rcrpy 0.1.0__tar.gz

rcrpy-0.1.0/.gitignore

@@ -0,0 +1,162 @@
+# ============================================================================
+# Python
+# ============================================================================
+__pycache__/
+*.py[cod]
+*$py.class
+*.pyo
+*.pyd
+
+# Distribution / packaging
+build/
+dist/
+wheels/
+*.egg
+*.egg-info/
+*.whl
+MANIFEST
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage
+.pytest_cache/
+.tox/
+.nox/
+.coverage
+.coverage.*
+htmlcov/
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+
+# Type checkers / linters
+.mypy_cache/
+.dmypy.json
+dmypy.json
+.pyre/
+.ruff_cache/
+.pytype/
+
+# Jupyter
+.ipynb_checkpoints/
+*.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# Sphinx documentation
+docs/_build/
+
+# ============================================================================
+# Virtual environments (project-specific names + the standard ones)
+# ============================================================================
+.venv/
+.venv-*/
+venv/
+env/
+ENV/
+env.bak/
+venv.bak/
+
+# ============================================================================
+# C++ build artifacts
+# ============================================================================
+*.o
+*.obj
+*.a
+*.lib
+*.so
+*.so.*
+*.dylib
+*.dll
+*.exe
+*.out
+*.app
+
+# CMake
+CMakeCache.txt
+CMakeFiles/
+CMakeScripts/
+cmake_install.cmake
+install_manifest.txt
+CTestTestfile.cmake
+_deps/
+
+# Build directories
+build/
+build-*/
+cmake-build-*/
+
+# ============================================================================
+# IDE / editor state
+# ============================================================================
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+.project
+.settings/
+.classpath
+
+# JetBrains
+*.iml
+.idea_modules/
+
+# ============================================================================
+# OS junk
+# ============================================================================
+.DS_Store
+.DS_Store?
+._*
+.Spotlight-V100
+.Trashes
+ehthumbs.db
+Thumbs.db
+desktop.ini
+$RECYCLE.BIN/
+
+# ============================================================================
+# Claude / agent local state (conversation transcripts live here on the user's
+# machine; do not commit even by accident if a .claude/ ends up in the repo).
+# ============================================================================
+.claude/
+.aider*
+CLAUDE.md.local
+
+# ============================================================================
+# Logs and temporary files
+# ============================================================================
+*.log
+*.tmp
+*.bak
+*.orig
+*.rej
+
+# ============================================================================
+# Local secrets / credentials (defense in depth — nothing should ever be here)
+# ============================================================================
+.env
+.env.local
+.env.*.local
+*.pem
+*.key
+secrets/
+
+# ============================================================================
+# Project-specific
+# ============================================================================
+# Compiled rcr extension lands here after pip install -e cpp/
+cpp/rcr.*.pyd
+cpp/rcr.*.so
+cpp/build/
+
+# Background-task transcripts (Claude Code writes these locally during sessions)
+*.output
+
+# Editor swap files in nested dirs
+**/*~

rcrpy-0.1.0/LICENSE

@@ -0,0 +1,48 @@
+COPYRIGHT AND PERMISSION NOTICE
+UNC Software "Robust Chauvenet Outlier Rejection" (Ref No. 17-0086)
+Copyright (C) 2020 The University of North Carolina at Chapel Hill
+All rights reserved.
+
+The University of North Carolina at Chapel Hill ("UNC") and the developers ("Developers") of Robust
+Chauvenet Outlier Rejection software ("Software") give recipient ("Recipient") and Recipient's Institution
+("Institution") permission to use, copy and redistribute the software in source and binary forms, with or without
+modification for non-commercial purposes only provided that the following conditions are met:
+
+1) All copies and redistributions of Software in binary form and/or source code, related documentation
+and/or other materials provided with the Software must reproduce and retain the above copyright notice,
+this list of conditions and the following disclaimer.
+
+2) Recipient will provide the Developers with feedback on the use of the Software in their research. The
+Developers and UNC are permitted to use any information Recipient provides in making changes to the
+Software. All feedback, bug reports and technical questions shall be sent to: Prof. Daniel Reichart
+(reichart@email.unc.edu).
+
+3) Recipient acknowledges that the Developers, UNC and its licensees may develop modifications to
+Software that may be substantially similar to Recipient's modifications of Software, and that the
+Developers, UNC and its licensees shall not be constrained in any way by Recipient in UNC's or its
+licensees' use or management of such modifications. Recipient acknowledges the right of the
+Developers and UNC to prepare and publish modifications to Software that may be substantially similar
+or functionally equivalent to your modifications and improvements, and if Recipient or Institution
+obtains patent protection for any modification or improvement to Software, Recipient and Institution
+agree not to allege or enjoin infringement of their patent by the Developers, UNC or any of UNC's
+licensees obtaining modifications or improvements to Software from the UNC or the Developers.
+
+4) Recipient and Developer will acknowledge in their respective publications the contributions made to
+each other's research involving or based on the Software. The current citations for Software are:
+M. P. Maples, D. E. Reichart, N. C. Konz, T. A. Berger, A. S. Trotter, J. R. Martin, D. A. Dutton, M. L.
+Paggen, R. E. Joyner, C. P. Salemi, 2018 ApJS, 238.1, p. 2
+
+5) Any party desiring a license to use the Software for commercial purposes shall contact UNC's Office of
+Commercialization and Economic Development at 919-966-3929 or oced@unc.edu.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS, CONTRIBUTORS, AND THE
+UNIVERSITY OF NORTH CAROLINA AT CHAPEL HILL "AS IS" AND ANY EXPRESS OR IMPLIED
+WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
+EVENT SHALL THE COPYRIGHT OWNER, CONTRIBUTORS OR THE UNIVERSITY OF NORTH
+CAROLINA AT CHAPEL HILL BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
+BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
+IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.

rcrpy-0.1.0/PKG-INFO

@@ -0,0 +1,247 @@
+Metadata-Version: 2.4
+Name: rcrpy
+Version: 0.1.0
+Summary: Robust Chauvenet Rejection (RCR), Python reimplementation
+Project-URL: Homepage, https://github.com/nickk124/RCR
+Project-URL: Repository, https://github.com/nickk124/RCR
+Project-URL: Issues, https://github.com/nickk124/RCR/issues
+Project-URL: Paper, https://arxiv.org/abs/1807.05276
+Author-email: Haislip <haislip@physics.unc.edu>
+Maintainer-email: UNC Physics RCR successors <haislip@physics.unc.edu>
+License: COPYRIGHT AND PERMISSION NOTICE
+        UNC Software "Robust Chauvenet Outlier Rejection" (Ref No. 17-0086)
+        Copyright (C) 2020 The University of North Carolina at Chapel Hill
+        All rights reserved.
+        
+        The University of North Carolina at Chapel Hill ("UNC") and the developers ("Developers") of Robust
+        Chauvenet Outlier Rejection software ("Software") give recipient ("Recipient") and Recipient's Institution
+        ("Institution") permission to use, copy and redistribute the software in source and binary forms, with or without
+        modification for non-commercial purposes only provided that the following conditions are met:
+        
+        1) All copies and redistributions of Software in binary form and/or source code, related documentation
+        and/or other materials provided with the Software must reproduce and retain the above copyright notice,
+        this list of conditions and the following disclaimer.
+        
+        2) Recipient will provide the Developers with feedback on the use of the Software in their research. The
+        Developers and UNC are permitted to use any information Recipient provides in making changes to the
+        Software. All feedback, bug reports and technical questions shall be sent to: Prof. Daniel Reichart
+        (reichart@email.unc.edu).
+        
+        3) Recipient acknowledges that the Developers, UNC and its licensees may develop modifications to
+        Software that may be substantially similar to Recipient's modifications of Software, and that the
+        Developers, UNC and its licensees shall not be constrained in any way by Recipient in UNC's or its
+        licensees' use or management of such modifications. Recipient acknowledges the right of the
+        Developers and UNC to prepare and publish modifications to Software that may be substantially similar
+        or functionally equivalent to your modifications and improvements, and if Recipient or Institution
+        obtains patent protection for any modification or improvement to Software, Recipient and Institution
+        agree not to allege or enjoin infringement of their patent by the Developers, UNC or any of UNC's
+        licensees obtaining modifications or improvements to Software from the UNC or the Developers.
+        
+        4) Recipient and Developer will acknowledge in their respective publications the contributions made to
+        each other's research involving or based on the Software. The current citations for Software are:
+        M. P. Maples, D. E. Reichart, N. C. Konz, T. A. Berger, A. S. Trotter, J. R. Martin, D. A. Dutton, M. L.
+        Paggen, R. E. Joyner, C. P. Salemi, 2018 ApJS, 238.1, p. 2
+        
+        5) Any party desiring a license to use the Software for commercial purposes shall contact UNC's Office of
+        Commercialization and Economic Development at 919-966-3929 or oced@unc.edu.
+        THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS, CONTRIBUTORS, AND THE
+        UNIVERSITY OF NORTH CAROLINA AT CHAPEL HILL "AS IS" AND ANY EXPRESS OR IMPLIED
+        WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+        MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
+        EVENT SHALL THE COPYRIGHT OWNER, CONTRIBUTORS OR THE UNIVERSITY OF NORTH
+        CAROLINA AT CHAPEL HILL BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+        EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+        PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
+        BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
+        IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+        ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+        POSSIBILITY OF SUCH DAMAGE.
+License-File: LICENSE
+Keywords: astronomy,chauvenet,outlier-rejection,robust-statistics,statistics
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: Other/Proprietary License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering
+Classifier: Topic :: Scientific/Engineering :: Astronomy
+Classifier: Topic :: Scientific/Engineering :: Physics
+Requires-Python: >=3.11
+Requires-Dist: numpy>=1.26
+Requires-Dist: scipy>=1.11
+Description-Content-Type: text/markdown
+
+# rcrpy
+
+Python reimplementation of Robust Chauvenet Rejection (RCR) — statistical
+outlier rejection that stays accurate on samples that are >85% contaminated.
+
+Based on the algorithm in
+[Maples et al. 2018, ApJS 238.1](https://arxiv.org/abs/1807.05276).
+This package is a numpy / scipy port of the original C++/pybind11
+implementation at [`../cpp/`](../cpp/), preserving algorithmic parity
+while running pure Python — no compiler toolchain required.
+
+## Install
+
+```bash
+pip install rcrpy
+```
+
+Runtime deps: `numpy>=1.26`, `scipy>=1.11`. Python ≥3.11.
+
+## Quick start
+
+### Single-value RCR
+
+Reject outliers from a 1-D dataset and recover the underlying central
+value and width.
+
+```python
+import rcrpy
+import numpy as np
+
+# Heavily contaminated data: mu=0, sigma=1, plus one-sided outliers
+rng = np.random.default_rng(42)
+y = np.concatenate([
+    rng.normal(0, 1, size=150),                # clean
+    np.abs(rng.normal(0, 10, size=850)),       # 85% contamination
+])
+
+r = rcrpy.RCR(rcrpy.RejectionTech.LS_MODE_68)
+r.perform_rejection(y.tolist())
+
+print(f"Recovered mu     = {r.result.mu:.3f}    (truth: 0)")
+print(f"Recovered sigma  = {r.result.sigma:.3f} (truth: 1)")
+print(f"Points kept      = {r.result.flags.sum()} / {len(y)}")
+```
+
+### Functional-form RCR (linear model fit with outlier rejection)
+
+Fit a parametric model to (x, y) data while rejecting outliers.
+
+```python
+import rcrpy
+import numpy as np
+
+# Linear data with outliers
+rng = np.random.default_rng(0)
+x = np.linspace(-5, 5, 100)
+y = 2.0 + 1.5 * x + rng.normal(0, 0.3, size=x.size)
+# Inject 20 outliers
+out = rng.choice(x.size, size=20, replace=False)
+y[out] += rng.normal(15, 5, size=20)
+
+def linear(x, params):
+    return params[0] + params[1] * x
+
+def d_linear_b(x, params):
+    return 1.0
+
+def d_linear_m(x, params):
+    return x
+
+model = rcrpy.FunctionalForm(
+    linear, x, y, [d_linear_b, d_linear_m], guess=[0.0, 0.0],
+)
+r = rcrpy.RCR(rcrpy.RejectionTech.LS_MODE_68)
+r.set_parametric_model(model)
+r.perform_rejection(y.tolist())
+
+b, m = model.result.parameters
+print(f"Recovered b = {b:.3f} (truth: 2.0)")
+print(f"Recovered m = {m:.3f} (truth: 1.5)")
+print(f"Kept {int(r.result.flags.sum())} / {len(y)} points")
+```
+
+## Rejection techniques
+
+| `RejectionTech.*` | Use case |
+|---|---|
+| `LS_MODE_68` | Symmetric uncontaminated distribution, one-sided contaminants |
+| `LS_MODE_DL` | Mixed one-sided + two-sided contaminants |
+| `SS_MEDIAN_DL` | Symmetric uncontaminated distribution, two-sided contaminants |
+| `ES_MODE_DL` | Mildly asymmetric uncontaminated distribution and/or very small N |
+
+> ⚠️ **Known algorithm bug**: `ES_MODE_DL` combined with `FunctionalForm`
+> (parametric models) exhibits a runaway-rejection cascade that drops
+> ~75% of inliers even on perfectly clean data. **This is an algorithm bug
+> inherited from the C++ source** — the C++ oracle exhibits the same
+> behavior; it is not a port deficiency. Empirical characterization is in
+> [`benchmarks/es_mode_dl_truth_test.py`](benchmarks/es_mode_dl_truth_test.py).
+> All other combinations of rejection technique × model type work
+> correctly. **For functional-form fits, use one of `LS_MODE_68`,
+> `LS_MODE_DL`, or `SS_MEDIAN_DL`.** `ES_MODE_DL` remains correct for
+> single-value (non-parametric) rejection.
+
+See [agents/python_vs_rust_plan.md](../agents/python_vs_rust_plan.md) and
+[Maples et al. 2018](https://arxiv.org/abs/1807.05276) for the science.
+
+## More invocations
+
+| Want… | Call |
+|---|---|
+| Bulk rejection (faster on large N) | `r.perform_bulk_rejection(y)` |
+| Weighted RCR | `r.perform_rejection(y, w=weights)` |
+| User-defined "candidate" mu | subclass `rcrpy.NonParametric`, `r.set_non_parametric_model(model)` |
+| Add Gaussian / bounded priors | `rcrpy.Priors(prior_type=..., gaussian_params=...)` → pass via `priors=` to `FunctionalForm` |
+| Pivot search for power-law / exponential fits | pass `pivot_function=...` and `pivot_guess=...` to `FunctionalForm` |
+| ND independent variable | pass `xdata` as `(N, D)` array; user functions take vector `x` |
+
+## Status
+
+| Slice | Status |
+|---|---|
+| Phase 1 single-value RCR (4 techs × iter/bulk × weighted/unweighted) | ✅ 100%; bit-identical to C++ oracle at rtol=1e-12 (worst case 1.7e-15 = 8 ULPs) |
+| Phase 2 functional form (1D + ND, priors, pivot search, paramuncertainty) | ✅ 99%; parity with C++ at rtol=5% on 6 sweep scenarios (RNG-limited) |
+| Packaging / CI / docs | 🚧 0.1.0 release; collecting real-user feedback before 1.0 |
+
+85 tests passing (+ 1 documented xfail). Reproduce parity with
+[`benchmarks/reflect_parity.py`](benchmarks/reflect_parity.py); benchmark
+with [`benchmarks/diagnostics.py`](benchmarks/diagnostics.py) and
+[`benchmarks/diagnostics_functional.py`](benchmarks/diagnostics_functional.py).
+
+## Layout
+
+```
+python/
+├── pyproject.toml
+├── README.md             (you are here)
+├── LICENSE               (mirrors ../cpp/LICENSE — academic / non-commercial)
+├── src/rcrpy/
+│   ├── __init__.py       public API re-exports
+│   ├── api.py            RCR class, RCRResults, RejectionTech, MuType
+│   ├── stats.py          median, halfSampleMode, robust-sigma helpers
+│   ├── rejection.py      iterative + bulk loops (lower/single/each sigma)
+│   ├── tables.py         unity tables ported verbatim from cpp/src/RCR.cpp
+│   ├── nonparametric.py  NonParametric base class
+│   └── functional.py     FunctionalForm + Priors for parametric fits
+├── tests/                pytest suite (85 + 1 xfailed)
+├── docs/                 markdown tutorials + PUBLISHING checklist
+├── benchmarks/           parity + perf scripts (not part of the wheel)
+└── scripts/              one-off codegen + post-install smoke check
+```
+
+## Citing
+
+If you use `rcrpy` in academic work, please cite the original paper:
+
+```bibtex
+@article{maples2018robust,
+    title  = {Robust Chauvenet Outlier Rejection},
+    author = {Maples, M.P. and Reichart, D.E. and Konz, N.C. and Berger, T.A.
+              and Trotter, A.S. and Martin, J.R. and Dutton, D.A.
+              and Paggen, M.L. and Joyner, R.E. and Salemi, C.P.},
+    journal = {The Astrophysical Journal Supplement Series},
+    volume  = {238}, number = {1}, pages = {2}, year = {2018},
+}
+```
+
+## License
+
+See [LICENSE](LICENSE) — mirrors the upstream C++ license. Free for
+academic and non-commercial use; contact UNC's Office of
+Commercialization and Economic Development for commercial licensing.

rcrpy-0.1.0/README.md

@@ -0,0 +1,171 @@
+# rcrpy
+
+Python reimplementation of Robust Chauvenet Rejection (RCR) — statistical
+outlier rejection that stays accurate on samples that are >85% contaminated.
+
+Based on the algorithm in
+[Maples et al. 2018, ApJS 238.1](https://arxiv.org/abs/1807.05276).
+This package is a numpy / scipy port of the original C++/pybind11
+implementation at [`../cpp/`](../cpp/), preserving algorithmic parity
+while running pure Python — no compiler toolchain required.
+
+## Install
+
+```bash
+pip install rcrpy
+```
+
+Runtime deps: `numpy>=1.26`, `scipy>=1.11`. Python ≥3.11.
+
+## Quick start
+
+### Single-value RCR
+
+Reject outliers from a 1-D dataset and recover the underlying central
+value and width.
+
+```python
+import rcrpy
+import numpy as np
+
+# Heavily contaminated data: mu=0, sigma=1, plus one-sided outliers
+rng = np.random.default_rng(42)
+y = np.concatenate([
+    rng.normal(0, 1, size=150),                # clean
+    np.abs(rng.normal(0, 10, size=850)),       # 85% contamination
+])
+
+r = rcrpy.RCR(rcrpy.RejectionTech.LS_MODE_68)
+r.perform_rejection(y.tolist())
+
+print(f"Recovered mu     = {r.result.mu:.3f}    (truth: 0)")
+print(f"Recovered sigma  = {r.result.sigma:.3f} (truth: 1)")
+print(f"Points kept      = {r.result.flags.sum()} / {len(y)}")
+```
+
+### Functional-form RCR (linear model fit with outlier rejection)
+
+Fit a parametric model to (x, y) data while rejecting outliers.
+
+```python
+import rcrpy
+import numpy as np
+
+# Linear data with outliers
+rng = np.random.default_rng(0)
+x = np.linspace(-5, 5, 100)
+y = 2.0 + 1.5 * x + rng.normal(0, 0.3, size=x.size)
+# Inject 20 outliers
+out = rng.choice(x.size, size=20, replace=False)
+y[out] += rng.normal(15, 5, size=20)
+
+def linear(x, params):
+    return params[0] + params[1] * x
+
+def d_linear_b(x, params):
+    return 1.0
+
+def d_linear_m(x, params):
+    return x
+
+model = rcrpy.FunctionalForm(
+    linear, x, y, [d_linear_b, d_linear_m], guess=[0.0, 0.0],
+)
+r = rcrpy.RCR(rcrpy.RejectionTech.LS_MODE_68)
+r.set_parametric_model(model)
+r.perform_rejection(y.tolist())
+
+b, m = model.result.parameters
+print(f"Recovered b = {b:.3f} (truth: 2.0)")
+print(f"Recovered m = {m:.3f} (truth: 1.5)")
+print(f"Kept {int(r.result.flags.sum())} / {len(y)} points")
+```
+
+## Rejection techniques
+
+| `RejectionTech.*` | Use case |
+|---|---|
+| `LS_MODE_68` | Symmetric uncontaminated distribution, one-sided contaminants |
+| `LS_MODE_DL` | Mixed one-sided + two-sided contaminants |
+| `SS_MEDIAN_DL` | Symmetric uncontaminated distribution, two-sided contaminants |
+| `ES_MODE_DL` | Mildly asymmetric uncontaminated distribution and/or very small N |
+
+> ⚠️ **Known algorithm bug**: `ES_MODE_DL` combined with `FunctionalForm`
+> (parametric models) exhibits a runaway-rejection cascade that drops
+> ~75% of inliers even on perfectly clean data. **This is an algorithm bug
+> inherited from the C++ source** — the C++ oracle exhibits the same
+> behavior; it is not a port deficiency. Empirical characterization is in
+> [`benchmarks/es_mode_dl_truth_test.py`](benchmarks/es_mode_dl_truth_test.py).
+> All other combinations of rejection technique × model type work
+> correctly. **For functional-form fits, use one of `LS_MODE_68`,
+> `LS_MODE_DL`, or `SS_MEDIAN_DL`.** `ES_MODE_DL` remains correct for
+> single-value (non-parametric) rejection.
+
+See [agents/python_vs_rust_plan.md](../agents/python_vs_rust_plan.md) and
+[Maples et al. 2018](https://arxiv.org/abs/1807.05276) for the science.
+
+## More invocations
+
+| Want… | Call |
+|---|---|
+| Bulk rejection (faster on large N) | `r.perform_bulk_rejection(y)` |
+| Weighted RCR | `r.perform_rejection(y, w=weights)` |
+| User-defined "candidate" mu | subclass `rcrpy.NonParametric`, `r.set_non_parametric_model(model)` |
+| Add Gaussian / bounded priors | `rcrpy.Priors(prior_type=..., gaussian_params=...)` → pass via `priors=` to `FunctionalForm` |
+| Pivot search for power-law / exponential fits | pass `pivot_function=...` and `pivot_guess=...` to `FunctionalForm` |
+| ND independent variable | pass `xdata` as `(N, D)` array; user functions take vector `x` |
+
+## Status
+
+| Slice | Status |
+|---|---|
+| Phase 1 single-value RCR (4 techs × iter/bulk × weighted/unweighted) | ✅ 100%; bit-identical to C++ oracle at rtol=1e-12 (worst case 1.7e-15 = 8 ULPs) |
+| Phase 2 functional form (1D + ND, priors, pivot search, paramuncertainty) | ✅ 99%; parity with C++ at rtol=5% on 6 sweep scenarios (RNG-limited) |
+| Packaging / CI / docs | 🚧 0.1.0 release; collecting real-user feedback before 1.0 |
+
+85 tests passing (+ 1 documented xfail). Reproduce parity with
+[`benchmarks/reflect_parity.py`](benchmarks/reflect_parity.py); benchmark
+with [`benchmarks/diagnostics.py`](benchmarks/diagnostics.py) and
+[`benchmarks/diagnostics_functional.py`](benchmarks/diagnostics_functional.py).
+
+## Layout
+
+```
+python/
+├── pyproject.toml
+├── README.md             (you are here)
+├── LICENSE               (mirrors ../cpp/LICENSE — academic / non-commercial)
+├── src/rcrpy/
+│   ├── __init__.py       public API re-exports
+│   ├── api.py            RCR class, RCRResults, RejectionTech, MuType
+│   ├── stats.py          median, halfSampleMode, robust-sigma helpers
+│   ├── rejection.py      iterative + bulk loops (lower/single/each sigma)
+│   ├── tables.py         unity tables ported verbatim from cpp/src/RCR.cpp
+│   ├── nonparametric.py  NonParametric base class
+│   └── functional.py     FunctionalForm + Priors for parametric fits
+├── tests/                pytest suite (85 + 1 xfailed)
+├── docs/                 markdown tutorials + PUBLISHING checklist
+├── benchmarks/           parity + perf scripts (not part of the wheel)
+└── scripts/              one-off codegen + post-install smoke check
+```
+
+## Citing
+
+If you use `rcrpy` in academic work, please cite the original paper:
+
+```bibtex
+@article{maples2018robust,
+    title  = {Robust Chauvenet Outlier Rejection},
+    author = {Maples, M.P. and Reichart, D.E. and Konz, N.C. and Berger, T.A.
+              and Trotter, A.S. and Martin, J.R. and Dutton, D.A.
+              and Paggen, M.L. and Joyner, R.E. and Salemi, C.P.},
+    journal = {The Astrophysical Journal Supplement Series},
+    volume  = {238}, number = {1}, pages = {2}, year = {2018},
+}
+```
+
+## License
+
+See [LICENSE](LICENSE) — mirrors the upstream C++ license. Free for
+academic and non-commercial use; contact UNC's Office of
+Commercialization and Economic Development for commercial licensing.

rcrpy-0.1.0/benchmarks/README.md

@@ -0,0 +1,26 @@
+# `rcrpy` benchmarks
+
+Standalone scripts for measuring `rcrpy` correctness and performance.
+None of these are part of the installed package — they're maintainer
+tools that live alongside the source.
+
+| Script | What it does |
+|---|---|
+| [`reflect_parity.py`](reflect_parity.py) | Sweep all 20 Phase 1 parity scenarios (single-value RCR × every tech × iterative+bulk × weighted+unweighted). Reports max relative scalar disagreement vs the C++ oracle. Worst case: 1.7×10⁻¹⁵. |
+| [`diagnostics.py`](diagnostics.py) | Per-scenario wall-clock + precision sweep for single-value RCR (port vs. C++ oracle). |
+| [`diagnostics_functional.py`](diagnostics_functional.py) | Same for functional-form RCR. Demonstrates the port is FASTER than the oracle on this workload (80×–6000× depending on N). |
+| [`bench_port_vs_oracle.py`](bench_port_vs_oracle.py) | One-shot bench on `data_singlevalue.csv` for quick perf checks. |
+
+## Running
+
+Each script is self-contained. From the repo root:
+
+```bash
+python rcrpy/benchmarks/reflect_parity.py
+python rcrpy/benchmarks/diagnostics.py
+python rcrpy/benchmarks/diagnostics_functional.py
+```
+
+All four require the legacy `rcr` C++ module (the oracle) plus `rcrpy`
+itself installed in the active environment — i.e., `pip install -e
+./python[dev]` covers everything.