Cov refactor

[beykyle]

A major refactor separating nuisance parameters from covariance instances, and both from observed data, allowing composition of nuisance terms and true hierarchical Bayesian inference. See design doc for details.

[Copilot]

Pull request overview

This PR implements a major architecture refactor that moves covariance/uncertainty modeling out of the old “likelihood model zoo” and into an explicit, composable stacked-covariance layer (rxmc.covariance). Constraints now own a single multivariate likelihood over their stacked observations, with correlated modes expressed as additive covariance Terms; Evidence becomes a weighted sum over independent constraints with parametric constraints auto-detected.

Changes:

• Introduces stacked covariance primitives (Term, ConstraintCovariance, factory helpers) and refactors Constraint, Evidence, Walker, and configs to use constraint-scoped covariance/likelihood parameters.
• Redefines Observation (and reaction observation subclasses) as “pure data + statistical term”, with reported systematics retained as inert metadata and only composed into covariance when explicitly requested.
• Adds GP predictive utilities (rxmc.predictive), extensive new tests/regression pins, and updates docs/CI to match the new API and workflows.

Reviewed changes

Copilot reviewed 40 out of 92 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
test/test_sampler.py	Updates sampler tests to use covariance Terms and adds Gibbs-weighting/validation tests.
test/test_regression.py	Adds regression pins for the systematic-default behavior change and covariance fast paths/coupling.
test/test_reaction_observation.py	Adjusts reaction observation tests for “statistical-only + explicit terms” contract; adds unit/systematics retention checks.
test/test_reaction_models.py	Adds real-solver smoke tests for reaction models.
test/test_predictive.py	Adds unit tests for new predictive GP/band utilities.
test/test_observation.py	Refactors observation tests to exercise statistical_term/systematic_terms and stacked covariance assembly.
test/test_evidence.py	Refactors evidence tests for auto-detected parametric constraints and cross-constraint parameter validation.
test/test_covariance.py	Adds comprehensive unit coverage for covariance term primitives, caching, and block/dense equivalence.
test/test_constraint.py	Adds extensive tests for stacked constraints, case A/B, param-count/name validation, and singular-guard behavior.
test/test_config.py	Updates calibration config tests for constraint-scoped params, tempering, and parametric-index alignment.
test/helpers.py	Adds shared test helpers (make_ctx, dense MVN reference loglike).
test/conftest.py	Ensures helpers is importable under pytest import modes.
src/rxmc/walker.py	Updates parameter consistency checks and applies evidence weighting in Gibbs conditional likelihood updates.
src/rxmc/predictive.py	Adds GP posterior predictive helpers and total predictive-band propagation.
src/rxmc/physical_model.py	Adds ScaledModel and PerObservationScaledModel for latent mean normalization on the model side.
src/rxmc/observation.py	Reworks Observation to emit a statistical Term and opt-in systematic term composition; removes fixed-covariance subclass.
src/rxmc/observation_from_measurement.py	Replaces observation setup with shared validation + normalized error kwargs helper.
src/rxmc/ias_pn_observation.py	Converts IAS observation to inherit Observation; composes errors via normalized kwargs; removes dynamic inheritance.
src/rxmc/evidence.py	Refactors to a single constraint list with parametric auto-detection + constraint-scoped parameter validation + weighted marginal helper.
src/rxmc/elastic_diffxs_observation.py	Converts elastic observation to inherit Observation; clarifies normalization conversions and error handling.
src/rxmc/covariance.py	Adds the new stacked covariance engine (terms, kernels, caching, block/dense distance evaluation).
src/rxmc/correlated_discrepancy_likelihood_model.py	Removes the old sklearn-GP discrepancy likelihood model implementation.
src/rxmc/constraint.py	Refactors constraint to stacked MVN with explicit covariance terms and likelihood functionals; adds Σ accessor and strong validation.
src/rxmc/config.py	Updates config validation, prediction ordering, and Gibbs conditional tempering to align with new evidence/constraint semantics.
src/rxmc/__init__.py	Exposes new covariance and predictive modules; removes old discrepancy module export.
README.md	Updates conceptual model and quickstart for explicit covariance terms + behavior-change note.
pyproject.toml	Updates isort first-party list and pytest defaults to run unit tests only by default.
examples/sampling_algos.ipynb	Updates notebook examples to new term-based covariance API.
examples/calibration_config_emcee_dynesty.ipynb	Updates notebook to new Constraint(likelihood=...) API and output text normalization.
examples/30s_optical_potential_calibration.ipynb	Updates notebook to new Constraint(likelihood=...) API.
environment.yml	Removes the conda environment file.
docs/installation.rst	Removes conda/mamba installation instructions.
docs/index.rst	Updates docs index wording and adds design to the toctree.
docs/generated/rxmc.predictive.total_predictive_band.rst	Adds generated API stub for predictive band helper.
docs/generated/rxmc.predictive.predictive_band.rst	Adds generated API stub for predictive band helper.
docs/generated/rxmc.predictive.gp_posterior_predictive.rst	Adds generated API stub for GP posterior predictive helper.
docs/generated/rxmc.physical_model.ScaledModel.rst	Adds generated API stub for ScaledModel.
docs/generated/rxmc.physical_model.PerObservationScaledModel.rst	Adds generated API stub for PerObservationScaledModel.
docs/generated/rxmc.observation.Observation.rst	Updates generated API stub to reflect removed/added Observation methods.
docs/generated/rxmc.observation.FixedCovarianceObservation.rst	Removes generated API stub for deleted class.
docs/generated/rxmc.likelihood_model.*.rst	Removes generated stubs for deleted likelihood-model-zoo classes; adds new functional likelihood stubs.
docs/generated/rxmc.constraint.Constraint.rst	Updates generated stub to include covariance matrix accessor and remove old API.
docs/generated/rxmc.config.CalibrationConfig.rst	Updates generated stub to include predict_parametric.
docs/examples.rst	Adds/organizes example notebooks in docs.
docs/design.md	Adds design doc describing stacked covariance architecture and scope decisions.
docs/conf.py	Updates exclude patterns for built docs.
docs/api.rst	Updates API docs to describe covariance terms, likelihood functionals, and predictive utilities.
.gitignore	Adds ignores for uv.lock, agent config, and built docs artifacts.
.github/workflows/ci.yml	Expands Python matrix, adds EXFOR DB caching/warming, parallel notebook runs, and adjusts timeouts/install steps.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.