calm-benchmark 0.1.3__tar.gz

calm_benchmark-0.1.3/.github/workflows/ci.yml

@@ -0,0 +1,18 @@
+name: CI
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: astral-sh/setup-uv@v7
+      - run: uv sync --all-groups
+      - run: uv run ruff check
+      - run: uv run ruff format --check
+      - run: uv run mypy
+      - run: uv run pytest

calm_benchmark-0.1.3/.github/workflows/publish-pypi.yml

@@ -0,0 +1,45 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  contents: read
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - name: Check out source
+        uses: actions/checkout@v7
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b # v8.1.0
+        with:
+          enable-cache: true
+
+      - name: Install Python
+        run: uv python install 3.14
+
+      - name: Verify release version
+        run: test "v$(uv version --short)" = "$GITHUB_REF_NAME"
+
+      - name: Run release gate
+        run: |
+          uv sync --locked
+          uv run ruff check
+          uv run ruff format --check
+          uv run mypy
+          uv run pytest --cov=calm
+
+      - name: Build distributions
+        run: uv build
+
+      - name: Publish distributions
+        run: uv publish

calm_benchmark-0.1.3/.gitignore

@@ -0,0 +1,19 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv
+.env
+
+# Test and documentation output
+.coverage
+coverage.xml
+htmlcov/
+site/
+runs/
+data/

calm_benchmark-0.1.3/.hypothesis/.gitignore

@@ -0,0 +1,9 @@
+# This .gitignore file was automatically created by Hypothesis. Hypothesis gitignores
+# .hypothesis by default, because we generally recommend that .hypothesis not be checked
+# into version control.
+#
+# If you *would* like to check .hypothesis into version control, you should delete this
+# file. Hypothesis will not re-create this .gitignore unless .hypothesis is deleted (and
+# if it does, that's a bug - please report it!)
+
+*

calm_benchmark-0.1.3/.hypothesis/constants/0c4ab3f4807b1659

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/cohort.py
+# hypothesis_version: 6.156.7
+
+['both', 'not_female', 'outside_age_policy']

calm_benchmark-0.1.3/.hypothesis/constants/0fc300e77578a183

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/orchestration/benchmark.py
+# hypothesis_version: 6.156.7
+
+['0.1.3', '2017-2018', 'AGE', 'RHD043', 'RHQ031', 'RIAGENDR', 'RIDAGEYR', 'SEQN', 'SEX', 'acquire', 'attribute', 'calibrate', 'calm', 'cohort', 'demographics', 'evaluate', 'features', 'full_range', 'ingest', 'labels', 'questionnaire', 'real-data', 'report', 'split', 'synthetic', 'synthetic://', 'train', 'unavailable']

calm_benchmark-0.1.3/.hypothesis/constants/12d99eb2fb0e31cc

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/services/explainer/calm_explainer/contracts.py
+# hypothesis_version: 6.156.7
+
+['1', 'explained', 'refused']

calm_benchmark-0.1.3/.hypothesis/constants/14075edae418ffe6

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/evaluate.py
+# hypothesis_version: 6.156.7
+
+[0.5, 0.95, 200, 'age_', 'auroc', 'balanced_accuracy', 'brier', 'f1', 'nan', 'no_call_rate', 'pr_auc', 'recall_premenopausal']

calm_benchmark-0.1.3/.hypothesis/constants/15208dff27bebba0

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/orchestration/manifest.py
+# hypothesis_version: 6.156.7
+
+['manifest.json']

calm_benchmark-0.1.3/.hypothesis/constants/1c43ea5aad69db0f

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/report.py
+# hypothesis_version: 6.156.7
+
+[', ', '- None', '- Not estimated', 'None', 'attributions.json', 'benchmark_card.md', 'full_range', 'metrics', 'metrics.json', 'reliability.svg', 'subgroups', 'value']

calm_benchmark-0.1.3/.hypothesis/constants/1de1c580569e9cea

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/domain/errors.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/2170d90bc85b3eb8

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/config/__init__.py
+# hypothesis_version: 6.156.7
+
+['RunConfig', 'load_config']

calm_benchmark-0.1.3/.hypothesis/constants/24be4ab7866beb98

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/observability/logging.py
+# hypothesis_version: 6.156.7
+
+['%(message)s', 'iso']

calm_benchmark-0.1.3/.hypothesis/constants/2a676ced37d4cce1

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/config/models.py
+# hypothesis_version: 6.156.7
+
+[0.95, 100, 120, 200, '1', '^\\d{4}-\\d{4}$', 'after', 'demographics', 'examination', 'exclude', 'forbid', 'full_range', 'gradient_boosting', 'isotonic', 'laboratory', 'logistic_regression', 'natural_menopause', 'premenopausal', 'preselected', 'questionnaire', 'restricted', 'sigmoid']

calm_benchmark-0.1.3/.hypothesis/constants/31dd52fcb4e2a2b3

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/domain/leakage.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/39073abd55cb81c8

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/ports/data_source.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/3ad8f61b62a82266

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/adapters/__init__.py
+# hypothesis_version: 6.156.7
+
+['NhanesSource', 'SyntheticSource']

calm_benchmark-0.1.3/.hypothesis/constants/42cb07f95690e4b3

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/services/explainer/calm_explainer/service.py
+# hypothesis_version: 6.156.7
+
+['causal', 'causes', 'certain', 'diagnos', 'doctor', 'explained', 'medication', 'refused', 'therapy', 'treat']

calm_benchmark-0.1.3/.hypothesis/constants/4446fa7afd69f802

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/calibrate.py
+# hypothesis_version: 6.156.7
+
+[0.5]

calm_benchmark-0.1.3/.hypothesis/constants/570fa43e156bc904

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/split.py
+# hypothesis_version: 6.156.7
+
+['labels', 'seed', 'test_fraction', 'test_ids', 'training_ids', 'validation_fraction', 'validation_ids']

calm_benchmark-0.1.3/.hypothesis/constants/5c3118ac0156bb09

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/adapters/codebook.py
+# hypothesis_version: 6.156.7
+
+[' - ', 'h3', 'html.parser', 'table', 'td', 'tr']

calm_benchmark-0.1.3/.hypothesis/constants/6166398d30c22845

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/domain/entities.py
+# hypothesis_version: 6.156.7
+
+['-', 'unversioned']

calm_benchmark-0.1.3/.hypothesis/constants/71e00ec495499f44

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/adapters/nhanes.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/736d6fe7be657c15

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/orchestration/__init__.py
+# hypothesis_version: 6.156.7
+
+['RunManifest', 'StageRecord', 'StageRunner']

calm_benchmark-0.1.3/.hypothesis/constants/743c6e1dfb631705

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/services/__init__.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/76508aa4a4175b66

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/ingest.py
+# hypothesis_version: 6.156.7
+
+['.xpt', 'SEQN', 'polars']

calm_benchmark-0.1.3/.hypothesis/constants/7662d704b8d35622

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/__init__.py
+# hypothesis_version: 6.156.7
+
+['0.1.3', 'CalibratedModel', 'EvaluationResult', 'FeatureAttribution', 'FrozenSplit', 'LabelResult', 'PredictionResult', 'ReportArtifacts', 'ReportWriter', 'RunConfig', 'Subgroup', 'TrainedModel', 'app', 'attribute', 'build_labels', 'calibrate', 'evaluate', 'get_label_rule', 'load_config', 'predict', 'select_features', 'stratified_split', 'train']

calm_benchmark-0.1.3/.hypothesis/constants/76bc977e7eb5cd52

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/cli.py
+# hypothesis_version: 6.156.7
+
+[0.2, 0.45, 0.55, 200, '--config', '.cache', '.raw', '2017-2018', 'calm', 'exclude', 'logistic_regression', 'marker_one', 'marker_two', 'max_iter', 'nhanes-2017-2018', 'public_cessation', 'restricted', 'run-nhanes', 'run-synthetic', 'runs', 'sigmoid', 'synthetic-demo', 'synthetic-run']

calm_benchmark-0.1.3/.hypothesis/constants/7afaf1638e6eb750

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/__init__.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/7c805b6af0508d82

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/observability/__init__.py
+# hypothesis_version: 6.156.7
+
+['ContentHash', 'Provenance', 'configure_logging', 'hash_bytes', 'hash_config', 'hash_json']

calm_benchmark-0.1.3/.hypothesis/constants/803886b0ec08beb9

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/attribution.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/8a6e0aa798244d9c

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/config/loading.py
+# hypothesis_version: 6.156.7
+
+['rb']

calm_benchmark-0.1.3/.hypothesis/constants/8fba44fbd7419531

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/observability/provenance.py
+# hypothesis_version: 6.156.7
+
+['json']

calm_benchmark-0.1.3/.hypothesis/constants/92f6bd8fa9a6a6a0

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/labels.py
+# hypothesis_version: 6.156.7
+
+['1', '2', '3', '7', 'RHD043', 'RHD143', 'RHD280', 'RHQ031', 'RHQ200', 'RHQ305', 'RHQ540', 'SEQN', 'exclude', 'exogenous_hormones', 'hysterectomy', 'lactation', 'natural-menopause-1', 'natural_menopause', 'no', 'perimenopause', 'pregnancy', 'premenopausal', 'public_cessation', 'surgical_menopause', 'yes']

calm_benchmark-0.1.3/.hypothesis/constants/ac61764b214cded0

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/train.py
+# hypothesis_version: 6.156.7
+
+[0.0, 'estimator', 'gradient_boosting', 'imputer', 'logistic_regression', 'median', 'random_state']

calm_benchmark-0.1.3/.hypothesis/constants/b7273513001c7382

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/adapters/synthetic.py
+# hypothesis_version: 6.156.7
+
+[2026, 'AGE', 'Age', 'Female', 'Hysterectomy', 'Male', 'No', 'RHD043', 'RHD143', 'RHD280', 'RHQ031', 'RHQ200', 'RHQ305', 'RHQ540', 'SEQN', 'SEX', 'STATUS', 'Sex', 'Synthetic marker one', 'Synthetic marker two', 'Yes', 'demographics', 'demographics.sav', 'laboratory', 'laboratory.sav', 'marker_one', 'marker_two', 'questionnaire', 'questionnaire.sav']

calm_benchmark-0.1.3/.hypothesis/constants/c3d801b66e25d28e

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/adapters/nhanes_2017_2018.py
+# hypothesis_version: 6.156.7
+
+['1', '2', '2017-2018', '3', '7', 'BMX_J.XPT', 'BPX_J.XPT', 'Breast feeding', 'CBC_J.XPT', 'DEMO_J.XPT', 'GHB_J.XPT', 'HDL_J.XPT', 'Hysterectomy', 'No', 'Pregnancy', 'RHD043', 'RHQ031', 'RHQ_J.XPT', 'RIAGENDR', 'RIDAGEYR', 'SEQN', 'TCHOL_J.XPT', 'Yes', 'blood_pressure', 'body_measures', 'complete_blood_count', 'demographics', 'glycated_hemoglobin', 'hdl', 'questionnaire', 'total_cholesterol']

calm_benchmark-0.1.3/.hypothesis/constants/cb383feb990bc16b

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/features.py
+# hypothesis_version: 6.156.7
+
+['SEQN', 'left']

calm_benchmark-0.1.3/.hypothesis/constants/cc255fd23cd54b1b

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/services/explainer/__init__.py
+# hypothesis_version: 6.156.7
+
+[]

calm_benchmark-0.1.3/.hypothesis/constants/d1e39bf8bc23207e

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/observability/hashing.py
+# hypothesis_version: 6.156.7
+
+[',', ':', 'utf-8']

calm_benchmark-0.1.3/.hypothesis/constants/d920e5be5304d03b

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/ports/__init__.py
+# hypothesis_version: 6.156.7
+
+['DataSource']

calm_benchmark-0.1.3/.hypothesis/constants/e00e778e1fc5a182

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/orchestration/runner.py
+# hypothesis_version: 6.156.7
+
+['Result', 'stage_completed', 'stage_started']

calm_benchmark-0.1.3/.hypothesis/constants/e56700b7fce7099f

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/.venv/bin/pytest
+# hypothesis_version: 6.156.7
+
+['-script.pyw', '.exe', '__main__']

calm_benchmark-0.1.3/.hypothesis/constants/f573a9d979b3995e

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/pipeline/verification.py
+# hypothesis_version: 6.156.7
+
+[0.95, '40-44', '45-55', '56-60', 'RHD043', 'RHD280', 'RHQ305', 'RHQ540', 'RIDAGEYR', 'public_cessation']

calm_benchmark-0.1.3/.hypothesis/constants/fa19d69dfd8d556c

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/src/calm/domain/__init__.py
+# hypothesis_version: 6.156.7
+
+['Cohort', 'LabelSet', 'VariableCode']

calm_benchmark-0.1.3/.hypothesis/constants/fa99e419bd9005fc

@@ -0,0 +1,4 @@
+# file: /home/runner/work/calm/calm/services/explainer/calm_explainer/__init__.py
+# hypothesis_version: 6.156.7
+
+['AttributionInput', 'CalmHandoff', 'ExplanationResponse', 'explain']

calm_benchmark-0.1.3/.pre-commit-config.yaml

@@ -0,0 +1,12 @@
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.15.22
+    hooks:
+      - id: ruff-check
+        args: [--fix]
+      - id: ruff-format
+  - repo: https://github.com/pre-commit/mirrors-mypy
+    rev: v1.19.1
+    hooks:
+      - id: mypy
+        additional_dependencies: [pydantic, typer]

calm_benchmark-0.1.3/.python-version

@@ -0,0 +1 @@
+3.14

calm_benchmark-0.1.3/AGENT.md

@@ -0,0 +1,155 @@
+# CALM
+
+This document governs how the autonomous engineering agent builds **CALM**. It is binding. When this
+document and any instinct to move faster conflict, this document wins.
+
+---
+
+## 1. Mission
+
+Build **CALM** (Calibrated Menopausal-status benchmark): an open, reproducible, rigorously tested
+benchmark that predicts premenopausal vs. postmenopausal status from routine, low-cost markers derived
+from NHANES. CALM is a **library and CLI** application. It produces versioned, provenance-tracked
+artifacts that downstream systems consume.
+
+Read `DESIGN.md` for the architecture and `TODOS.md` for the phased plan before writing code.
+
+---
+
+## 2. Prime directives (non-negotiable)
+
+1. **Reproducibility over speed.** Any result must be regenerable from config plus pinned inputs. A change
+   that breaks reproducibility is a defect, not a trade-off.
+2. **Determinism.** All randomness flows from a single injected RNG seeded by config. No wall-clock and no
+   ambient state in domain logic.
+3. **No fabrication.** Never invent data, variable codes, response codings, or metric values.
+   Unknown is not a licence to guess.
+4. **Verify domain facts from source metadata, not memory.** NHANES variable codes and value codings are
+   read from ingested `.XPT` value-label metadata and asserted. If a required code or coding cannot be
+   verified from metadata, **stop and escalate** (§10). Do not assume `1 = yes`.
+5. **The leakage boundary is sacred.** Variables used to construct the label must never reach the feature
+   layer. Enforce with a runtime invariant and a property-based test.
+6. **Tests gate every change.** No implementation is done without unit tests. Pipeline invariants get
+   property-based tests. Metrics get golden tests.
+7. **No unnecessary comments.** Types and names document structure; Google-style docstrings document the
+   public API. Inline comments exist only to justify a non-obvious *why*, and rarely.
+8. **Small, verifiable increments.** One task at a time, each ending green. Never advance a phase with a
+   red gate.
+9. **Every step is recorded.** Each pipeline stage is instrumented (timing, input/output content hashes,
+   emitted metrics), and each run writes a durable manifest.
+
+---
+
+## 3. Operating loop (repeat per task)
+
+1. Open `TODOS.md`; select the next unblocked task in the lowest incomplete phase.
+2. Write the failing test(s) that define the task's success.
+3. Implement the **minimal** typed code to pass them, respecting layer boundaries (§6).
+4. Run the gate:
+   `uv run ruff check && uv run ruff format --check && uv run mypy && uv run pytest --cov=calm`.
+   All must pass; coverage must not regress.
+5. Add a `CHANGELOG.md` entry under `Unreleased`.
+6. Emit a checkpoint report (§5) and check the task box.
+7. Stop. Do not chain tasks silently.
+
+---
+
+## 4. Definition of done
+
+- **Global:** gates green (lint, format, types, tests, coverage ≥ threshold); docstrings on public
+  symbols; changelog updated; no undocumented TODO/FIXME left in the tree.
+- **Per stage:** typed input/output contract; unit tests; relevant invariants tested; provenance emitted;
+  instrumented by the orchestrator.
+- **Per phase:** all phase tasks checked **and** the phase acceptance tests pass.
+
+---
+
+## 5. Monitoring & reporting
+
+After every task, emit exactly:
+
+```
+[phase X · task Y] <what changed>
+  tests: +N (unit/property/golden) | gates: ruff ✓ mypy ✓ pytest ✓ | cov: A% (Δ)
+  artifacts / manifests touched: <...>
+  next: <task>   blockers: <none | ...>
+```
+
+Never report a phase complete unless its acceptance criteria in `TODOS.md` pass. Surface coverage or type
+regressions immediately instead of proceeding.
+
+---
+
+## 6. Coding standards
+
+- Target **Python 3.14**, managed with **uv**. Source under `src/calm/`.
+- **Full type hints**; no `Any` in public signatures. Prefer `Protocol` for ports and
+  `@dataclass(frozen=True)` or pydantic for data.
+- **Layer purity:** `calm/domain/` imports no I/O, no framework, and nothing from `calm/adapters`.
+  Dependencies point inward (domain ← pipeline ← orchestration ← adapters / CLI).
+- **Docstrings:** Google style, on public modules, classes, and functions only. Enforced via Ruff `D`
+  rules.
+- **Config** is pydantic; never read environment or files inside domain logic.
+- **Errors** are typed exceptions with actionable messages. Domain uncertainty raises a dedicated
+  `DomainVerificationError` that halts the run.
+
+---
+
+## 7. Determinism rules
+
+- One RNG (numpy `Generator`) constructed from `config.seed`, injected where needed.
+- Artifacts are content-addressed: the hash of (normalized config + input hashes) identifies an output.
+- Splits, model initialization, and any sampling use only the injected RNG.
+
+---
+
+## 8. Domain guardrails
+
+- **Label construction** from the Reproductive Health questionnaire: read the value-label metadata and
+  assert the response coding before applying it. Apply and log exclusions (surgical menopause, pregnancy
+  and lactation, exogenous hormones) with counts.
+- **Age confounding:** honor the configured policy — restricted transition window, or full range with
+  mandatory age-band reporting. Never headline a full-range AUROC without the age-band breakdown.
+- **Survey weights:** CALM models at the individual level; weights are not applied to metrics unless
+  explicitly configured, and the choice is recorded in the artifacts.
+- **No clinical claims** anywhere in code, docstrings, or artifacts. CALM predicts and evaluates; it never
+  diagnoses.
+
+---
+
+## 9. Tooling
+
+```bash
+uv sync                          # install
+uv run calm-benchmark --help     # CLI
+uv run ruff check                # lint
+uv run ruff format --check       # format
+uv run mypy                      # types
+uv run pytest --cov=calm         # tests + coverage
+uv run pre-commit run -a         # all hooks
+uv run mkdocs serve              # docs preview
+```
+
+---
+
+## 10. Escalation / stop conditions
+
+Halt and surface a concise decision request when:
+
+- a required NHANES code or response coding cannot be verified from ingested metadata;
+- the specification is ambiguous in a way that changes results (for example, an undefined perimenopause
+  policy for the data at hand);
+- a gate cannot pass without weakening a guarantee (reproducibility, leakage boundary, determinism);
+- proceeding would require inventing data or numbers.
+
+Escalation is success, not failure. Guessing on any of the above is a defect.
+
+---
+
+## 11. Prohibited
+
+- Committing red gates or skipping tests to "unblock."
+- Placing I/O in the domain layer or bypassing ports.
+- Hardcoding variable codes without metadata verification.
+- Emitting a metric computed outside the evaluation stage.
+- Adding UI, serving, or LLM features to CALM core.

calm_benchmark-0.1.3/CHANGELOG.md

@@ -0,0 +1,75 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+### Changed
+
+- Aligned package metadata, runtime provenance, citation, and release artifacts to version 0.1.3.
+
+- Renamed the PyPI distribution to `calm-benchmark` to avoid collision with the unrelated `calm`
+  package, renamed the CLI to `calm-benchmark`, and added tag-driven uv Trusted Publishing automation.
+
+- Completed Phase 10: added provenance-bound feature-association exports and an isolated OpenAI
+  Structured-Outputs explanation service with immutable hand-offs and adversarial safety refusals.
+
+- Completed Phase 9 methodological validation: withheld out-of-range predictions; deterministic bootstrap
+  confidence intervals; subgroup denominators and eligibility; enforced full-range age-band context;
+  calibration feasibility and preselected-method recording; and run-derived benchmark-card disclosures.
+
+- Require a configured minimum number of validation examples per outcome class for calibration and retain fit sample
+  counts on calibrated model artifacts.
+
+- Added source-derived age-band subgroups to every benchmark evaluation, with a full-range guard against
+  reporting an aggregate without age context.
+
+- Made the no-call execution boundary explicit: out-of-range or low-confidence scores have no
+  actionable class label, and documented the range-checking limitation.
+
+- Added the pinned public NHANES 2017–2018 CLI run, checksum-verified local input adapter, versioned
+  RHQ metadata contract, and real `SEQN`-joined multi-component feature assembly.
+
+- Added an architecture and delivery audit to the roadmap, distinguishing implemented unit-level
+  safeguards from the remaining end-to-end benchmark, methodological, attribution, and separate LLM
+  communication-layer work.
+- Replaced the hand-assembled CLI demo with a config-driven synthetic execution of every benchmark stage,
+  including a required complete run manifest.
+- Added a validated TOML-driven synthetic CLI command and a default run configuration.
+- Made the range-based out-of-distribution safeguard a mandatory no-call condition.
+- Corrected the 2017–2018 NHANES access note: `RHQ031` and `RHD043` are public-use fields; only the
+  clean surgical-menopause exclusion for women aged 45 and older requires the optional RDC variant.
+- Added a separately versioned public-use cessation label rule that avoids claiming unavailable natural-
+  menopause exclusions.
+- Reconciled README and architecture claims with shipped artifacts, marking NHANES production runs,
+  stored model artifacts, generated model cards, and SHAP attribution as planned.
+- Added branch-tested public NHANES age-band coverage verification for label-variant selection.
+
+## [0.1.3] - 2026-07-19
+
+### Added
+
+- Initial project scaffold and development tooling.
+- Immutable Pydantic run-configuration contracts and TOML loading.
+- Frozen domain value objects and entities with alignment and shape invariants.
+- Deterministic content hashing, normalized configuration hashes, and immutable provenance records.
+- Structured JSON logging, injected-clock stage instrumentation, and durable run-manifest writing.
+- Phase 1 acceptance tests for configuration failures, hash stability, and no-op manifest shape.
+- Configurable NHANES raw-file acquisition port with byte checksums and source lineage.
+- Deterministic offline synthetic source and pyreadstat ingestion with recoverable value-label metadata.
+- Configurable cohort selection with participant drop-count accounting.
+- Registered natural-menopause label construction with metadata-verified response codings,
+  auditable exclusion and class-balance counts, configurable perimenopause treatment, and an optional
+  early-natural-menopause secondary outcome.
+- Leakage-safe allowlist feature selection and deterministic, content-addressed stratified split
+  artifacts, including property-tested partition invariants.
+- Registered logistic-regression and gradient-boosting training with injected RNGs, validation-only
+  sigmoid or isotonic calibration, abstention bands, and training-range distribution-shift flags.
+- Frozen-test evaluation with a metric registry, reliability data, abstention and distribution-shift
+  accounting, subgroup reports, and deterministic metrics, SVG reliability plot, benchmark-card, and
+  manifest exports.
+- Frozen public API, offline CLI demonstration, generated API documentation, extension guidance, and
+  Citation File Format metadata. A DOI has not been assigned.

calm_benchmark-0.1.3/CITATION.cff

@@ -0,0 +1,12 @@
+cff-version: 1.2.0
+message: "If you use CALM, please cite it using this metadata."
+title: "CALM: Calibrated Menopausal-status Benchmark"
+version: 0.1.3
+date-released: 2026-07-19
+license: MIT
+abstract: "A reproducible research benchmark for menopausal-status prediction from routine health markers."
+keywords:
+  - reproducibility
+  - NHANES
+  - menopausal status
+  - benchmark