One click, cryptographically verified, current tools in your hands — no installer dance, no waiting for v2.5.

Your laptop, your data

Cross-platform desktop (Windows, Mac, Linux). No cloud lock-in. Install once, keep forever. Your numbers never leave the machine you chose.

Where to go next

Introduction What Open Choice is and how to read this guide.
Getting Started Install the app, install your first plugin, run your first task.
Plugins Browse the registry, understand trust, manage what you install.
Tasks The .oce file format reference.