cost-core 1.0.1__tar.gz

cost_core-1.0.1/CHANGELOG.md

@@ -0,0 +1,331 @@
+# Changelog
+
+All notable changes to `cost_core` are recorded here.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and the project follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+Entries say what moved and by how much. A number that changes is not a
+housekeeping detail here, because someone may have put the old one in a budget.
+
+## [Unreleased]
+
+### Fixed
+
+- The buy S-curve is readable again when the simulation throws a long tail.
+  The axis followed the single most extreme draw, and parameter uncertainty on
+  a short series is a t with a few degrees of freedom, so one iteration in five
+  thousand landing at 2.6x the median squeezed the whole curve into a vertical
+  line against the left spine. It now shows the 0.5th through 99.5th
+  percentile, widened to keep P50, P80, P90 and the point estimate inside it,
+  and the axis label says so whenever anything is cropped. No number moves: the
+  curve is still drawn in full and the percentiles are what they were.
+
+- `ce-core fit-lots` says which units its costs are in. The engine is scale
+  free and hands back whatever it is given, but its column headings read `($K)`,
+  carried over from the desktop tool whose input column is `AUC ($K)`, while
+  the charts format the same numbers as dollars. Feed it a file in dollars, as
+  the bundled `example_lots.csv` is, and a first-unit cost of $7.5M prints as
+  `T1 ($K) 7,477,686`, which is a 1,000x misread waiting to happen. The command
+  now states the convention in its output and in `--csv --help`, and the README
+  says it as well. The headings are unchanged, because they are the engine's
+  column names: the goldens pin them, the workbook writer builds sheets from
+  them and the desktop tool reads them.
+
+## [1.0.1] - 2026-09-11
+
+A tests-only release. The engine's code and every number it produces are the
+same as 1.0.0's; the one thing that reads differently is the version in the
+provenance row. What changed is the source distribution, which ships the test
+suite, and 1.0.0's failed on Linux.
+
+### Fixed
+
+- The test suite passes off Windows. Every golden and every frozen draw in it
+  was captured on Windows, and the first push of 1.0.0 failed the same six
+  tests on all six Linux lanes. Nothing in `cost_core` changed and no golden
+  was rewritten; the tests assumed more was portable than is.
+
+  Under Linux the same numpy on the same machine draws about 0.5% of the
+  analytic risk model's values one to nine ulps away from Windows, so the four
+  frozen-draw hashes can't hold there. That model is now pinned at p50, p80
+  and p90 on every platform, to 1e-12 relative, and at the hashes on Windows
+  only, where they still catch a change like scipy 1.18.1's that leaves the
+  percentiles alone.
+
+  The same last-bit differences move the MUPE and ZMPE refits in the
+  learning-curve block of `tests/goldens/lots_cost_core`, because
+  `cost_core.learning_curve` fits with central differences: fifty leaves, up
+  to 3.8e-8 relative on T1, and the `scatter_0.3` MUPE loop stops at 12 steps
+  instead of 11. Which leaves move depends on the CPU as well, through the
+  kernel OpenBLAS picks for it: three GitHub lanes, and OpenBLAS forced onto
+  its kernels without FMA, also moved an iteration count or two and flipped
+  the sign of the `Mean bias` the two exact-fit fixtures print as zero, both
+  ways between `-0.00%` and `+0.00%`. Off Windows, every MUPE and ZMPE statistic
+  in that block is now compared under a measured allowance,
+  `COMPARE_POLICY.expected_to_move_across_platforms`, the
+  three iteration counts aren't compared, and those `Mean bias` cells pass
+  when both sides print a zero. On Windows all of it stays at 1e-9. Nothing
+  else in any golden moves, anywhere. That makes Windows the place a numpy or
+  scipy cap raise has to be measured: under Linux, numpy 2.5.3 and scipy
+  1.18.1 now both pass the whole suite.
+
+  And a masked path in an error text keeps the separator of the machine that
+  wrote it, so off Windows the error-path comparison now reads
+  `<error_inputs>\` and `<error_inputs>/` as the same.
+
+## [1.0.0] - 2026-09-09
+
+The first release. This is the point where the library stopped being the back
+half of a desktop tool and became something installable on its own: the lot cost
+engine, the WBS roll-up and the shared estimator all live here, and the window
+lives in [lot-cost-model](https://github.com/MichaelFowler1/lot-cost-model),
+which today carries its own copy of the deterministic fit and imports this
+library for the risk half. The version says the interface is now something worth
+keeping stable, and the goldens are what hold it to that.
+
+### Added
+
+- A golden-master corpus under `tests/goldens`, read by
+  `tests/test_goldens.py`. It holds what the engine produced on 2026-09-06 for
+  36 engine cases, eleven WBS programmes (two of them `FULL_WBS` at a
+  different buy scale), the CSV and Excel front door, three command line runs
+  and the error texts, at full float precision. Every case
+  is rebuilt from inputs stored beside it and compared under
+  `COMPARE_POLICY.json`, so the tolerances are written down instead of hidden
+  in assertions. It exists because the steps that follow re-express the
+  engine's arithmetic, and a claim that nothing moved beyond one part in a
+  billion needs something to be measured against.
+
+- `cost_core.program`, the WBS roll-up. Several elements, each fitted to its
+  own analogy history or derived as a percentage or a flat amount, priced
+  against one shared lot schedule and correlated into one programme estimate.
+  It came from the desktop tool, where it was `wbs.py` and could not be used
+  without importing a window.
+
+- An `empirical` distribution type in `cost_core.monte_carlo`. Its spec is
+  `{"type": "empirical", "draws": ...}` and it is the draws themselves rather
+  than a family fitted to them. The quantile function is `numpy.interp` over
+  the Hazen plotting positions of the sorted draws, which returns each knot
+  bit for bit and clamps instead of extrapolating. Draws must be at least two,
+  finite and non-negative. The last is required rather than merely expected,
+  because the sampler clamps costs at zero after drawing them and one negative
+  draw would silently stop the column being a permutation of the draws.
+
+### Changed
+
+- The distribution name on PyPI is `cost-core`. The import name does not move:
+  `import cost_core` is unchanged, and setuptools normalises the two spellings
+  to the same artefact names, so the hyphen is only what you type after
+  `pip install`.
+
+- The sdist carries the whole of `tests/`, plus `example_lots.csv`,
+  `requirements.txt` and this file, through a new `MANIFEST.in`. setuptools'
+  default sweep picks up `tests/test_*.py` and nothing those modules import, so
+  the sdist used to unpack into a suite that could not run: `pytest tests/ -q`
+  never reached a test, because collection was interrupted by
+  `ModuleNotFoundError: No module named 'golden_support'`, and `--collect-only`
+  reported 704 tests collected and one error, the 48 in `test_goldens.py` being
+  the ones it lost. It now unpacks into the same suite the repository runs,
+  goldens included, and `pytest tests/ -q` gives the same 752 passed from an
+  unpacked release as from a checkout.
+
+- `requires-python` drops from `>=3.11` to `>=3.9`. Nothing here needed 3.11.
+  The PEP 604 unions that looked like they needed 3.10 all sit in annotations
+  in files carrying `from __future__ import annotations`, so none of them is
+  evaluated at runtime, and the suite passes on 3.9 unmodified. The last
+  3.9-capable releases of numpy, pandas, scipy and matplotlib -- 2.0.2, 2.3.3,
+  1.13.1 and 3.9.4 -- all satisfy the dependency floors, so the floors did not
+  have to move either.
+
+- CI runs 3.9, 3.10, 3.11, 3.12, 3.13 and 3.14 on every push, up from 3.11 and
+  3.12. Every lane now installs the package from the version range it declares
+  instead of from `requirements.txt`, because four of the five pins in that file
+  will not install on 3.9 and three of them will not install on 3.10, so a
+  pinned install could not have been what the lower lanes ran. One lane applies
+  the pins afterwards, so the goldens are still checked against the stack they
+  were recorded on. A lane on every version also uninstalls matplotlib and
+  imports the engine without it, which is the only thing that keeps the claim
+  below honest.
+
+  The 3.9 lane names `ubuntu-24.04` instead of `ubuntu-latest`. setup-python
+  takes its interpreters from the manifest in `actions/python-versions`, which
+  has linux builds of 3.9 up to 24.04 and no further while every other version in
+  the matrix has a 26.04 build. `ubuntu-latest` is 24.04 today, so the lane runs;
+  the day it moves on, a lane left following it would stop finding an interpreter
+  and 3.9 is too far past end of life for a new build to appear. Naming the image
+  keeps the floor actually tested rather than quietly skipped.
+
+- **numpy and scipy now carry an upper bound, `numpy<2.5` and `scipy<1.18`.**
+  These are reproducibility bounds, not compatibility ones. Above them the
+  library imports and runs; what it stops doing is reproducing the numbers this
+  release pins, and both effects were measured on Python 3.12 one package at a
+  time.
+
+  numpy 2.5.3 moves the unbiased refits in `tests/goldens/lots_cost_core`. The
+  MUPE iteration lands somewhere slightly different and sometimes takes a
+  different number of steps, 13 instead of 11 on the `scatter_0.3` case, so T1
+  goes 2,712.4962079 -> 2,712.4962290, about 8 parts in a billion, and the
+  nominally zero mean percentage error goes -4.3e-13 -> -2.6e-11. scipy 1.18.1
+  moves the same refits further, T1 to 2,712.4963311, about 4.5 parts in a
+  hundred million, and it also changes the draws the risk model produces, so
+  all four frozen-draw hashes in `tests/test_monte_carlo.py` fail as well.
+  Forty-one golden leaves move under scipy and forty-two under numpy, against
+  a tolerance of 1e-9 relative and 1e-12 absolute.
+
+  None of that is a bug in either library, and neither version was excluded
+  before. Both declare Requires-Python >=3.12, so the 3.12, 3.13 and 3.14 lanes
+  were all new enough to be offered them, and with nothing stopping them a lane
+  resolved both together and failed five tests on numbers: 71 golden leaves in
+  one case and the four frozen-draw hashes. The lanes whose tests actually ran
+  on a resolved stack were 3.12 and 3.13; 3.14 resolves the same pair and is
+  protected only because it applies `requirements.txt` before it runs the suite.
+  Raising a bound means rebaselining the goldens and the frozen draws under the
+  new stack and recording the movement, the way `COMPARE_POLICY.json` records
+  every other rebaseline here, rather than quietly widening a tolerance until
+  the failure stops.
+
+- **matplotlib is an optional extra, `cost-core[plots]`, rather than a
+  dependency.** It draws the charts and does nothing else here, so a caller who
+  wants numbers should not have to install a plotting stack to get them: the lot
+  engine, the CERs, the risk simulation and the Excel workbooks all import and
+  run without it. `cost_core.monte_carlo` no longer imports pyplot at module
+  scope, `plot_distribution` imports it when called, and anything that draws --
+  `cost_core.reporting.charts`, `ce-core full-run`, `ce-core fit-lots --simulate
+  --out` -- fails with a message naming the extra rather than a bare
+  `No module named 'matplotlib'`. `fit-lots --out` on its own draws nothing and
+  needs nothing: the buy S-curve is the only chart that command writes and it
+  only exists with `--simulate`, so the eight tables and the `ASSUMPTIONS.md`
+  come out of a bare install unchanged. **This changes what a `pip install` pulls: an
+  install that used to produce charts now needs `pip install 'cost-core[plots]'`.**
+
+- The lot engine, the analyst summary and the Excel workbook writer are the
+  desktop tool's current versions rather than the copy taken from it on 21
+  August. The engine itself had not moved in between; the summary had gained
+  three provenance rows, and the workbook writer three risk sheets, data
+  labels, a prediction band and an S-curve. Every function was carried across
+  from the tool's own source rather than retyped, so no number moved.
+
+- `save_complete_excel_workbook` now lives in `cost_core.reporting`, with the
+  other reporting code, and `cost_core.lotmodel` re-exports it. Importing
+  `cost_core.reporting` no longer pulls matplotlib: its chart and pipeline
+  names load on first use, so the workbook writers can be imported without it.
+
+- `generate_analyst_summary` takes an optional `provenance` mapping, so a
+  caller can stamp its own version rather than the library's.
+
+- The three lot models are fitted through `cost_core.fitting` instead of
+  through the private least squares, MUPE and ZMPE the lot engine carried.
+  The coefficients move by about a part in a hundred billion, and that move is
+  the old error leaving: against an exact solve in sixty digits the old normal
+  equations sat 1.1e-11 out and the new path sits 1.5e-14. Model selection
+  does not change on any of the 36 pinned cases and the analyst summary is
+  identical character for character.
+
+  Two things do move further than that. The unbiased refits on the
+  `Fit_Methods` table shift by up to 3.3e-7 relative on their coefficients,
+  because the shared estimator's iteration stops on its own least squares
+  tolerance and its ZMPE starts from the MUPE answer rather than from ordinary
+  least squares. One leaf goes further in relative terms and not in absolute
+  ones: the ZMPE learning exponent of the flat reference series is 0.0036265630
+  against 0.0036265683, which is 1.5e-6 relative on a move of 5.3e-9, and it is
+  large only because the number it divides into is nearly zero. That is why the
+  exponent carries an absolute allowance as well as a relative one;
+  `COMPARE_POLICY._atol_on_the_exponents` has the reasoning. The ZMPE row was
+  never reproducible better than about 1e-7 against itself, measured by
+  perturbing its own seed, so this is inside the noise that row always had.
+  And `DFFITS` now reports zero below two degrees
+  of freedom, where the leave-one-out variance is zero divided by zero and the
+  old formula returned whatever the rounding produced.
+
+- The significance gate no longer forms a rate t-statistic when the fit has no
+  residual spread to measure against. A series priced straight off the curve
+  reproduces itself exactly, so the coefficient and its standard error are
+  both at the floating-point floor and their ratio is decided by the last bit:
+  two solvers agreeing to twelve digits, pricing every lot identically, could
+  recommend different models. The t is reported as unavailable there instead,
+  which is the test `cost_core.cer.diagnostics` already applies before
+  dividing by the same quantity, and leaves the learning curve as the default.
+  Across every lot fit the test suite performs the ratio is either below
+  1.1e-14 or above 2.6e-4, so nothing real sits near the line.
+
+- The programme roll-up hands each fitted element's simulated buy totals to
+  the risk model directly, as an `empirical` marginal, instead of summarising
+  them as a two-parameter lognormal first. An element total is a sum of
+  correlated lognormal lot costs and is not itself lognormal, so no
+  two-parameter fit reproduced it. Because the draw count equals the
+  programme's iteration count, the copula returns a permutation of those very
+  draws, so **each element's percentiles inside the programme simulation are
+  now identical to its standalone percentiles**, at all 99 levels, under `==`.
+  Under the old handoff none of the 99 levels agreed and the worst was 0.79%
+  to 1.23% out.
+
+  The programme percentiles move as a result. They fall through the working
+  range and rise in the far tail, because the fitted lognormal was too fat in
+  the shoulder and too thin at the extreme:
+
+  | figure | TEST_PROGRAM | | FULL_WBS | |
+  | --- | --- | --- | --- | --- |
+  | | before -> after | change | before -> after | change |
+  | P50 | 197,359,861.00 -> 197,248,653.81 | -0.056% | 235,806,082.37 -> 235,679,973.43 | -0.053% |
+  | P80 | 199,334,505.70 -> 198,764,987.35 | -0.286% | 238,045,329.47 -> 237,399,495.65 | -0.271% |
+  | P90 | 200,431,772.55 -> 199,833,437.80 | -0.299% | 239,289,630.07 -> 238,611,118.46 | -0.284% |
+  | P99 | 202,989,641.04 -> 204,171,353.96 | +0.582% | 242,190,252.94 -> 243,530,315.40 | +0.553% |
+  | standard deviation | 2,361,702.58 -> 2,337,709.99 | -1.016% | 2,678,170.73 -> 2,650,963.13 | -1.016% |
+  | reserve to P80 | 2,168,821.67 -> 1,599,303.32 | -26.3% | 2,459,443.78 -> 1,813,609.96 | -26.3% |
+  | point estimate sits at | P46.3875 -> P47.9625 | +1.575 points | P46.3875 -> P47.9625 | +1.575 points |
+
+  All of these are at 8,000 iterations, seed 11, element correlation 0.25 and
+  lot correlation 0.30. The reserve is the headline: the old P80 was carrying
+  $569,518 on `TEST_PROGRAM` and $645,834 on `FULL_WBS` that came from the
+  summary's own error rather than from anything measured.
+
+  The standard deviation falls 1.02% although the element variances rise about
+  4%. A Gaussian copula attenuates a heavier-tailed marginal more, so the
+  achieved Pearson correlation between the three fitted elements drops from a
+  mean of 0.262 to 0.216 while Spearman does not move: pair by pair, 0.279,
+  0.241 and 0.267 become 0.225, 0.189 and 0.235. Variance rose, covariance
+  fell further, and the total narrowed.
+
+  Nothing else moved. Each element's own simulated buy, the deterministic
+  roll-up and every engine number are byte-identical, and so is every
+  non-empirical marginal in the risk model, which is now pinned by a frozen
+  seeded simulation in `tests/test_monte_carlo.py`.
+
+- `tests/goldens/wbs_TEST_PROGRAM.json.gz` and
+  `tests/goldens/wbs_FULL_WBS.json.gz` are rebaselined in their
+  `program_level_percentiles` block, and only there, to the numbers above.
+  They are still compared leaf by leaf rather than excluded, so the new values
+  are pinned as tightly as the old ones were.
+  `COMPARE_POLICY.exclude_after_step3` records what was rebaselined, when, why
+  and by how much, and `tests/goldens/README.md` carries the same account.
+
+- The command line moved from a top-level `cli` package to `cost_core.cli`.
+  A wheel that installs a package called `cli` into site-packages shares that
+  name with every other distribution that does, and installing or removing
+  either can break the other. The `ce-core` command is unchanged, and
+  `python -m cost_core.cli` replaces `python -m cli.ce_core_cli`.
+
+### Removed
+
+- `cost_core.gui` and the `ce-core gui` subcommand. It was a copy of the
+  desktop tool's window taken on 21 August, already older than the tool by the
+  time it landed, and two windows over one engine is one too many. The window
+  to keep is [lot-cost-model](https://github.com/MichaelFowler1/lot-cost-model).
+  **Anyone importing `cost_core.gui` or
+  running `ce-core gui` has to move to that repository.**
+
+- `cost_core.lotmodel.mathx.ols_fit` and `.solve_model`, the lot engine's
+  private least squares and its midpoint iteration, replaced by
+  `cost_core.lotmodel.models` on top of `cost_core.fitting`. `mathx` keeps the
+  lot midpoint, the unit tracking and the two column helpers, which are
+  geometry and parsing. The private MUPE and ZMPE in
+  `cost_core.lotmodel.enrich` are gone the same way.
+
+- `cost_core.program.SPEC_TOLERANCE` and `cost_core.program._lognormal_spec`,
+  along with the test that bounded the fit against the element it summarised.
+  There is no longer an approximation there to bound.
+
+[Unreleased]: https://github.com/MichaelFowler1/cost-risk-toolkit/compare/v1.0.1...HEAD
+[1.0.1]: https://github.com/MichaelFowler1/cost-risk-toolkit/compare/v1.0.0...v1.0.1
+[1.0.0]: https://github.com/MichaelFowler1/cost-risk-toolkit/releases/tag/v1.0.0

cost_core-1.0.1/CITATION.cff

@@ -0,0 +1,31 @@
+cff-version: 1.2.0
+message: "If you use this software, please cite it as below."
+type: software
+title: "cost-risk-toolkit"
+abstract: >-
+  A Python library and CLI for defense cost estimating. Generates synthetic
+  CSDR/SRDR contractor cost data, normalises it through an ingest pipeline with
+  a persisted WBS crosswalk and full row-level provenance, fits Wright and
+  Crawford learning curves and parametric cost estimating relationships by OLS,
+  MUPE and ZMPE, and runs correlated Monte Carlo risk simulation with discrete
+  risks, tornado analysis and P50/P80/P90 thresholds. Every run emits an
+  assumptions and provenance log organised around the four characteristics of a
+  reliable estimate in the GAO Cost Estimating and Assessment Guide. All data
+  used is synthetic; no real or proprietary contractor data is included.
+authors:
+  - family-names: "Fowler"
+    given-names: "Michael"
+    alias: "MichaelFowler1"
+repository-code: "https://github.com/MichaelFowler1/cost-risk-toolkit"
+url: "https://github.com/MichaelFowler1/cost-risk-toolkit"
+keywords:
+  - cost-estimating
+  - monte-carlo
+  - learning-curve
+  - risk-analysis
+  - csdr
+license: "Apache-2.0"
+version: "1.0.1"
+# CFF pairs `version` with `date-released`: together they describe the release
+# being cited, not the first commit. Each release moves both.
+date-released: "2026-09-11"

cost_core-1.0.1/LICENSE

@@ -0,0 +1,202 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Michael Fowler
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+

cost_core-1.0.1/MANIFEST.in

@@ -0,0 +1,33 @@
+# What goes into the sdist, beyond the files setuptools adds on its own.
+#
+# Without this file setuptools' default sweep picks up `tests/test_*.py` and
+# nothing those modules import: no conftest.py, no golden_support.py, and none
+# of the recorded goldens. An sdist like that unpacks into a suite that collects
+# and then errors on `import golden_support`, which is worse than shipping no
+# tests at all. So graft the whole directory: the suite in the sdist is then the
+# suite in the repo, goldens included, and `pytest tests/ -q` works from an
+# unpacked source release.
+graft tests
+
+# example_lots.csv is test input, not decoration. tests/golden_support.py reads
+# it from the repository root for the `example_lots_csv` and `cli_fit_lots`
+# golden cases, so a grafted tests/ without it is still a suite that cannot run.
+include example_lots.csv
+
+# A source release should carry the history and the citation metadata, not only
+# the code. README.md and LICENSE arrive on their own through `readme` and the
+# license field; requirements.txt is the exact stack the goldens were recorded
+# against, which is worth having beside them.
+include CHANGELOG.md
+include CITATION.cff
+include requirements.txt
+
+# The README tells the reader to regenerate the hero figure with this script, so
+# a source release that omits it is a source release the README lies about. The
+# figure itself is not here: the README now points at it by absolute URL, which
+# is what makes it render on PyPI, where nothing from the sdist is served.
+include make_hero.py
+
+# Build leftovers are not source.
+global-exclude *.py[cod]
+prune tests/__pycache__