diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 237fc87..46aab0d 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -18,6 +18,19 @@ jobs:
           cache-dependency-glob: pyproject.toml
       - run: uvx ruff check .
 
+  docs:
+    # Catches broken links and API-docs errors before they reach the published site.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v10.2.0
+        with:
+          python-version: "3.12"
+          activate-environment: true
+          cache-dependency-glob: pyproject.toml
+      - run: uv pip install -e ".[docs]"
+      - run: mkdocs build --strict
+
   test:
     runs-on: ubuntu-latest
     strategy:
diff --git a/.github/workflows/docs.yml b/.github/workflows/docs.yml
new file mode 100644
index 0000000..22f7bd8
--- /dev/null
+++ b/.github/workflows/docs.yml
@@ -0,0 +1,50 @@
+name: Docs
+
+# Builds the documentation site and publishes it to GitHub Pages whenever main changes.
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - "docs/**"
+      - "mkdocs.yml"
+      - "src/**"
+      - "CHANGELOG.md"
+      - "pyproject.toml"
+      - ".github/workflows/docs.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v10.2.0
+        with:
+          python-version: "3.12"
+          activate-environment: true
+          cache-dependency-glob: pyproject.toml
+      - run: uv pip install -e ".[docs]"
+      - run: mkdocs build --strict
+      - uses: actions/upload-pages-artifact@v5
+        with:
+          path: site
+
+  deploy:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - id: deployment
+        uses: actions/deploy-pages@v5
diff --git a/.gitignore b/.gitignore
index 3882666..a09584d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,3 +12,4 @@ dist/
 .ipynb_checkpoints/
 benchmarks/results/*.log
 private/
+site/
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1133c7b..3d8cfed 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,7 +5,11 @@ All notable changes to graphdml are listed here. Versions follow
 
 ## [Unreleased]
 
+## [0.2.0] - 2026-10-09
+
 ### Added
+- Documentation site (https://samrdreyfuss.github.io/graphdml/): guides, methodology,
+  validation, example datasets and an API reference generated from the docstrings.
 - `graphdml.selftest()` and `make_signal_check()`: a built-in check, on synthetic data with
   planted effects and a null twin that has none, that the estimator finds real signals and
   does not invent false ones. Reports bias, coverage, power and false-alarm rates against
@@ -53,5 +57,6 @@ First release.
 - Neo4j integration: load graphs, write results back, GDS FastRP features, demo.
 - Validation suite and benchmarks.
 
-[Unreleased]: https://github.com/Samrdreyfuss/graphdml/compare/v0.1.0...HEAD
+[Unreleased]: https://github.com/Samrdreyfuss/graphdml/compare/v0.2.0...HEAD
+[0.2.0]: https://github.com/Samrdreyfuss/graphdml/compare/v0.1.0...v0.2.0
 [0.1.0]: https://github.com/Samrdreyfuss/graphdml/releases/tag/v0.1.0
diff --git a/CITATION.cff b/CITATION.cff
index 3583a1a..ba74368 100644
--- a/CITATION.cff
+++ b/CITATION.cff
@@ -6,7 +6,7 @@ authors:
   - family-names: Dreyfuss
     given-names: Sam
 license: MIT
-version: 0.1.0
+version: 0.2.0
 repository-code: "https://github.com/Samrdreyfuss/graphdml"
 references:
   - type: conference-paper
diff --git a/README.md b/README.md
index 45e82fa..cdbf8e2 100644
--- a/README.md
+++ b/README.md
@@ -7,6 +7,8 @@
 
 **Causal effects on networks, with honest confidence intervals.**
 
+**[Documentation](https://samrdreyfuss.github.io/graphdml/)** · [PyPI](https://pypi.org/project/graphdml/) · [Changelog](https://github.com/Samrdreyfuss/graphdml/blob/main/CHANGELOG.md)
+
 When a treatment reaches one person, it often reaches their friends too: a vaccine protects
 the neighbors, a coupon gets shared, a training spreads by word of mouth. graphdml estimates:
 
diff --git a/benchmarks/report.py b/benchmarks/report.py
index 0a37370..3529a85 100644
--- a/benchmarks/report.py
+++ b/benchmarks/report.py
@@ -112,7 +112,7 @@ def lastfm_section() -> str:
         "# Validation results\n\n"
         f"Generated by `benchmarks/report.py` on {dt.date.today()} for graphdml "
         f"{graphdml.__version__}. The protocol and pass criteria are in "
-        "[the validation protocol](../docs/validation.md); raw per-seed results are in "
+        "[the validation protocol](validation.md); raw per-seed results are in "
         "`benchmarks/results/*.csv`.\n"
     )
     DOC.write_text("\n\n".join([header, coverage_section(), design_section(),
diff --git a/docs/api.md b/docs/api.md
new file mode 100644
index 0000000..ff4a269
--- /dev/null
+++ b/docs/api.md
@@ -0,0 +1,65 @@
+# API reference
+
+## Estimator
+
+::: graphdml.GraphDML
+
+## Data
+
+::: graphdml.GraphData
+
+## Exposure maps
+
+::: graphdml.SumExposure
+
+::: graphdml.MeanExposure
+
+::: graphdml.WeightedExposure
+
+::: graphdml.TwoHopExposure
+
+::: graphdml.MatrixExposure
+
+## Features
+
+::: graphdml.NeighborhoodFeatures
+
+::: graphdml.OwnFeatures
+
+::: graphdml.PrecomputedFeatures
+
+## Diagnostics and checks
+
+::: graphdml.placebo_test
+
+::: graphdml.negative_control_test
+
+::: graphdml.two_hop_test
+
+::: graphdml.selftest
+
+## Datasets
+
+::: graphdml.datasets.GraphDataset
+
+::: graphdml.datasets.make_flu_town
+
+::: graphdml.datasets.make_referral_app
+
+::: graphdml.datasets.make_classroom_tutoring
+
+::: graphdml.datasets.make_homophily_trap
+
+::: graphdml.datasets.make_signal_check
+
+::: graphdml.datasets.make_lastfm_promo
+
+::: graphdml.datasets.load_insurance_experiment
+
+## Neo4j
+
+::: graphdml.io.neo4j.Neo4jGraphSource
+
+::: graphdml.io.neo4j.Neo4jResultWriter
+
+::: graphdml.io.neo4j.GDSEmbeddings
diff --git a/docs/changelog.md b/docs/changelog.md
new file mode 100644
index 0000000..786b75d
--- /dev/null
+++ b/docs/changelog.md
@@ -0,0 +1 @@
+--8<-- "CHANGELOG.md"
diff --git a/docs/datasets.md b/docs/datasets.md
new file mode 100644
index 0000000..92e5e29
--- /dev/null
+++ b/docs/datasets.md
@@ -0,0 +1,36 @@
+# Example datasets
+Every dataset is a `GraphDataset` with the data, the true effects (where known) and a story
+card: `print(ds.DESCR)`. Five generate their data offline; `make_lastfm_promo` downloads a
+public network on first use; `load_insurance_experiment` needs a free download.
+
+---
+
+--8<-- "src/graphdml/datasets/descr/flu_town.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/referral_app.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/classroom_tutoring.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/homophily_trap.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/signal_check.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/lastfm_promo.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/insurance_experiment.md"
+
+---
+
+--8<-- "src/graphdml/datasets/descr/toy_graph.md"
diff --git a/docs/index.md b/docs/index.md
new file mode 100644
index 0000000..e898621
--- /dev/null
+++ b/docs/index.md
@@ -0,0 +1,64 @@
+# graphdml
+
+**Causal effects on networks, with honest confidence intervals.**
+
+When a treatment reaches one person, it often reaches their friends too: a vaccine protects
+the neighbors, a coupon gets shared, a training spreads by word of mouth. graphdml estimates
+
+* the **ATE**: the average effect of treating everyone vs no one, including the ripple
+* the **direct effect** (ADE): what the treatment does for the person who receives it
+* the **peer effect** (APE): what it does for the people connected to them
+
+using double machine learning with any scikit-learn model, adjusting for confounding that
+runs through the network, with confidence intervals that account for neighbors being
+dependent. Graphs can come from pandas, networkx or Neo4j.
+
+## Install
+
+```bash
+pip install graphdml               # core
+pip install "graphdml[neo4j]"      # with Neo4j support
+```
+
+Requires Python 3.10 or newer.
+
+## Quickstart
+
+```python
+from graphdml import GraphDML
+from graphdml.datasets import make_flu_town
+
+ds = make_flu_town()      # 5,000 residents; true effects: -2.0 and -0.5 sick days
+model = GraphDML(exposure="sum", random_state=0).fit(ds.data)
+print(model.summary())
+```
+
+## Where to go next
+
+| | |
+|---|---|
+| [Example datasets](datasets.md) | Eight datasets with known truth, and the story behind each |
+| [Methodology](methodology.md) | Exactly what `fit` computes, and why |
+| [Design choices](design.md) | Each default, with the evidence behind it |
+| [Validation](validation.md) | The protocol, and the [results](validation-results.md) |
+| [Neo4j integration](neo4j.md) | Load graphs, write results back, GDS embeddings |
+| [API reference](api.md) | Every public class and function |
+
+## Does it find real signals?
+
+```python
+import graphdml
+print(graphdml.selftest())
+```
+
+Simulates a world with planted effects and a null twin with none, and checks that graphdml
+finds the first and stays quiet on the second, while a method that ignores the network
+does not. See the [README](https://github.com/Samrdreyfuss/graphdml#does-it-find-real-signals-run-the-self-test)
+for sample output.
+
+## Citing
+
+The estimator builds on Khatami, Parikh, Chen, Roy and Salimi,
+[*Graph Machine Learning based Doubly Robust Estimator for Network Causal Effects*](https://proceedings.mlr.press/v258/khatami25a.html)
+(AISTATS 2025). Please cite their paper along with graphdml; see
+[CITATION.cff](https://github.com/Samrdreyfuss/graphdml/blob/main/CITATION.cff).
diff --git a/docs/javascripts/mathjax.js b/docs/javascripts/mathjax.js
new file mode 100644
index 0000000..e9656e9
--- /dev/null
+++ b/docs/javascripts/mathjax.js
@@ -0,0 +1,9 @@
+window.MathJax = {
+  tex: {
+    inlineMath: [["\\(", "\\)"]],
+    displayMath: [["\\[", "\\]"]],
+    processEscapes: true,
+    processEnvironments: true,
+  },
+  options: { ignoreHtmlClass: ".*|", processHtmlClass: "arithmatex" },
+};
diff --git a/docs/validation-results.md b/docs/validation-results.md
index 80ed112..4d07866 100644
--- a/docs/validation-results.md
+++ b/docs/validation-results.md
@@ -1,6 +1,6 @@
 # Validation results
 
-Generated by `benchmarks/report.py` on 2026-10-06 for graphdml 0.1.0. The protocol and pass criteria are in [the validation protocol](../docs/validation.md); raw per-seed results are in `benchmarks/results/*.csv`.
+Generated by `benchmarks/report.py` on 2026-10-09 for graphdml 0.1.0. The protocol and pass criteria are in [the validation protocol](validation.md); raw per-seed results are in `benchmarks/results/*.csv`.
 
 
 ## Coverage
diff --git a/mkdocs.yml b/mkdocs.yml
new file mode 100644
index 0000000..fc78036
--- /dev/null
+++ b/mkdocs.yml
@@ -0,0 +1,63 @@
+site_name: graphdml
+site_description: Causal effects on networks with double machine learning
+site_url: https://samrdreyfuss.github.io/graphdml/
+repo_url: https://github.com/Samrdreyfuss/graphdml
+repo_name: Samrdreyfuss/graphdml
+edit_uri: edit/main/docs/
+
+theme:
+  name: material
+  features:
+    - navigation.sections
+    - navigation.top
+    - content.code.copy
+    - search.suggest
+  palette:
+    - media: "(prefers-color-scheme: light)"
+      scheme: default
+      toggle: {icon: material/weather-night, name: Switch to dark mode}
+    - media: "(prefers-color-scheme: dark)"
+      scheme: slate
+      toggle: {icon: material/weather-sunny, name: Switch to light mode}
+
+nav:
+  - Home: index.md
+  - Guides:
+      - Example datasets: datasets.md
+      - Neo4j integration: neo4j.md
+  - Method:
+      - Methodology: methodology.md
+      - Design choices: design.md
+  - Validation:
+      - Protocol: validation.md
+      - Results: validation-results.md
+  - API reference: api.md
+  - Changelog: changelog.md
+
+markdown_extensions:
+  - admonition
+  - tables
+  - toc:
+      permalink: true
+  - pymdownx.arithmatex:
+      generic: true
+  - pymdownx.highlight
+  - pymdownx.superfences
+  - pymdownx.snippets:
+      check_paths: true
+
+extra_javascript:
+  - javascripts/mathjax.js
+  - https://unpkg.com/mathjax@3/es5/tex-mml-chtml.js
+
+plugins:
+  - search
+  - mkdocstrings:
+      handlers:
+        python:
+          options:
+            docstring_style: numpy
+            show_source: false
+            show_root_heading: true
+            heading_level: 3
+            members_order: source
diff --git a/pyproject.toml b/pyproject.toml
index 7404f5a..a0bc7e1 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "graphdml"
-version = "0.1.0"
+version = "0.2.0"
 description = "Causal effects on networks with double machine learning: ATE, direct and peer effects with confidence intervals that account for network dependence."
 readme = "README.md"
 license = "MIT"
@@ -43,6 +43,7 @@ dependencies = [
 
 [project.optional-dependencies]
 neo4j = ["neo4j>=5.14"]
+docs = ["mkdocs-material>=9.5", "mkdocstrings[python]>=0.25"]
 validation = ["statsmodels>=0.14", "doubleml>=0.9", "joblib>=1.3", "tabulate>=0.9"]
 dev = [
     "graphdml[neo4j]",
@@ -54,7 +55,7 @@ dev = [
 
 [project.urls]
 Homepage = "https://github.com/Samrdreyfuss/graphdml"
-Documentation = "https://github.com/Samrdreyfuss/graphdml/tree/main/docs"
+Documentation = "https://samrdreyfuss.github.io/graphdml/"
 Source = "https://github.com/Samrdreyfuss/graphdml"
 Issues = "https://github.com/Samrdreyfuss/graphdml/issues"
 Changelog = "https://github.com/Samrdreyfuss/graphdml/blob/main/CHANGELOG.md"
diff --git a/src/graphdml/__init__.py b/src/graphdml/__init__.py
index 6cc061e..bdaf948 100644
--- a/src/graphdml/__init__.py
+++ b/src/graphdml/__init__.py
@@ -20,7 +20,7 @@
 from graphdml.features import NeighborhoodFeatures, OwnFeatures, PrecomputedFeatures
 from graphdml.selftest import selftest
 
-__version__ = "0.1.0"
+__version__ = "0.2.0"
 
 __all__ = [
     "ExposureMap",
diff --git a/src/graphdml/selftest.py b/src/graphdml/selftest.py
index bfa70e7..0bb19a3 100644
--- a/src/graphdml/selftest.py
+++ b/src/graphdml/selftest.py
@@ -87,17 +87,36 @@ def selftest(
 
     Parameters
     ----------
-    n_reps : repetitions per world; each fits one model on a fresh ``n``-node graph.
-    estimator : an unfitted :class:`GraphDML` to test (default: ``GraphDML(exposure="mean")``
-        with the library's default learners). Pass your own to test your settings.
-
-    Checks (set from statistics, not tuned to results; the bands use 3 Monte Carlo
-    standard errors, so they loosen when ``n_reps`` is small):
+    n_reps : int, default=30
+        Repetitions per world; each fits one model on a fresh ``n``-node graph.
+    n : int, default=6000
+        Nodes per simulated graph.
+    direct_effect, peer_effect : float
+        The effects planted in the signal world (the null world has none).
+    estimator : GraphDML, optional
+        An unfitted estimator to test (default: ``GraphDML(exposure="mean")`` with the
+        library's default learners). Pass your own to test your settings.
+    alpha : float, default=0.05
+        Significance level used for detection and false alarms.
+    n_jobs : int, default=-1
+        Parallel workers (joblib).
+    random_state : int, default=0
+        Seeds the repetitions.
+
+    Returns
+    -------
+    SelfTestResult
+        ``passed``, a results ``table`` and the individual ``checks``.
+
+    Notes
+    -----
+    The criteria are set from statistics, not tuned to results. The bands use 3 Monte
+    Carlo standard errors, so they loosen when ``n_reps`` is small.
 
     * **coverage**: the share of 95% intervals containing the truth is at least
-      ``0.95 - 3 * SE`` in the planted world, for every effect;
+      ``0.95 - 3 * SE`` in the planted world, for every effect.
     * **power**: the planted direct and peer effects are detected in at least 80% of runs,
-      with the correct sign;
+      with the correct sign.
     * **false alarms**: in the null world, the share of runs that claim an effect is at
       most ``alpha + 3 * SE``, for every effect.
     """