calibroute-ai 0.3.0__tar.gz

calibroute_ai-0.3.0/.github/workflows/ci.yml

@@ -0,0 +1,24 @@
+name: ci
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - run: python -m pip install -e ".[dev]"
+      - run: ruff check src tests examples/financial_ner/run_case_study.py
+      - run: ruff format --check src tests examples/financial_ner/run_case_study.py
+      - run: python -m unittest discover -s tests -v
+      - run: python examples/financial_ner/run_case_study.py
+      - run: python -m build
+      - run: python -m twine check dist/*

calibroute_ai-0.3.0/.github/workflows/publish.yml

@@ -0,0 +1,47 @@
+name: Publish to PyPI
+
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: pypi-publication
+  cancel-in-progress: false
+
+jobs:
+  build:
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: python -m pip install -e ".[dev]"
+      - run: python -m unittest discover -s tests -v
+      - run: ruff check src tests examples/financial_ner/run_case_study.py
+      - run: python -m build
+      - run: python -m twine check dist/*
+      - uses: actions/upload-artifact@v4
+        with:
+          name: pypi-distributions
+          path: dist/
+          if-no-files-found: error
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/project/calibroute-ai/
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: pypi-distributions
+          path: dist/
+      - name: Publish verified distributions using Trusted Publishing
+        uses: pypa/gh-action-pypi-publish@release/v1

calibroute_ai-0.3.0/.gitignore

@@ -0,0 +1,13 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+.pytest_cache/
+.ruff_cache/
+.venv/
+venv/
+dist/
+build/
+coverage.xml
+.coverage
+examples/output/
+workcheck/

calibroute_ai-0.3.0/CHANGELOG.md

@@ -0,0 +1,29 @@
+# Changelog
+
+## 0.3.0 - 2026-09-20
+
+- Fix tie groups in threshold fitting, risk-coverage curves, and selective reports.
+- Add independent holdout validation with CLI pass/fail exit codes.
+- Clarify pointwise fit bounds versus independent fixed-policy risk assessment.
+- Replace quadratic AUROC with tie-aware O(n log n) rank aggregation.
+- Make missing shift threshold default to adaptive monitoring; validate inputs.
+- Regenerate case-study reports and record provenance and reproducibility limits.
+- Align package/citation versions; expand CI to lint, build, and check distributions.
+
+Refit 0.2 policies before deployment. The 0.2 policy selection and audit reports
+could split ties and depend on input order. No 0.2 PyPI release is intended.
+
+## 0.2.0 - 2026-09-19
+
+- Add exact Clopper-Pearson upper risk bounds to policy fitting.
+- Add batch-size-aware JSD calibration and explicit small-batch handling.
+- Add a Financial NER sentence-output adapter and multi-domain case study.
+- Expand policy schema to record declared risk and calibration settings.
+
+## 0.1.0 - 2026-09-19
+
+- Initial model-agnostic confidence audit.
+- Validation-grounded selective policy fitting.
+- Batch confidence-shift detection.
+- Auditable accept, human-review, and abstain routing.
+- Dependency-free Python API and CLI.

calibroute_ai-0.3.0/CITATION.cff

@@ -0,0 +1,11 @@
+cff-version: 1.2.0
+message: "If CalibRoute supports your research, please cite the software release."
+title: "CalibRoute: Uncertainty-Aware Evaluation and Decision Control for AI Systems"
+type: software
+version: 0.3.0
+date-released: 2026-09-20
+repository-code: "https://github.com/Garyouki/calibroute"
+authors:
+  - family-names: "Zheng"
+    given-names: "Zihao"
+license: MIT

calibroute_ai-0.3.0/CONTRIBUTING.md

@@ -0,0 +1,23 @@
+# Contributing
+
+Contributions are welcome, especially new metrics, shift-monitor adapters,
+task-specific examples, and tests that expose unsafe assumptions.
+
+## Local workflow
+
+1. Create a focused branch.
+2. Install the package with `python -m pip install -e .`.
+3. Run `python -m unittest discover -s tests -v`.
+4. Add tests for every behavior change.
+5. Document whether a new metric is descriptive, empirically calibrated, or
+   accompanied by a formal guarantee.
+
+Please do not commit private, proprietary, personally identifiable, or
+license-restricted datasets. Prefer download scripts, hashes, and synthetic
+fixtures for examples.
+
+## Pull requests
+
+Keep pull requests narrow. Explain the uncertainty assumption, expected failure
+mode, and how the change affects accept/review/abstain decisions.
+

calibroute_ai-0.3.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Zihao Zheng and CalibRoute contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

calibroute_ai-0.3.0/PKG-INFO

@@ -0,0 +1,168 @@
+Metadata-Version: 2.5
+Name: calibroute-ai
+Version: 0.3.0
+Summary: Uncertainty-aware evaluation and decision control for AI systems.
+Project-URL: Repository, https://github.com/Garyouki/calibroute
+Project-URL: Issues, https://github.com/Garyouki/calibroute/issues
+Project-URL: Documentation, https://github.com/Garyouki/calibroute/blob/main/README.md
+Author: Zihao Zheng
+License: MIT License
+        
+        Copyright (c) 2026 Zihao Zheng and CalibRoute contributors
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: abstention,ai-evaluation,distribution-shift,selective-prediction,trustworthy-ai,uncertainty
+Classifier: Development Status :: 3 - Alpha
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Requires-Dist: ruff>=0.8; extra == 'dev'
+Requires-Dist: twine>=6; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# CalibRoute
+
+CalibRoute is a lightweight Python toolkit for confidence evaluation and
+uncertainty-aware routing. It converts model predictions into three actions:
+`accept`, `human_review`, or `abstain`.
+
+## Features
+
+- Confidence calibration and error-ranking metrics
+- Risk-coverage analysis
+- Validation-only threshold fitting
+- Tie-aware threshold fitting and independent holdout risk assessment
+- Batch-size-aware confidence-shift detection
+- Financial NER encoder and generative-output adapter
+- CSV and JSONL support
+- Dependency-free Python API and CLI
+
+## Install
+
+Requires Python 3.10 or newer.
+
+Install with `python -m pip install calibroute-ai`.
+
+To run the checked-in examples or contribute, clone the repository:
+
+```bash
+git clone https://github.com/Garyouki/calibroute.git
+cd calibroute
+python -m pip install -e .
+```
+
+The distribution name is `calibroute-ai`; the Python import and command are
+`calibroute`. No API key or model service is needed.
+
+## Quick start
+
+```bash
+# Audit labeled predictions
+calibroute audit \
+  --input examples/validation.csv \
+  --output examples/audit.md
+
+# Fit an acceptance policy on validation data
+calibroute fit \
+  --input examples/validation.csv \
+  --max-risk 0.20 \
+  --min-coverage 0.25 \
+  --risk-method empirical \
+  --output examples/policy.json
+
+# Route a new prediction batch
+calibroute route \
+  --input examples/production_batch.csv \
+  --policy examples/policy.json \
+  --output examples/decisions.csv \
+  --summary examples/routing-summary.json
+```
+
+## Input format
+
+| Field | Required | Description |
+|---|---|---|
+| `id` | recommended | Prediction identifier |
+| `confidence` | yes | Number between 0 and 1 |
+| `correct` | `audit` and `fit` only | Boolean or `1`/`0` label |
+| `domain` | no | Evaluation slice or deployment domain |
+
+Additional columns are preserved as metadata.
+
+The default fit searches thresholds using pointwise 95% Clopper-Pearson bounds.
+Threshold selection on the same labels does **not** provide a 95% guarantee for
+the selected policy. Freeze the policy, then assess it on a separate IID holdout:
+
+```bash
+calibroute validate --input holdout.csv --policy examples/policy.json --output holdout-report.json
+```
+
+Exit code 0 means the holdout upper bound meets the declared risk limit; 1 means
+it does not (including zero accepted samples); 2 means invalid input. Never tune
+against this holdout or reuse it to select among multiple policies. A pass does
+not cover distribution shift. See [statistical scope](docs/design.md).
+
+## Python API
+
+```python
+from calibroute import PredictionRecord, fit_policy, route_batch
+
+validation = [
+    PredictionRecord("a", 0.98, True),
+    PredictionRecord("b", 0.82, True),
+    PredictionRecord("c", 0.55, False),
+]
+
+policy = fit_policy(validation, max_risk=0.10, min_coverage=0.50,
+                    risk_method="empirical")  # Tiny illustrative sample only.
+decisions, summary = route_batch(
+    [PredictionRecord("new", 0.74)],
+    policy,
+)
+```
+
+## Limitations
+
+CalibRoute is an evaluation and routing tool, not a safety certification.
+Thresholds should be revalidated after changes to the model, task, prompt, or
+deployment distribution. See [design principles](docs/design.md) for details.
+
+## Development
+
+```bash
+python -m unittest discover -s tests -v
+```
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) and [ROADMAP.md](ROADMAP.md).
+Release preparation is documented in [releasing](docs/releasing.md).
+The [Financial NER case study](examples/financial_ner/README.md) demonstrates
+the adapter and cross-domain failure pattern on 2,098 derived prediction rows.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

calibroute_ai-0.3.0/README.md

@@ -0,0 +1,121 @@
+# CalibRoute
+
+CalibRoute is a lightweight Python toolkit for confidence evaluation and
+uncertainty-aware routing. It converts model predictions into three actions:
+`accept`, `human_review`, or `abstain`.
+
+## Features
+
+- Confidence calibration and error-ranking metrics
+- Risk-coverage analysis
+- Validation-only threshold fitting
+- Tie-aware threshold fitting and independent holdout risk assessment
+- Batch-size-aware confidence-shift detection
+- Financial NER encoder and generative-output adapter
+- CSV and JSONL support
+- Dependency-free Python API and CLI
+
+## Install
+
+Requires Python 3.10 or newer.
+
+Install with `python -m pip install calibroute-ai`.
+
+To run the checked-in examples or contribute, clone the repository:
+
+```bash
+git clone https://github.com/Garyouki/calibroute.git
+cd calibroute
+python -m pip install -e .
+```
+
+The distribution name is `calibroute-ai`; the Python import and command are
+`calibroute`. No API key or model service is needed.
+
+## Quick start
+
+```bash
+# Audit labeled predictions
+calibroute audit \
+  --input examples/validation.csv \
+  --output examples/audit.md
+
+# Fit an acceptance policy on validation data
+calibroute fit \
+  --input examples/validation.csv \
+  --max-risk 0.20 \
+  --min-coverage 0.25 \
+  --risk-method empirical \
+  --output examples/policy.json
+
+# Route a new prediction batch
+calibroute route \
+  --input examples/production_batch.csv \
+  --policy examples/policy.json \
+  --output examples/decisions.csv \
+  --summary examples/routing-summary.json
+```
+
+## Input format
+
+| Field | Required | Description |
+|---|---|---|
+| `id` | recommended | Prediction identifier |
+| `confidence` | yes | Number between 0 and 1 |
+| `correct` | `audit` and `fit` only | Boolean or `1`/`0` label |
+| `domain` | no | Evaluation slice or deployment domain |
+
+Additional columns are preserved as metadata.
+
+The default fit searches thresholds using pointwise 95% Clopper-Pearson bounds.
+Threshold selection on the same labels does **not** provide a 95% guarantee for
+the selected policy. Freeze the policy, then assess it on a separate IID holdout:
+
+```bash
+calibroute validate --input holdout.csv --policy examples/policy.json --output holdout-report.json
+```
+
+Exit code 0 means the holdout upper bound meets the declared risk limit; 1 means
+it does not (including zero accepted samples); 2 means invalid input. Never tune
+against this holdout or reuse it to select among multiple policies. A pass does
+not cover distribution shift. See [statistical scope](docs/design.md).
+
+## Python API
+
+```python
+from calibroute import PredictionRecord, fit_policy, route_batch
+
+validation = [
+    PredictionRecord("a", 0.98, True),
+    PredictionRecord("b", 0.82, True),
+    PredictionRecord("c", 0.55, False),
+]
+
+policy = fit_policy(validation, max_risk=0.10, min_coverage=0.50,
+                    risk_method="empirical")  # Tiny illustrative sample only.
+decisions, summary = route_batch(
+    [PredictionRecord("new", 0.74)],
+    policy,
+)
+```
+
+## Limitations
+
+CalibRoute is an evaluation and routing tool, not a safety certification.
+Thresholds should be revalidated after changes to the model, task, prompt, or
+deployment distribution. See [design principles](docs/design.md) for details.
+
+## Development
+
+```bash
+python -m unittest discover -s tests -v
+```
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) and [ROADMAP.md](ROADMAP.md).
+Release preparation is documented in [releasing](docs/releasing.md).
+The [Financial NER case study](examples/financial_ner/README.md) demonstrates
+the adapter and cross-domain failure pattern on 2,098 derived prediction rows.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

calibroute_ai-0.3.0/ROADMAP.md

@@ -0,0 +1,45 @@
+# Roadmap
+
+## 0.1 - Core control contract
+
+- CSV and JSONL input
+- confidence, calibration, and risk-coverage audit
+- validation-only threshold fitting
+- shift-first accept/review/abstain routing
+- deterministic tests and GitHub Actions
+
+## 0.2 - Financial NER reference adapter
+
+- [x] convert encoder and generative sentence outputs into the common schema
+- [x] reproduce multi-domain confidence and risk analyses
+- [x] add exact one-sided binomial risk bounds
+- [x] calibrate shift thresholds by batch size and handle small batches
+- [ ] add entity-level policies
+- [ ] publish a complete external data acquisition manifest
+
+## 0.3 - Correctness and reproducible evaluation
+
+- [x] tie-aware fitting and audit operating points
+- [x] independent frozen-policy holdout assessment
+- [x] O(n log n) tie-aware AUROC
+- [x] input validation, package metadata, provenance, and build checks
+
+## Future uncertainty monitoring
+
+- pluggable embedding and feature-shift detectors
+- bootstrap uncertainty for audit metrics
+- drift alerts over time windows
+
+## 0.4 - Agent and language-system adapters
+
+- structured-output and tool-call correctness adapters
+- cost-sensitive escalation policies
+- regression gates for persistent agent updates
+- policy comparison reports
+
+## 1.0 - Stable public API
+
+- versioned policy and report schemas
+- expanded documentation and examples
+- independent reproducibility report
+- PyPI release and archival DOI

calibroute_ai-0.3.0/SECURITY.md

@@ -0,0 +1,12 @@
+# Security policy
+
+CalibRoute can influence whether AI outputs are automated or reviewed. Please do
+not disclose an exploitable routing bypass in a public issue before maintainers
+have had a reasonable opportunity to assess it.
+
+Until a private reporting address is published, prepare a minimal reproduction
+without confidential data and contact a maintainer through a private channel.
+
+CalibRoute does not execute model outputs or downstream actions. Integrators
+should preserve that separation and apply least-privilege controls to any system
+that consumes CalibRoute decisions.

calibroute_ai-0.3.0/docs/design.md

@@ -0,0 +1,80 @@
+# Design principles
+
+## Shift first, confidence second
+
+A model can be confidently wrong when deployment inputs differ from validation
+data. CalibRoute therefore performs a batch-level shift check before applying an
+instance-level confidence threshold. Severe shift results in human review rather
+than silently trusting a threshold learned under a different regime.
+
+## Validation-only fitting
+
+Policies are fitted on labeled validation records. Test or production labels are
+for evaluation, not threshold selection. The serialized policy records its
+achieved validation risk and coverage so an operating point can be audited.
+
+## Bounded actions
+
+CalibRoute emits one of three explicit actions:
+
+- `accept`: permit the configured downstream automation;
+- `human_review`: pause automation and request qualified review;
+- `abstain`: withhold the output when confidence is below the review band.
+
+CalibRoute itself does not execute downstream actions. The calling application
+decides what each action is allowed to do.
+
+## Observable reasons
+
+Every decision carries a stable reason code and the batch shift score. This is
+more useful for audit than returning a binary answer without the control path.
+
+## Current boundary
+
+CalibRoute uses the Jensen-Shannon divergence between validation and deployment
+confidence histograms as a transparent baseline. A confidence distribution is
+not a complete representation of input shift. Planned adapters will add feature,
+embedding, and task-specific shift detectors without changing the control API.
+
+## Statistical scope
+
+`fit_policy` evaluates only complete confidence groups: its reported accepted
+set is exactly `confidence >= accept_threshold`. Equal scores are never split.
+The same rule applies to audit curves. Requested coverage rounds upward to the
+next attainable threshold; AURC is the right-step integral weighted by coverage
+increments, not the average over unique thresholds.
+
+The Clopper-Pearson bound is an exact one-sided binomial interval for a fixed
+acceptance rule on independent identically distributed labeled examples.
+Searching thresholds on those same labels invalidates a claim of nominal
+coverage for the selected threshold. The fit bound is therefore a selection
+diagnostic. `validate_policy` / `calibroute validate` assess a frozen threshold
+on a separate IID holdout. The caller must enforce independence; the software
+cannot establish it from identifiers. Repeated policy selection against the
+holdout also invalidates the interpretation. No accepted examples means no
+evidence and a failed validation, not zero risk.
+
+The validation API assesses threshold acceptance before the batch shift gate.
+It does not certify the conditional risk of the full batch-dependent pipeline,
+future domains, correlated records, or changing models/prompts.
+
+JSD calibration simulates batches from the empirical reference histogram. It
+does not account for uncertainty in that reference or provide a distribution-
+free false-alarm guarantee. The default 0.02 effect floor is a heuristic and
+must be chosen on development data for the application. Batches smaller than
+20 are marked `insufficient_batch` and use instance thresholds; callers may
+instead queue them or require review. Unchanged confidence histograms can hide
+large correctness changes. Monitoring confidence cannot replace labeled audits.
+
+## Migrating from 0.2
+
+Refit policies and regenerate audits. Old policies can retain thresholds selected
+inside ties. Missing `max_js_divergence` now means adaptive monitoring, matching
+the constructor; specify `0.10` explicitly to retain that older default.
+Policy schema is 1.2 and audit schema is 1.1. Earlier AURC and selective-point
+values with ties are not directly comparable to the corrected reports.
+
+AUROC uses an O(n log n) sort and tie-aware rank aggregation. Exact binomial
+threshold search can still be expensive for many unique scores and errors;
+the dependency-free implementation targets modest validation sets. Use the
+empirical mode for exploratory large audits, then validate a frozen threshold.

calibroute_ai-0.3.0/docs/releasing.md

@@ -0,0 +1,50 @@
+# Release workflow
+
+0.3.0 is the first intended PyPI release. Do not publish 0.2.0.
+Public Git history preserves the earlier defect and its correction.
+
+## Trusted Publishing setup
+
+On PyPI, verify the account email and complete required two-factor setup.
+Under account Publishing, add a pending GitHub publisher for the first release:
+
+- PyPI project: `calibroute-ai`
+- GitHub owner: `Garyouki`
+- Repository: `calibroute`
+- Workflow filename: `publish.yml`
+- Environment: `pypi`
+
+The workflow `.github/workflows/publish.yml` is manually dispatched from `main`.
+It tests and builds the selected commit, then publishes the resulting artifacts
+in a separate job with short-lived OIDC credentials. No password or API token
+belongs in this repository. Configure environment protection on GitHub if
+additional release approval is desired. Do not dispatch until the corresponding
+PyPI publisher has been configured and release metadata is finalized.
+
+## Release checks
+
+1. Run tests, lint, the case study, and build checks below.
+2. Check package version, `__version__`, CITATION, and changelog agree.
+3. On actual release, set CITATION `date-released` to that release date and
+   replace the changelog's pending marker. Create an annotated tag `v0.3.0`
+   and a GitHub Release pointing at the verified commit.
+4. Publish built artifacts using PyPI Trusted Publishing or a token configured
+   locally. Never put a token in chat, source, or a committed `.pypirc`.
+5. Verify installation from PyPI in a fresh environment before describing the
+   package as published. Add the PyPI install command as the default README path.
+
+```bash
+python -m pip install -e ".[dev]"
+python -m unittest discover -s tests -v
+ruff check src tests examples/financial_ner/run_case_study.py
+ruff format --check src tests examples/financial_ner/run_case_study.py
+python examples/financial_ner/run_case_study.py
+python -m build
+python -m twine check dist/*
+```
+
+Preserve the release URL, commit, checksums, CI result, and reproducible outputs.
+For an impact record, distinguish authored software from independent adoption:
+record only verifiable external uses, citations, contributed issues, and feedback
+with their dates and links. A repository or passing CI alone does not establish
+independent use or practical impact.