Move paper to paper/ subdirectory and add tutorial figures

- Move paper.md and paper.bib into paper/ to avoid top-level clutter
- Add hillstorm_dte.png and oregon_ldte_costs_comparison.png from tutorials
- Add figure captions showing simple vs ML-adjusted comparison for both
  the Hillstrom email campaign and Oregon Health Insurance Experiment
- Cross-reference figures in Research Impact Statement
- Add dataset citations (Hillstrom 2008, Finkelstein et al. 2012)
- Add GitHub Action (draft-pdf.yml) to auto-build JOSS PDF on push
  using paper-path: paper/paper.md

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: TomeHirata

.github/workflows/draft-pdf.yml

@@ -0,0 +1,19 @@
+on: [push]
+
+jobs:
+  paper:
+    runs-on: ubuntu-latest
+    name: Paper Draft
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Build draft PDF
+        uses: openjournals/openjournals-draft-action@master
+        with:
+          journal: joss
+          paper-path: paper/paper.md
+      - name: Upload
+        uses: actions/upload-artifact@v4
+        with:
+          name: paper
+          path: paper/paper.pdf

paper/hillstorm_dte.png

@@ -116,6 +116,24 @@ @ARTICLE{2020SciPy-NMeth
   doi     = {10.1038/s41592-019-0686-2}
 }
 
+@misc{hillstrom2008,
+  title={The MineThatData E-Mail Analytics And Data Mining Challenge},
+  author={Hillstrom, Kevin},
+  year={2008},
+  url={https://blog.minethatdata.com/2008/03/minethatdata-e-mail-analytics-and-data.html}
+}
+
+@article{finkelstein2012,
+  title={The Oregon Health Insurance Experiment: Evidence from the First Year},
+  author={Finkelstein, Amy and Taubman, Sarah and Wright, Bill and Bernstein, Mira and Gruber, Jonathan and Newhouse, Joseph P. and Allen, Heidi and Baicker, Katherine and {Oregon Health Study Group}},
+  journal={The Quarterly Journal of Economics},
+  volume={127},
+  number={3},
+  pages={1057--1106},
+  year={2012},
+  doi={10.1093/qje/qjs020}
+}
+
 @article{kobrosly2020causalcurve,
   title={causal-curve: A Python Causal Inference Package to Estimate Causal Dose-Response Curves},
   author={Kobrosly, Roni W.},

paper/oregon_ldte_costs_comparison.png

@@ -60,9 +60,13 @@ In the R ecosystem, packages like `qte` provide quantile treatment effect estima
 
 All estimators implement a consistent API with three primary methods: `predict_dte()` for distributional treatment effects, `predict_pte()` for probability treatment effects over intervals, and `predict_qte()` for quantile treatment effects. The adjusted estimators use K-fold cross-fitting to prevent overfitting and support both single-task and multi-task learning modes [@hirata2025efficientscalableestimationdistributional] for computational efficiency. Bootstrap methods provide confidence intervals with multiple variance estimation approaches.
 
+![Distributional treatment effects for the Hillstrom email marketing dataset [@hillstrom2008], comparing Women's vs Men's email campaigns. The simple estimator (left, purple) and ML-adjusted estimator (right, green) show that adjustment substantially tightens confidence bands, demonstrating the variance reduction benefit of regression adjustment.\label{fig:hillstrom}](hillstrom_dte.png)
+
+![Local distributional treatment effects for emergency department costs in the Oregon Health Insurance Experiment [@finkelstein2012], estimated using `SimpleLocalDistributionEstimator` (left) and `AdjustedLocalDistributionEstimator` (right). Health insurance coverage shifts the distribution of ED costs, with ML adjustment again yielding narrower confidence intervals.\label{fig:oregon}](oregon_ldte_costs_comparison.png)
+
 # Research Impact Statement
 
-The methods implemented in `dte_adj` have been published at top machine learning venues: ICML 2024 [@byambadalai2024estimatingdistributionaltreatmenteffects] and ICML 2025 [@byambadalai2025efficientestimationdistributionaltreatment]. The package has been used internally at CyberAgent, Inc. for analyzing A/B tests where distributional impacts are critical, such as evaluating interventions on user engagement metrics where tail behavior matters more than averages. The documentation includes tutorials demonstrating applications to the Hillstrom email marketing dataset and the Oregon Health Insurance Experiment, facilitating adoption by researchers in economics, marketing, and healthcare.
+The methods implemented in `dte_adj` have been published at top machine learning venues: ICML 2024 [@byambadalai2024estimatingdistributionaltreatmenteffects] and ICML 2025 [@byambadalai2025efficientestimationdistributionaltreatment]. The package has been used internally at CyberAgent, Inc. for analyzing A/B tests where distributional impacts are critical, such as evaluating interventions on user engagement metrics where tail behavior matters more than averages. The documentation includes tutorials demonstrating applications to the Hillstrom email marketing dataset (\autoref{fig:hillstrom}) and the Oregon Health Insurance Experiment (\autoref{fig:oregon}), facilitating adoption by researchers in economics, marketing, and healthcare.
 
 # AI Usage Disclosure