emu-gmm 0.3.1__tar.gz

emu_gmm-0.3.1/LICENSE.org

@@ -0,0 +1,19 @@
+#+TITLE: License
+#+OPTIONS: toc:nil num:nil
+
+* Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)
+
+This work is licensed under the
+[[http://creativecommons.org/licenses/by-nc-sa/4.0/][Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License]].
+
+To view a copy of this license, visit:
+http://creativecommons.org/licenses/by-nc-sa/4.0/
+or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.
+
+** Summary (non-legal)
+- *Attribution*: credit "Ethan Ligon et al., emu-gmm project"
+- *Non-Commercial*: may not be used for commercial purposes
+- *ShareAlike*: derivative works must carry the same license
+- *No additional restrictions*: you may not impose legal or technological barriers that prevent others from exercising the license terms.
+
+Full legal code available at the above URL.

emu_gmm-0.3.1/PKG-INFO

@@ -0,0 +1,191 @@
+Metadata-Version: 2.4
+Name: emu-gmm
+Version: 0.3.1
+Summary: Measure-theoretic Generalized Method of Moments estimation; estimation via E_mu.
+License: CC-BY-NC-SA-4.0
+License-File: LICENSE.org
+Author: Ethan Ligon
+Author-email: ligon@berkeley.edu
+Requires-Python: >=3.11,<4.0
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: License :: Other/Proprietary License
+Classifier: Operating System :: OS Independent
+Classifier: Intended Audience :: Science/Research
+Classifier: Topic :: Scientific/Engineering
+Requires-Dist: haliax (>=1.4.dev452,<1.5)
+Requires-Dist: jax (>=0.4.25)
+Requires-Dist: jax_dataclasses (>=1.6.1)
+Requires-Dist: jaxtyping (>=0.2.30)
+Requires-Dist: numpy (>=1.26)
+Requires-Dist: optimistix (>=0.0.7)
+Requires-Dist: pandas (>=2.0)
+Requires-Dist: scipy (>=1.11)
+Description-Content-Type: text/plain
+
+#+TITLE: emu-gmm
+#+SUBTITLE: Measure-theoretic GMM; estimation via $\mathbb{E}_\mu$
+#+AUTHOR: Ethan Ligon
+#+OPTIONS: toc:nil num:nil
+
+=emu-gmm= is a JAX-native framework for Generalized Method of Moments estimation. The framework is named for the operator at its centre: =emu= reads as $\mathbb{E}_\mu$, the expectation under a measure $\mu$.  One operator interface --- implemented against empirical, analytical, or synthetic measures --- drives sample estimation, identification analysis, and simulation-based inference through a single computational pipeline.
+
+The most important architectural commitment is that variance construction is orthogonal to integration: $\mathbb{E}_\mu[\psi]$ is a property of the measure; $V_\mu(\theta)$ is a property of a separate =CovarianceStrategy= that can be swapped (iid vs cluster-robust vs replicate-weight) without changing what is being integrated.
+
+* Status
+
+v2 (v0.2.0): the v1 pipeline plus Riemannian-manifold parameter geometry and the stratified / design-aware covariance ladder. The implemented menu --- everything importable and composable today --- is tabulated in =docs/design.org= under "Contract surface (implemented)", the single source of truth for status claims; the green gate is =make check= (ruff + black + mypy + the full pytest suite). The three measure paths (synthetic, analytical, empirical) all run end-to-end against the bundled multi-asset Euler example, and the empirical path additionally has a real-data acceptance test (=tests/test_estimator_realdata.py=). The architecture and theoretical scope are pinned in =docs/=.
+
+* Quickstart
+
+Bootstrap and try the worked example:
+
+#+begin_src shell
+make setup                                 # poetry install + .venv/
+poetry run python examples/run_euler.py    # multi-asset Hansen-Singleton demo
+#+end_src
+
+The minimal estimation surface in code (synthetic-data variant):
+
+#+begin_src python
+import jax
+from emu_gmm import estimate, SyntheticMeasure, SyntheticCovariance
+from emu_gmm.examples.euler import (
+    EulerParams, euler_residual, euler_sampler_factory,
+)
+
+measure = SyntheticMeasure(
+    key=jax.random.PRNGKey(0),
+    n_sim=5000,
+    sampler=euler_sampler_factory(5000),
+)
+
+result = estimate(
+    model=euler_residual,
+    measure=measure,
+    covariance=SyntheticCovariance(),
+    theta_init=EulerParams(beta=0.9, gamma=1.0),
+)
+
+print(f"beta = {result.theta_hat.beta:.4f}  (truth 0.96)")
+print(f"gamma = {result.theta_hat.gamma:.4f}  (truth 2.00)")
+print(f"J-stat = {result.J_stat:.3f}  dof={result.J_dof}  p={result.J_pvalue:.3f}")
+print(result.to_pandas()["Sigma_theta"])
+#+end_src
+
+The same =euler_residual= drives the analytical and empirical variants of the demo; only the =Measure= / =CovarianceStrategy= change. See =docs/api-sketch.org= Section 5 for the three side-by-side, and =examples/run_euler.py= for the runnable script.
+
+* Documents
+
+- =docs/howto.org= --- *start here.* Integration HOWTO: the architecture
+  at the level you need to wire an application to =estimate()= --- what
+  code you write (the moment function, the parameter container, the
+  measure/covariance choice) and what the framework hands back.
+- =docs/design.org= --- architecture specification (four review rounds; stable). Its "Contract surface (implemented)" section is the single source of truth for the implemented menu.
+- =docs/api-sketch.org= --- the v1 API surface, retained as the v1 design record; superseded on status by =design.org='s contract-surface section.
+- =docs/implementation-plan.org= --- phased task list (Phases 1-7 complete, Phase 8 polish underway; the v2 roadmap is Section 13).
+- =docs/mcar-asymptotics.org= --- companion theoretical note; consistency, asymptotic normality, PD properties of the pairwise-overlap estimator under MCAR.
+- =docs/refs.bib= --- project-local bibliography (entries not in =~/bibtex/main.bib=).
+
+* Setup
+
+Requires Python >= 3.11 and [[https://python-poetry.org/][Poetry]].
+
+#+begin_src shell
+make setup
+#+end_src
+
+This runs =poetry install= and creates =.venv/= in the project root. Activate with =direnv allow= or =poetry shell=.
+
+* Development
+
+#+begin_src shell
+make check         # ruff + black + mypy + full pytest
+make quick-check   # same, skipping slow tests
+make test          # pytest only
+#+end_src
+
+To install pre-commit hooks (ruff + black on every commit):
+
+#+begin_src shell
+poetry run pre-commit install
+#+end_src
+
+* Layout
+
+#+begin_example
+docs/                      design specs and theoretical notes
+src/emu_gmm/
+    __init__.py            public API re-exports
+    types.py               protocols + EstimationResult / Diagnostics
+    estimator.py           estimate() / build_estimator() entry points
+    measures/              SyntheticMeasure, AnalyticalMeasure, EmpiricalMeasure
+    covariance/            the CovarianceStrategy ladder (iid, clustered,
+                           stratified / design-aware, sum, analytical, synthetic)
+    weighting.py           Identity, Fixed, IteratedWeighting, ContinuouslyUpdated
+    regularization.py      DiagonalTikhonov
+    penalty.py             TikhonovPenalty
+    optimizer.py           optimistix_lm, scipy_lm, linear_solver
+    manifolds/             parameter geometry (Euclidean, Positive, PSDFixedRank,
+                           Product, ManifoldLeaf) + riemannian_lm
+    parameter_space.py     ParameterSpace / on field-to-manifold declarations
+    inference/             j_test, k_statistic, k_confidence_set, bootstraps
+    numerics/              ridge_inverse
+    studies/               Monte Carlo driver (subpackage-only; not re-exported)
+    diagnostics.py         build_diagnostics, log_to_stdout
+    examples/              shared example models (euler.py)
+    _internal/             axes, params, cholesky, labels (private)
+tests/                     test suite mirroring src/emu_gmm/
+examples/                  runnable demo scripts
+Makefile                   build automation
+pyproject.toml             Poetry configuration
+poetry.lock                pinned dependencies (committed)
+#+end_example
+
+* Public API
+
+Everything user-facing is re-exported at the package top level
+(=from emu_gmm import ...=), with one deliberate exception: the Monte
+Carlo =studies= subpackage is subpackage-only (import from
+=emu_gmm.studies= directly). The implemented menu --- entry points,
+measures, covariance strategies, weighting, regularization, penalty,
+optimisers, manifold geometry, inference helpers, and result types ---
+is tabulated, one line of semantics and a module pointer per row, in
+=docs/design.org= under "Contract surface (implemented)". That table is
+the single source of truth this README defers to;
+=sorted(emu_gmm.__all__)= is the export list it tracks.
+
+* Using emu-gmm correctly (and reporting problems)
+
+=emu-gmm= is intended as a /spare set of correct interfaces/. Your only
+modelling input is a per-observation residual =psi(x_i, theta) -> R^M=;
+the package owns the moment expectation, the design-aware covariance
+=V_X=, the criterion, the \(J\)-statistic, standard errors, and
+\(p\)-values. Two rules keep estimation correct:
+
+1. *Read results off the package, never recompute them by hand.* Take the
+   criterion, =J_stat=, =standard_errors=, and \(p\)-values from
+   =EstimationResult= / =k_statistic= / the inference helpers. A
+   hand-rolled criterion is how subtle scaling bugs get reintroduced ---
+   notably the textbook habit of scaling moments by \(\sqrt{N}\) and
+   weighting by \(V_X\), which is correct only when every moment has the
+   /same/ number of observations. =emu-gmm= keeps moments as
+   per-coordinate means and folds all the per-moment \(N_j\) bookkeeping
+   into \(V_X\) (the \(1/(N_j N_k)\) normalisation); see =docs/design.org=
+   "Scaling convention" and =docs/mcar-asymptotics.org=.
+
+2. *Express per-moment observability only through the =mask=.* Different
+   moments may be observed for different numbers of units; the \((N, M)\)
+   =mask= on =EmpiricalMeasure= is the single place that lives. Do not
+   work around it by pre-aggregating or by assuming a common \(N\).
+
+If an interface is missing a knob, or you believe it gives a wrong
+answer: *file an issue against =emu-gmm=, do not reimplement the
+statistic in your own project.* A local reimplementation forfeits the
+"single correct implementation" guarantee for everyone downstream. The
+fix belongs in the shared package.
+
+* License
+
+[[file:LICENSE.org][CC-BY-NC-SA-4.0]].
+

emu_gmm-0.3.1/README.org

@@ -0,0 +1,165 @@
+#+TITLE: emu-gmm
+#+SUBTITLE: Measure-theoretic GMM; estimation via $\mathbb{E}_\mu$
+#+AUTHOR: Ethan Ligon
+#+OPTIONS: toc:nil num:nil
+
+=emu-gmm= is a JAX-native framework for Generalized Method of Moments estimation. The framework is named for the operator at its centre: =emu= reads as $\mathbb{E}_\mu$, the expectation under a measure $\mu$.  One operator interface --- implemented against empirical, analytical, or synthetic measures --- drives sample estimation, identification analysis, and simulation-based inference through a single computational pipeline.
+
+The most important architectural commitment is that variance construction is orthogonal to integration: $\mathbb{E}_\mu[\psi]$ is a property of the measure; $V_\mu(\theta)$ is a property of a separate =CovarianceStrategy= that can be swapped (iid vs cluster-robust vs replicate-weight) without changing what is being integrated.
+
+* Status
+
+v2 (v0.2.0): the v1 pipeline plus Riemannian-manifold parameter geometry and the stratified / design-aware covariance ladder. The implemented menu --- everything importable and composable today --- is tabulated in =docs/design.org= under "Contract surface (implemented)", the single source of truth for status claims; the green gate is =make check= (ruff + black + mypy + the full pytest suite). The three measure paths (synthetic, analytical, empirical) all run end-to-end against the bundled multi-asset Euler example, and the empirical path additionally has a real-data acceptance test (=tests/test_estimator_realdata.py=). The architecture and theoretical scope are pinned in =docs/=.
+
+* Quickstart
+
+Bootstrap and try the worked example:
+
+#+begin_src shell
+make setup                                 # poetry install + .venv/
+poetry run python examples/run_euler.py    # multi-asset Hansen-Singleton demo
+#+end_src
+
+The minimal estimation surface in code (synthetic-data variant):
+
+#+begin_src python
+import jax
+from emu_gmm import estimate, SyntheticMeasure, SyntheticCovariance
+from emu_gmm.examples.euler import (
+    EulerParams, euler_residual, euler_sampler_factory,
+)
+
+measure = SyntheticMeasure(
+    key=jax.random.PRNGKey(0),
+    n_sim=5000,
+    sampler=euler_sampler_factory(5000),
+)
+
+result = estimate(
+    model=euler_residual,
+    measure=measure,
+    covariance=SyntheticCovariance(),
+    theta_init=EulerParams(beta=0.9, gamma=1.0),
+)
+
+print(f"beta = {result.theta_hat.beta:.4f}  (truth 0.96)")
+print(f"gamma = {result.theta_hat.gamma:.4f}  (truth 2.00)")
+print(f"J-stat = {result.J_stat:.3f}  dof={result.J_dof}  p={result.J_pvalue:.3f}")
+print(result.to_pandas()["Sigma_theta"])
+#+end_src
+
+The same =euler_residual= drives the analytical and empirical variants of the demo; only the =Measure= / =CovarianceStrategy= change. See =docs/api-sketch.org= Section 5 for the three side-by-side, and =examples/run_euler.py= for the runnable script.
+
+* Documents
+
+- =docs/howto.org= --- *start here.* Integration HOWTO: the architecture
+  at the level you need to wire an application to =estimate()= --- what
+  code you write (the moment function, the parameter container, the
+  measure/covariance choice) and what the framework hands back.
+- =docs/design.org= --- architecture specification (four review rounds; stable). Its "Contract surface (implemented)" section is the single source of truth for the implemented menu.
+- =docs/api-sketch.org= --- the v1 API surface, retained as the v1 design record; superseded on status by =design.org='s contract-surface section.
+- =docs/implementation-plan.org= --- phased task list (Phases 1-7 complete, Phase 8 polish underway; the v2 roadmap is Section 13).
+- =docs/mcar-asymptotics.org= --- companion theoretical note; consistency, asymptotic normality, PD properties of the pairwise-overlap estimator under MCAR.
+- =docs/refs.bib= --- project-local bibliography (entries not in =~/bibtex/main.bib=).
+
+* Setup
+
+Requires Python >= 3.11 and [[https://python-poetry.org/][Poetry]].
+
+#+begin_src shell
+make setup
+#+end_src
+
+This runs =poetry install= and creates =.venv/= in the project root. Activate with =direnv allow= or =poetry shell=.
+
+* Development
+
+#+begin_src shell
+make check         # ruff + black + mypy + full pytest
+make quick-check   # same, skipping slow tests
+make test          # pytest only
+#+end_src
+
+To install pre-commit hooks (ruff + black on every commit):
+
+#+begin_src shell
+poetry run pre-commit install
+#+end_src
+
+* Layout
+
+#+begin_example
+docs/                      design specs and theoretical notes
+src/emu_gmm/
+    __init__.py            public API re-exports
+    types.py               protocols + EstimationResult / Diagnostics
+    estimator.py           estimate() / build_estimator() entry points
+    measures/              SyntheticMeasure, AnalyticalMeasure, EmpiricalMeasure
+    covariance/            the CovarianceStrategy ladder (iid, clustered,
+                           stratified / design-aware, sum, analytical, synthetic)
+    weighting.py           Identity, Fixed, IteratedWeighting, ContinuouslyUpdated
+    regularization.py      DiagonalTikhonov
+    penalty.py             TikhonovPenalty
+    optimizer.py           optimistix_lm, scipy_lm, linear_solver
+    manifolds/             parameter geometry (Euclidean, Positive, PSDFixedRank,
+                           Product, ManifoldLeaf) + riemannian_lm
+    parameter_space.py     ParameterSpace / on field-to-manifold declarations
+    inference/             j_test, k_statistic, k_confidence_set, bootstraps
+    numerics/              ridge_inverse
+    studies/               Monte Carlo driver (subpackage-only; not re-exported)
+    diagnostics.py         build_diagnostics, log_to_stdout
+    examples/              shared example models (euler.py)
+    _internal/             axes, params, cholesky, labels (private)
+tests/                     test suite mirroring src/emu_gmm/
+examples/                  runnable demo scripts
+Makefile                   build automation
+pyproject.toml             Poetry configuration
+poetry.lock                pinned dependencies (committed)
+#+end_example
+
+* Public API
+
+Everything user-facing is re-exported at the package top level
+(=from emu_gmm import ...=), with one deliberate exception: the Monte
+Carlo =studies= subpackage is subpackage-only (import from
+=emu_gmm.studies= directly). The implemented menu --- entry points,
+measures, covariance strategies, weighting, regularization, penalty,
+optimisers, manifold geometry, inference helpers, and result types ---
+is tabulated, one line of semantics and a module pointer per row, in
+=docs/design.org= under "Contract surface (implemented)". That table is
+the single source of truth this README defers to;
+=sorted(emu_gmm.__all__)= is the export list it tracks.
+
+* Using emu-gmm correctly (and reporting problems)
+
+=emu-gmm= is intended as a /spare set of correct interfaces/. Your only
+modelling input is a per-observation residual =psi(x_i, theta) -> R^M=;
+the package owns the moment expectation, the design-aware covariance
+=V_X=, the criterion, the \(J\)-statistic, standard errors, and
+\(p\)-values. Two rules keep estimation correct:
+
+1. *Read results off the package, never recompute them by hand.* Take the
+   criterion, =J_stat=, =standard_errors=, and \(p\)-values from
+   =EstimationResult= / =k_statistic= / the inference helpers. A
+   hand-rolled criterion is how subtle scaling bugs get reintroduced ---
+   notably the textbook habit of scaling moments by \(\sqrt{N}\) and
+   weighting by \(V_X\), which is correct only when every moment has the
+   /same/ number of observations. =emu-gmm= keeps moments as
+   per-coordinate means and folds all the per-moment \(N_j\) bookkeeping
+   into \(V_X\) (the \(1/(N_j N_k)\) normalisation); see =docs/design.org=
+   "Scaling convention" and =docs/mcar-asymptotics.org=.
+
+2. *Express per-moment observability only through the =mask=.* Different
+   moments may be observed for different numbers of units; the \((N, M)\)
+   =mask= on =EmpiricalMeasure= is the single place that lives. Do not
+   work around it by pre-aggregating or by assuming a common \(N\).
+
+If an interface is missing a knob, or you believe it gives a wrong
+answer: *file an issue against =emu-gmm=, do not reimplement the
+statistic in your own project.* A local reimplementation forfeits the
+"single correct implementation" guarantee for everyone downstream. The
+fix belongs in the shared package.
+
+* License
+
+[[file:LICENSE.org][CC-BY-NC-SA-4.0]].

emu_gmm-0.3.1/pyproject.toml

@@ -0,0 +1,96 @@
+[project]
+name = "emu-gmm"
+version = "0.3.1"
+description = "Measure-theoretic Generalized Method of Moments estimation; estimation via E_mu."
+readme = "README.org"
+requires-python = ">=3.11,<4.0"
+license = { text = "CC-BY-NC-SA-4.0" }
+authors = [{ name = "Ethan Ligon", email = "ligon@berkeley.edu" }]
+classifiers = [
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.11",
+  "License :: Other/Proprietary License",
+  "Operating System :: OS Independent",
+  "Intended Audience :: Science/Research",
+  "Topic :: Scientific/Engineering"
+]
+dependencies = [
+  "jax>=0.4.25",
+  "jaxtyping>=0.2.30",
+  "haliax>=1.4.dev452,<1.5",
+  "jax_dataclasses>=1.6.1",
+  "optimistix>=0.0.7",
+  "numpy>=1.26",
+  "scipy>=1.11",
+  "pandas>=2.0"
+]
+
+[tool.poetry]
+package-mode = true
+packages = [{ include = "emu_gmm", from = "src" }]
+
+[tool.poetry.dependencies]
+python = ">=3.11,<4.0"
+jax = ">=0.4.25"
+jaxtyping = ">=0.2.30"
+# PyPI's stable haliax 1.3 imports jax._src.tree_util.BuiltInKeyEntry, which
+# current JAX no longer exposes; upstream main is compatible but PyPI forbids
+# direct (git URL) dependencies in published metadata (the v0.3.0 publish
+# rejection). haliax auto-publishes dev wheels from main: 1.4.dev452 is built
+# from 280f49fb (2025-11-08) -- the EXACT revision the previous git pin had
+# locked and the whole v2 validation ran against. The pre-release specifier
+# is deliberate and PEP 440-explicit (it enables pre-release resolution for
+# this requirement); the <1.5 ceiling guards API drift, and a future stable
+# 1.4 satisfies the range. Suite re-validated against the PyPI wheel before
+# this change landed.
+haliax = { version = ">=1.4.dev452,<1.5", allow-prereleases = true }
+jax_dataclasses = ">=1.6.1"
+optimistix = ">=0.0.7"
+numpy = ">=1.26"
+scipy = ">=1.11"
+pandas = ">=2.0"
+
+[tool.poetry.group.dev.dependencies]
+pytest = "^8.2.0"
+pytest-xdist = "^3.0"
+ipython = "^8.26.0"
+ruff = "^0.5.4"
+black = "^24.4.2"
+mypy = "^1.10.0"
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+markers = [
+  "slow: tests that perform heavier computations"
+]
+
+[tool.black]
+target-version = ["py311"]
+line-length = 88
+
+[tool.ruff]
+target-version = "py311"
+line-length = 88
+extend-exclude = [".claude", ".coder-session", "examples"]
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "B", "UP"]
+# E501 (line length): handled by black.
+# F722 (syntax error in forward annotation), F821 (undefined name in
+# forward annotation), UP037 (remove quotes from type annotation):
+# all false positives on jaxtyping shape strings like
+# Float[Array, "M K"]. Suppressed project-wide.
+ignore = ["E501", "F722", "F821", "UP037"]
+
+[tool.ruff.lint.isort]
+known-third-party = ["jax", "jaxtyping", "haliax", "jax_dataclasses", "optimistix", "numpy", "scipy"]
+
+[tool.mypy]
+python_version = "3.11"
+warn_unused_ignores = true
+ignore_missing_imports = true
+strict_optional = true
+
+[build-system]
+requires = ["poetry-core>=1.7.0"]
+build-backend = "poetry.core.masonry.api"

emu_gmm-0.3.1/src/emu_gmm/__init__.py

@@ -0,0 +1,213 @@
+"""emu-gmm: Measure-theoretic GMM. Estimation via E_mu.
+
+The public API is re-exported here for convenience::
+
+    from emu_gmm import (
+        estimate,
+        SyntheticMeasure, SyntheticCovariance,
+        AnalyticalMeasure, AnalyticalCovariance,
+        EmpiricalMeasure, IIDCovariance, ClusteredCovariance,
+        Identity, Fixed, ContinuouslyUpdated, IteratedWeighting,
+        DiagonalTikhonov,
+        TikhonovPenalty,
+        optimistix_lm, scipy_lm,
+        EstimationResult, Diagnostics, OptimizerInfo,
+    )
+
+See docs/design.org for the architectural specification, docs/api-sketch.org
+for the v1 API surface, and docs/mcar-asymptotics.org for the asymptotic
+theory under MCAR.
+"""
+
+# ruff: noqa: E402
+# Imports below the float64 config update intentionally come after a
+# non-import statement.
+from importlib.metadata import PackageNotFoundError, version
+
+# Enable float64 in JAX *before* any module-under-this-package import
+# touches jax.numpy. JAX defaults to float32 (a deep-learning convention);
+# for a numerical-statistics framework where Cholesky pivots, condition
+# numbers, and gradient norms cross many orders of magnitude, float32
+# precision is the wrong baseline. Notable symptoms with float32:
+#   - optimistix LM cannot certify convergence at rtol=1e-8 because the
+#     float32 noise floor on a O(0.1)-magnitude whitened residual is
+#     around 2e-8;
+#   - theta_hat itself drifts at the third significant digit relative to
+#     the float64 solution.
+# Users who explicitly want float32 can override after import with:
+#     jax.config.update("jax_enable_x64", False)
+import jax as _jax
+
+_jax.config.update("jax_enable_x64", True)
+
+# Public re-exports.
+from emu_gmm.covariance import (
+    AnalyticalCovariance,
+    ClusteredCovariance,
+    DesignAwareCovariance,
+    IIDCovariance,
+    StratifiedCovariance,
+    SumCovariance,
+    SyntheticCovariance,
+)
+from emu_gmm.diagnostics import (
+    build_diagnostics,
+    build_optimizer_health,
+    compute_cond_info,
+)
+from emu_gmm.estimator import build_estimator, estimate
+from emu_gmm.inference import (
+    AdaptiveBootstrapResult,
+    BootstrapMean,
+    BootstrapPValue,
+    BootstrapQuantile,
+    BootstrapSE,
+    ClusterBootstrapResult,
+    JTestResult,
+    KConfidenceSet,
+    KStatisticResult,
+    WildBootstrapResult,
+    adaptive_bootstrap,
+    cluster_bootstrap,
+    j_test,
+    k_confidence_set,
+    k_statistic,
+    moment_wild_bootstrap,
+)
+from emu_gmm.manifolds import (
+    Euclidean,
+    ManifoldLeaf,
+    ManifoldParam,
+    Positive,
+    Product,
+    PSDFixedRank,
+    riemannian_lm,
+)
+from emu_gmm.measures import (
+    AnalyticalMeasure,
+    EmpiricalMeasure,
+    SyntheticMeasure,
+)
+from emu_gmm.numerics import ridge_inverse
+from emu_gmm.optimizer import linear_solver, optimistix_lm, scipy_lm
+from emu_gmm.parameter_space import ParameterSpace, on
+from emu_gmm.penalty import PenaltyStrategy, TikhonovPenalty
+from emu_gmm.regularization import DiagonalTikhonov
+from emu_gmm.types import (
+    CovarianceStrategy,
+    Diagnostics,
+    Emu_GMM_DimensionError,
+    EstimationResult,
+    FitRecord,
+    LabelContext,
+    Measure,
+    Optimizer,
+    OptimizerInfo,
+    RegularizationStrategy,
+    StructuralModel,
+    WeightingStrategy,
+)
+from emu_gmm.weighting import (
+    CUE,
+    ContinuouslyUpdated,
+    Fixed,
+    Identity,
+    IteratedWeighting,
+)
+
+try:
+    __version__ = version("emu-gmm")
+except PackageNotFoundError:
+    __version__ = "0.0.0"
+
+__all__ = [
+    "__version__",
+    # Entry point
+    "estimate",
+    "build_estimator",
+    # Inference
+    "k_confidence_set",
+    "k_statistic",
+    "KConfidenceSet",
+    "KStatisticResult",
+    # Measures
+    "SyntheticMeasure",
+    "AnalyticalMeasure",
+    "EmpiricalMeasure",
+    # Covariance strategies
+    "SyntheticCovariance",
+    "AnalyticalCovariance",
+    "IIDCovariance",
+    "ClusteredCovariance",
+    "StratifiedCovariance",
+    "DesignAwareCovariance",
+    "SumCovariance",
+    # Weighting strategies
+    "Identity",
+    "Fixed",
+    "ContinuouslyUpdated",
+    "CUE",
+    "IteratedWeighting",
+    # Regularization
+    "DiagonalTikhonov",
+    # Numerics helpers
+    "ridge_inverse",
+    # Penalty (in-objective parameter penalty)
+    "TikhonovPenalty",
+    # Optimisers
+    "optimistix_lm",
+    "scipy_lm",
+    "linear_solver",
+    "riemannian_lm",
+    # Inference helpers
+    "cluster_bootstrap",
+    "ClusterBootstrapResult",
+    "j_test",
+    "JTestResult",
+    # Result / diagnostics types
+    "EstimationResult",
+    "FitRecord",
+    "Diagnostics",
+    "OptimizerInfo",
+    "LabelContext",
+    # Errors
+    "Emu_GMM_DimensionError",
+    # Diagnostics builders
+    "build_diagnostics",
+    "build_optimizer_health",
+    "compute_cond_info",
+    # Inference
+    "moment_wild_bootstrap",
+    "WildBootstrapResult",
+    # Adaptive (precision-targeted) bootstrap stopping (#91)
+    "adaptive_bootstrap",
+    "AdaptiveBootstrapResult",
+    "BootstrapMean",
+    "BootstrapSE",
+    "BootstrapQuantile",
+    "BootstrapPValue",
+    # Manifolds (parameter geometry — a first-class problem-tuple menu, peer
+    # to Measures / Covariance strategies; wrap a non-scalar parameter leaf in
+    # ManifoldLeaf(array, manifold) and estimate via the ordinary entry point.
+    # ManifoldSpec / LeafSpec stay submodule-internal — they are machinery,
+    # not part of the user-composed tuple.)
+    "Euclidean",
+    "PSDFixedRank",
+    "Product",
+    "Positive",
+    "ManifoldLeaf",
+    # Parameter-space declaration layer (#107): declare field -> manifold
+    # geometry once in a class, then ParameterSpace.point([seed]) lowers to a
+    # ManifoldLeaf pytree (a valid theta_init) consumed by estimate(parameters=).
+    "ParameterSpace",
+    "on",
+    # Protocols (for type-checking user code)
+    "Measure",
+    "CovarianceStrategy",
+    "WeightingStrategy",
+    "RegularizationStrategy",
+    "PenaltyStrategy",
+    "Optimizer",
+    "StructuralModel",
+    "ManifoldParam",
+]