Add CLAUDE.md and update arviz usage

.github/workflows/ci_cron_weekly.yml

@@ -19,7 +19,7 @@ jobs:
     - name: Set up Python to build docs with sphinx
       uses: actions/setup-python@v4
       with:
-        python-version: '3.13'
+        python-version: '3.14'
     - name: Install base dependencies
       run: |
         python -m pip install --upgrade pip

.github/workflows/ci_tests.yml

@@ -24,19 +24,30 @@ jobs:
         include:
           - name: Code style checks
             os: ubuntu-latest
-            python: '3.13'
+            python: '3.14'
             toxenv: codestyle
 
-          - name: Python 3.13 with required dependencies and coverage checking
+          - name: Python 3.14 with required dependencies and coverage checking
+            os: macos-latest
+            python: '3.14'
+            toxenv: py314-test-cov
+
+          - name: Python 3.13 with required dependencies
             os: macos-latest
             python: '3.13'
-            toxenv: py313-test-cov
+            toxenv: py313-test
 
-          - name: Python 3.13 with nightly wheels for key dependencies
+          - name: Python 3.14 with nightly wheels for key dependencies (macOS)
             continue-on-error: true
             os: macos-latest
-            python: '3.13'
-            toxenv: py313-test-devdeps
+            python: '3.14'
+            toxenv: py314-test-devdeps
+
+          - name: Python 3.14 with nightly wheels for key dependencies (Linux)
+            continue-on-error: true
+            os: ubuntu-latest
+            python: '3.14'
+            toxenv: py314-test-devdeps
 
     steps:
     - name: Checkout code

CLAUDE.md

@@ -0,0 +1,68 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+`pytelpoint` is a Python library for analyzing telescope pointing performance at the MMT Observatory using Bayesian/probabilistic methods via PyMC. It fits standard TPOINT-compatible pointing models to az/el encoder data.
+
+## Environment Setup
+
+```bash
+pip install -e ".[test]"
+```
+
+On macOS, `fitting.py` enforces `/usr/bin/clang++` to avoid conflicts between conda-forge's clang and Xcode command-line tools.
+
+## Common Commands
+
+```bash
+# Run all tests
+pytest --pyargs pytelpoint docs
+
+# Run a single test file
+pytest pytelpoint/tests/test_io.py
+
+# Run a single test
+pytest pytelpoint/tests/test_io.py::test_read_azel_datfile
+
+# Check code style (max line length 132)
+flake8 pytelpoint --count --max-line-length=132
+
+# Format code
+black pytelpoint
+
+# Run via tox (uses conda for pymc)
+tox -e py313-test
+
+# Build docs
+tox -e build_docs
+```
+
+## Architecture
+
+All core modules use `SkyCoord` (AstropY) as the primary data carrier. The typical workflow is:
+
+1. **IO** (`io.py`): Read pointing run data → returns `(coo_ref, coo_meas)` pair of `SkyCoord` objects in an `AltAz` frame tied to `MMT_LOCATION`.
+2. **Stats** (`stats.py`): Compute sky RMS / population SD from coordinate pairs (TPOINT-equivalent diagnostics).
+3. **Transform** (`transform.py`): Apply an 8-term az/el pointing model (`azel_model`) analytically to raw encoder coordinates.
+4. **Fitting** (`fitting.py`): Fit the pointing model probabilistically using PyMC (`azel_fit`), returning an `arviz.InferenceData` object. `best_fit_pars` summarizes the posterior using `arviz_stats.summary`.
+5. **Visualization** (`visualization.py`): `plot_posterior` and `plot_corner` for inspecting PyMC fit results using `arviz_plots`.
+
+### Pointing model terms (TPOINT-compatible)
+
+All terms are in arcseconds. The 8 standard az/el terms are: `ia` (az index), `ie` (el index), `an` (N-S axis tilt), `aw` (E-W axis tilt), `ca` (collimation), `npae` (az/el non-perpendicularity), `tf` (tube flexure ∝ cos(el)), `tx` (tube flexure ∝ cot(el)).
+
+### Key design note
+
+`fitting.py` uses `pm.math` (Theano/pytensor ops) for the model functions, while `transform.py` uses `numpy` — they implement the same math but for different contexts (PyMC model graph vs. direct application).
+
+### Test data
+
+`pytelpoint/test_data/` contains three real pointing run files used in tests. `k_and_e.dat` is the processed (4-column) format; `point_20210821.dat` and `point_20250326.dat` are raw mount code output files.
+
+## Testing Notes
+
+- Tests live in `pytelpoint/tests/`. Doctests in `.rst` files under `docs/` are also run.
+- `test_mc_fitting` runs a real PyMC MCMC chain (with reduced `nsamp=200, ntune=200`) — it is slow.
+- `pytest.ini` options (in `pyproject.toml`) enable `--doctest-rst` and treat most warnings as errors.

pyproject.toml

@@ -12,7 +12,8 @@ dependencies = [
     "astropy",
     "scipy",
     "matplotlib",
-    "arviz",
+    "arviz-stats",
+    "arviz-plots",
     "corner",
     "pymc",
     "pandas",

pytelpoint/fitting.py

@@ -3,7 +3,7 @@
 
 import numpy as np
 
-import arviz
+import arviz_stats
 import pymc as pm
 
 import astropy.units as u
@@ -173,7 +173,7 @@ def best_fit_pars(idata):
     pointing_pars : dict
         Best-fit pointing parameters
     """
-    t_fit = arviz.summary(idata, round_to=8)
+    t_fit = arviz_stats.summary(idata, round_to=8)
     pointing_pars = {}
     for p in t_fit.index:
         pointing_pars[p] = t_fit.loc[p, "mean"]

pytelpoint/visualization.py

@@ -6,7 +6,7 @@
 import matplotlib.pyplot as plt
 
 import corner
-import arviz
+import arviz_plots
 
 import astropy.units as u
 from astropy.visualization import hist
@@ -332,8 +332,8 @@ def plot_posterior(idata):
 
     Returns
     -------
-    fig : `matplotlib.figure.Figure` instance
-        Figure object containing the corner plot.
+    pc : `~arviz_plots.PlotCollection`
+        PlotCollection containing the posterior distribution plots.
     """
-    fig = arviz.plot_posterior(idata)
-    return fig
+    pc = arviz_plots.plot_dist(idata)
+    return pc