v1.1.0

Fixed

• Inverse problem (critical): the forward model _T_clad_from_S was
  independent of S_mag because the holiday source magnitude was hardcoded
  in run_coupled (S[ins_sl.stop-1] = 1.0e-6) and S_mag was only used
  as a boolean (holiday=(S_mag > 0)). Golden-section search converged to
  the upper search bound, producing 150–1400% recovery error in
  run_all.py [6/9]. Fixed by parameterising the outer Dirichlet BC
  through _holiday_theta_outer(S_mag, S_ref):
  theta_outer = f * THETA_SAT + (1 - f) * THETA_INIT, f = min(S_mag/S_ref, 1).
  Recovery error is now < 1% on the identifiable range S ∈ [0, S_ref]
  with 0.03 K DFOS noise.
• Magic number in coupled_solver._Q_array: corrosion heat-flux to
  volumetric source conversion used a hardcoded 0.002 m. Replaced with
  the mesh-derived first-cell thickness mesh["r"][1] - mesh["r"][0].
• Misleading docstring in inverse_problem.py: removed the claim
  about an "adjoint gradient formulation" — only golden-section is
  implemented. The docstring now describes the actual single-parameter
  1D inverse and its identifiability range.
• Duplicate inline comments in constants.py: cleaned D_THETA0,
  BETA_THETA, ETA_MIXED annotations.

Changed

• run_coupled accepts S_mag and S_ref keyword arguments. Defaults
  (S_mag=S_ref=1e-6) preserve legacy baseline behaviour exactly
  (wl(10yr) = 0.1360 mm unchanged).
• moisture_step and apply_holiday_bc accept theta_outer=THETA_SAT
  as a keyword argument. Default preserves prior behaviour.
• run_all.py [6/9] inverse search range capped at S_hi = S_ref = 1e-6
  with synthetic test cases {2e-7, 5e-7, 8e-7} — all in the identifiable
  range. Previous range [0, 3e-6] produced spurious "recovery" inside
  the saturation plateau.

Added

• LICENSE file (RES-EDU-1.0) formalising the Research and Educational
  Use clause that previously appeared only as prose in README.md.
• CHANGELOG.md (this file).
• CONTRIBUTING.md with development setup and contribution guidelines.
• pyproject.toml with project metadata, ruff and pytest configuration.
• .gitignore covering Python build artifacts, generated assets, IDE
  files, and __pycache__.
• .github/workflows/ci.yml — GitHub Actions CI matrix
  (Python 3.11/3.12/3.13 × Ubuntu/macOS/Windows): ruff lint, pytest with
  coverage, validation benchmarks, and a run_all.py smoke test that
  uploads the generated figures and audit chain as artefacts.
• .github/workflows/release.yml — tag-triggered release workflow that
  attaches a sanitised source archive and the most recent CI artefacts.
• .github/dependabot.yml for weekly dependency and Actions updates.
• .github/ISSUE_TEMPLATE/ (bug, feature, physics-question) and
  .github/PULL_REQUEST_TEMPLATE.md.
• .pre-commit-config.yaml running ruff (lint + format) and basic
  hygiene hooks before every commit.
• Inverse-recovery regression test in tests/test_remaining.py
  asserting |S_opt - S_true| / S_true < 0.05 for at least one
  identifiable test case — locks the fix as a regression guard.

Documentation

• README.md updated to reflect the partial-holiday Dirichlet model in
  Eq. (2c) and the S ∈ [0, S_ref] identifiability constraint on the
  inverse problem in Eq. (6).