# Joint Inference Experiment (oTree + Heroku)

## Local testing

```bash
pip install -r requirements.txt
otree devserver
# open http://localhost:8000
```

## Heroku deployment

```bash
# from this directory
git init
git add .
git commit -m "Initial commit"

# create heroku app
heroku create your-app-name
heroku addons:create heroku-postgresql:essential-0
heroku addons:create heroku-redis:mini

# config vars
heroku config:set OTREE_PRODUCTION=1
heroku config:set OTREE_ADMIN_PASSWORD=YOUR_PASSWORD_HERE
heroku config:set OTREE_SECRET_KEY=$(python -c "import secrets; print(secrets.token_urlsafe(50))")
heroku config:set OTREE_AUTH_LEVEL=DEMO

# deploy
git push heroku main

# initialize the database
heroku run otree resetdb --noinput
```

The app will be at https://your-app-name.herokuapp.com.

## Structure

- `joint_inference/__init__.py` — models, pages, Bayesian helpers, trial generation
- `joint_inference/templates/joint_inference/` — HTML for each page
- `settings.py` — oTree session config

## Trial structure

10 rounds total: 2 practice + 8 paid (4 with n_i=10, 4 with n_i=50).

Per-trial elicitations: Q1 (prior estimate), Q2 (paired-lottery price list, 11 rows from 0% to 100% in 10% steps), Q3 (posterior estimate). One of {Q1, Q2, Q3} randomly selected for payment within each trial. Max bonus per trial: $1.

## Data export

From the oTree admin, go to "Data" → CSV. Each row is a (player, round) pair with all elicitation responses, trial latents (true p, p_imp, k_i, k_a, k_b, a_is_partner), and resolved payoffs. The Bayesian benchmarks (q_A^*, posterior mean) can be computed offline from the latents using the helpers in `__init__.py`.