judgekit 0.2.0__tar.gz → 0.2.2__tar.gz

{judgekit-0.2.0/src/judgekit.egg-info → judgekit-0.2.2}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: judgekit
-Version: 0.2.0
-Summary: Audit an LLM judge before you trust it: agreement, bias probes, calibration and consistency, with bootstrap confidence intervals.
+Version: 0.2.2
+Summary: Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability.
 Author: Hadi Mohammadi
 License: MIT
 Project-URL: Homepage, https://github.com/mohammadi-hadi/judgekit
@@ -30,11 +30,19 @@ Requires-Dist: krippendorff>=0.6; extra == "dev"
 Requires-Dist: statsmodels>=0.14; extra == "dev"
 Dynamic: license-file
 
-# judgekit
+<div align="center">
 
-Audit an LLM judge before you trust it.
+# judgekit
 
 [![ci](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/judgekit)](https://pypi.org/project/judgekit/)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21802868.svg)](https://doi.org/10.5281/zenodo.21802868)
+[![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/judgekit/blob/main/LICENSE)
+
+*Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability.*
+
+</div>
 
 An LLM judge is a model too. It can prefer whichever answer came first, reward
 padding, favor its own outputs, hug the middle of the scale, and say "90%
@@ -63,35 +71,40 @@ The row to sit with is `length-lover`: quadratic-weighted kappa with humans of
 0.80 — a number most eval writeups would celebrate — while carrying a verbosity
 bias of +0.82. Agreement is not an audit.
 
-`make demo` regenerates this table, the [full report](results/report.md) and
+`make demo` regenerates this table, the [full report](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) and
 the figures from fixed seeds; CI rebuilds them from pinned dependencies and
 fails if a committed number differs from what the code produces.
 
-![position preference](results/figures/position.png)
+![position preference](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/position.png)
 
-![reliability](results/figures/reliability.png)
+![reliability](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/reliability.png)
 
 ## Two real judges, audited
 
-[`examples/audit_ollama.py`](examples/audit_ollama.py) builds a small
+[`examples/audit_ollama.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/examples/audit_ollama.py) builds a small
 adversarial arithmetic benchmark — one answer provably correct by
 construction, and half the time the wrong answer is the long, worked-through
 one — then asks a local Ollama model to judge every pair in both presentation
 orders. Verdicts and full reports are committed under
-[examples/results](examples/results/).
+[examples/results](https://github.com/mohammadi-hadi/judgekit/tree/main/examples/results/).
 
 | judge (temperature 0) | truth agreement | position preference | verbosity excess | flips under swap | decisive on identical pairs |
 |---|---|---|---|---|---|
+| aya-expanse:8b | 0.433 | **0.910** (flag) | +0.106 | 0.819 | 1.000 |
 | qwen2.5:14b | 0.667 | **0.768** (flag) | +0.131 | 0.568 | 0.200 |
 | llama3.1:8b | 0.443 | 0.576 | **+0.483** (flag) | 0.171 | 1.000 |
 
-Same task, two different failure modes. The 14B judge follows slot order:
-presented-first wins 76.8% of decisive verdicts and swapping the order flips
-56.8% of matched pairs. The 8B judge grades effort instead, picking the longer
-answer 48 points more often than correctness warrants, agreeing with ground
-truth less often than a coin flip, and never once declaring a tie between two
-identical answers. n=105 pairs, adversarial by construction — an illustration
-of what the probes see, not a model leaderboard.
+Same task, different failure modes at different severities. Two judges follow
+slot order — aya picks whatever is presented first 91.0% of the time and flips
+81.9% of pairs when the order is swapped; qwen sits at 76.8% and 56.8%. The
+llama judge grades effort instead, picking the longer answer 48 points more
+often than correctness warrants and scoring 0.443 against ground truth. Two of
+the three never once declare a tie between identical answers, which is also
+why that 0.443 is softer than it looks: those 15 pairs are unwinnable for a
+judge that never ties, and on the 90 adversarial pairs alone the llama judge
+reaches 0.517 — level with chance rather than below it. n=105 pairs,
+adversarial by construction — an illustration of what the probes see, not a
+model leaderboard.
 
 ## Install
 
@@ -99,13 +112,9 @@ of what the probes see, not a model leaderboard.
 pip install judgekit
 ```
 
-Or the latest development version straight from GitHub:
-
-```
-pip install git+https://github.com/mohammadi-hadi/judgekit
-```
-
-Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
+Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib. The
+development version installs with
+`pip install git+https://github.com/mohammadi-hadi/judgekit`.
 
 ## Quickstart
 
@@ -177,7 +186,7 @@ judge.
   silent on the clean judge, including a world where longer answers genuinely
   are better and the raw length correlation would convict an honest judge.
 - **Drift-checked results.** Every number above and in
-  [results/report.md](results/report.md) is regenerated by CI from pinned
+  [results/report.md](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) is regenerated by CI from pinned
   dependencies and diffed against the committed copy.
 
 ## Design notes
@@ -187,7 +196,7 @@ judge.
   defect-free judge could plausibly occupy — position 0.45–0.55, correlations
   ±0.10, self-preference ±0.20 scale points, spread 0.90–1.10, ECE ≤ 0.05 —
   and flags only when the whole 95% interval leaves it. The bands are opinions,
-  stated in one place ([`bias.py`](src/judgekit/bias.py)) and easy to disagree
+  stated in one place ([`bias.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/src/judgekit/bias.py)) and easy to disagree
   with.
 - **Item-level bootstrap.** Re-judgments and swapped presentations of an item
   are resampled together; treating correlated rows as independent evidence
@@ -230,10 +239,27 @@ the LLM-as-a-judge survey ([Gu et al., 2024](https://arxiv.org/abs/2411.15594)).
 judgekit's contribution is packaging the checks as one auditable tool with
 uncertainty on every number and validation by implantation.
 
-Companion projects: [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
-measures what outcome-only judges miss on agent trajectories;
-[EvalMORAAL](https://github.com/mohammadi-hadi/EvalMORAAL) applies
-chain-of-thought judging to moral alignment across 20 models.
+## Part of evalstack
+
+[evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
+eleven evaluation tools: what each one measures, what it found on real data,
+and the two chains that run end to end.
+
+## Citation
+
+If judgekit is useful in your research, please cite it (see
+[CITATION.cff](https://github.com/mohammadi-hadi/judgekit/blob/main/CITATION.cff)):
+
+```bibtex
+@software{mohammadi_judgekit,
+  author  = {Mohammadi, Hadi},
+  title   = {judgekit: audit an LLM judge before you trust it},
+  url     = {https://github.com/mohammadi-hadi/judgekit},
+  doi     = {10.5281/zenodo.21802868},
+  version = {0.2.2},
+  year    = {2026}
+}
+```
 
 ## License
 

{judgekit-0.2.0 → judgekit-0.2.2}/README.md

@@ -1,8 +1,16 @@
-# judgekit
+<div align="center">
 
-Audit an LLM judge before you trust it.
+# judgekit
 
 [![ci](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/judgekit)](https://pypi.org/project/judgekit/)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21802868.svg)](https://doi.org/10.5281/zenodo.21802868)
+[![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/judgekit/blob/main/LICENSE)
+
+*Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability.*
+
+</div>
 
 An LLM judge is a model too. It can prefer whichever answer came first, reward
 padding, favor its own outputs, hug the middle of the scale, and say "90%
@@ -31,35 +39,40 @@ The row to sit with is `length-lover`: quadratic-weighted kappa with humans of
 0.80 — a number most eval writeups would celebrate — while carrying a verbosity
 bias of +0.82. Agreement is not an audit.
 
-`make demo` regenerates this table, the [full report](results/report.md) and
+`make demo` regenerates this table, the [full report](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) and
 the figures from fixed seeds; CI rebuilds them from pinned dependencies and
 fails if a committed number differs from what the code produces.
 
-![position preference](results/figures/position.png)
+![position preference](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/position.png)
 
-![reliability](results/figures/reliability.png)
+![reliability](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/reliability.png)
 
 ## Two real judges, audited
 
-[`examples/audit_ollama.py`](examples/audit_ollama.py) builds a small
+[`examples/audit_ollama.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/examples/audit_ollama.py) builds a small
 adversarial arithmetic benchmark — one answer provably correct by
 construction, and half the time the wrong answer is the long, worked-through
 one — then asks a local Ollama model to judge every pair in both presentation
 orders. Verdicts and full reports are committed under
-[examples/results](examples/results/).
+[examples/results](https://github.com/mohammadi-hadi/judgekit/tree/main/examples/results/).
 
 | judge (temperature 0) | truth agreement | position preference | verbosity excess | flips under swap | decisive on identical pairs |
 |---|---|---|---|---|---|
+| aya-expanse:8b | 0.433 | **0.910** (flag) | +0.106 | 0.819 | 1.000 |
 | qwen2.5:14b | 0.667 | **0.768** (flag) | +0.131 | 0.568 | 0.200 |
 | llama3.1:8b | 0.443 | 0.576 | **+0.483** (flag) | 0.171 | 1.000 |
 
-Same task, two different failure modes. The 14B judge follows slot order:
-presented-first wins 76.8% of decisive verdicts and swapping the order flips
-56.8% of matched pairs. The 8B judge grades effort instead, picking the longer
-answer 48 points more often than correctness warrants, agreeing with ground
-truth less often than a coin flip, and never once declaring a tie between two
-identical answers. n=105 pairs, adversarial by construction — an illustration
-of what the probes see, not a model leaderboard.
+Same task, different failure modes at different severities. Two judges follow
+slot order — aya picks whatever is presented first 91.0% of the time and flips
+81.9% of pairs when the order is swapped; qwen sits at 76.8% and 56.8%. The
+llama judge grades effort instead, picking the longer answer 48 points more
+often than correctness warrants and scoring 0.443 against ground truth. Two of
+the three never once declare a tie between identical answers, which is also
+why that 0.443 is softer than it looks: those 15 pairs are unwinnable for a
+judge that never ties, and on the 90 adversarial pairs alone the llama judge
+reaches 0.517 — level with chance rather than below it. n=105 pairs,
+adversarial by construction — an illustration of what the probes see, not a
+model leaderboard.
 
 ## Install
 
@@ -67,13 +80,9 @@ of what the probes see, not a model leaderboard.
 pip install judgekit
 ```
 
-Or the latest development version straight from GitHub:
-
-```
-pip install git+https://github.com/mohammadi-hadi/judgekit
-```
-
-Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
+Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib. The
+development version installs with
+`pip install git+https://github.com/mohammadi-hadi/judgekit`.
 
 ## Quickstart
 
@@ -145,7 +154,7 @@ judge.
   silent on the clean judge, including a world where longer answers genuinely
   are better and the raw length correlation would convict an honest judge.
 - **Drift-checked results.** Every number above and in
-  [results/report.md](results/report.md) is regenerated by CI from pinned
+  [results/report.md](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) is regenerated by CI from pinned
   dependencies and diffed against the committed copy.
 
 ## Design notes
@@ -155,7 +164,7 @@ judge.
   defect-free judge could plausibly occupy — position 0.45–0.55, correlations
   ±0.10, self-preference ±0.20 scale points, spread 0.90–1.10, ECE ≤ 0.05 —
   and flags only when the whole 95% interval leaves it. The bands are opinions,
-  stated in one place ([`bias.py`](src/judgekit/bias.py)) and easy to disagree
+  stated in one place ([`bias.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/src/judgekit/bias.py)) and easy to disagree
   with.
 - **Item-level bootstrap.** Re-judgments and swapped presentations of an item
   are resampled together; treating correlated rows as independent evidence
@@ -198,10 +207,27 @@ the LLM-as-a-judge survey ([Gu et al., 2024](https://arxiv.org/abs/2411.15594)).
 judgekit's contribution is packaging the checks as one auditable tool with
 uncertainty on every number and validation by implantation.
 
-Companion projects: [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
-measures what outcome-only judges miss on agent trajectories;
-[EvalMORAAL](https://github.com/mohammadi-hadi/EvalMORAAL) applies
-chain-of-thought judging to moral alignment across 20 models.
+## Part of evalstack
+
+[evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
+eleven evaluation tools: what each one measures, what it found on real data,
+and the two chains that run end to end.
+
+## Citation
+
+If judgekit is useful in your research, please cite it (see
+[CITATION.cff](https://github.com/mohammadi-hadi/judgekit/blob/main/CITATION.cff)):
+
+```bibtex
+@software{mohammadi_judgekit,
+  author  = {Mohammadi, Hadi},
+  title   = {judgekit: audit an LLM judge before you trust it},
+  url     = {https://github.com/mohammadi-hadi/judgekit},
+  doi     = {10.5281/zenodo.21802868},
+  version = {0.2.2},
+  year    = {2026}
+}
+```
 
 ## License
 

{judgekit-0.2.0 → judgekit-0.2.2}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "judgekit"
-version = "0.2.0"
-description = "Audit an LLM judge before you trust it: agreement, bias probes, calibration and consistency, with bootstrap confidence intervals."
+version = "0.2.2"
+description = "Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability."
 readme = "README.md"
 license = { text = "MIT" }
 authors = [{ name = "Hadi Mohammadi" }]

{judgekit-0.2.0 → judgekit-0.2.2}/src/judgekit/__init__.py

@@ -1,11 +1,11 @@
-"""judgekit: audit an LLM judge before you trust it."""
+"""judgekit: bias probes for LLM judges, each with a bootstrap confidence interval."""
 
 from judgekit.audit import AuditReport, run_audit
 from judgekit.io import dump_verdicts, load_verdicts
 from judgekit.result import ProbeResult, SkippedProbe
 from judgekit.schema import BinaryVerdict, GradedVerdict, PairwiseVerdict
 
-__version__ = "0.2.0"
+__version__ = "0.2.2"
 
 __all__ = [
     "AuditReport",

{judgekit-0.2.0 → judgekit-0.2.2/src/judgekit.egg-info}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: judgekit
-Version: 0.2.0
-Summary: Audit an LLM judge before you trust it: agreement, bias probes, calibration and consistency, with bootstrap confidence intervals.
+Version: 0.2.2
+Summary: Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability.
 Author: Hadi Mohammadi
 License: MIT
 Project-URL: Homepage, https://github.com/mohammadi-hadi/judgekit
@@ -30,11 +30,19 @@ Requires-Dist: krippendorff>=0.6; extra == "dev"
 Requires-Dist: statsmodels>=0.14; extra == "dev"
 Dynamic: license-file
 
-# judgekit
+<div align="center">
 
-Audit an LLM judge before you trust it.
+# judgekit
 
 [![ci](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/judgekit/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/judgekit)](https://pypi.org/project/judgekit/)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21802868.svg)](https://doi.org/10.5281/zenodo.21802868)
+[![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/judgekit/blob/main/LICENSE)
+
+*Bias probes for LLM judges, each with a bootstrap confidence interval: position, verbosity, self-preference, calibration and stability.*
+
+</div>
 
 An LLM judge is a model too. It can prefer whichever answer came first, reward
 padding, favor its own outputs, hug the middle of the scale, and say "90%
@@ -63,35 +71,40 @@ The row to sit with is `length-lover`: quadratic-weighted kappa with humans of
 0.80 — a number most eval writeups would celebrate — while carrying a verbosity
 bias of +0.82. Agreement is not an audit.
 
-`make demo` regenerates this table, the [full report](results/report.md) and
+`make demo` regenerates this table, the [full report](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) and
 the figures from fixed seeds; CI rebuilds them from pinned dependencies and
 fails if a committed number differs from what the code produces.
 
-![position preference](results/figures/position.png)
+![position preference](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/position.png)
 
-![reliability](results/figures/reliability.png)
+![reliability](https://raw.githubusercontent.com/mohammadi-hadi/judgekit/main/results/figures/reliability.png)
 
 ## Two real judges, audited
 
-[`examples/audit_ollama.py`](examples/audit_ollama.py) builds a small
+[`examples/audit_ollama.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/examples/audit_ollama.py) builds a small
 adversarial arithmetic benchmark — one answer provably correct by
 construction, and half the time the wrong answer is the long, worked-through
 one — then asks a local Ollama model to judge every pair in both presentation
 orders. Verdicts and full reports are committed under
-[examples/results](examples/results/).
+[examples/results](https://github.com/mohammadi-hadi/judgekit/tree/main/examples/results/).
 
 | judge (temperature 0) | truth agreement | position preference | verbosity excess | flips under swap | decisive on identical pairs |
 |---|---|---|---|---|---|
+| aya-expanse:8b | 0.433 | **0.910** (flag) | +0.106 | 0.819 | 1.000 |
 | qwen2.5:14b | 0.667 | **0.768** (flag) | +0.131 | 0.568 | 0.200 |
 | llama3.1:8b | 0.443 | 0.576 | **+0.483** (flag) | 0.171 | 1.000 |
 
-Same task, two different failure modes. The 14B judge follows slot order:
-presented-first wins 76.8% of decisive verdicts and swapping the order flips
-56.8% of matched pairs. The 8B judge grades effort instead, picking the longer
-answer 48 points more often than correctness warrants, agreeing with ground
-truth less often than a coin flip, and never once declaring a tie between two
-identical answers. n=105 pairs, adversarial by construction — an illustration
-of what the probes see, not a model leaderboard.
+Same task, different failure modes at different severities. Two judges follow
+slot order — aya picks whatever is presented first 91.0% of the time and flips
+81.9% of pairs when the order is swapped; qwen sits at 76.8% and 56.8%. The
+llama judge grades effort instead, picking the longer answer 48 points more
+often than correctness warrants and scoring 0.443 against ground truth. Two of
+the three never once declare a tie between identical answers, which is also
+why that 0.443 is softer than it looks: those 15 pairs are unwinnable for a
+judge that never ties, and on the 90 adversarial pairs alone the llama judge
+reaches 0.517 — level with chance rather than below it. n=105 pairs,
+adversarial by construction — an illustration of what the probes see, not a
+model leaderboard.
 
 ## Install
 
@@ -99,13 +112,9 @@ of what the probes see, not a model leaderboard.
 pip install judgekit
 ```
 
-Or the latest development version straight from GitHub:
-
-```
-pip install git+https://github.com/mohammadi-hadi/judgekit
-```
-
-Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
+Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib. The
+development version installs with
+`pip install git+https://github.com/mohammadi-hadi/judgekit`.
 
 ## Quickstart
 
@@ -177,7 +186,7 @@ judge.
   silent on the clean judge, including a world where longer answers genuinely
   are better and the raw length correlation would convict an honest judge.
 - **Drift-checked results.** Every number above and in
-  [results/report.md](results/report.md) is regenerated by CI from pinned
+  [results/report.md](https://github.com/mohammadi-hadi/judgekit/blob/main/results/report.md) is regenerated by CI from pinned
   dependencies and diffed against the committed copy.
 
 ## Design notes
@@ -187,7 +196,7 @@ judge.
   defect-free judge could plausibly occupy — position 0.45–0.55, correlations
   ±0.10, self-preference ±0.20 scale points, spread 0.90–1.10, ECE ≤ 0.05 —
   and flags only when the whole 95% interval leaves it. The bands are opinions,
-  stated in one place ([`bias.py`](src/judgekit/bias.py)) and easy to disagree
+  stated in one place ([`bias.py`](https://github.com/mohammadi-hadi/judgekit/blob/main/src/judgekit/bias.py)) and easy to disagree
   with.
 - **Item-level bootstrap.** Re-judgments and swapped presentations of an item
   are resampled together; treating correlated rows as independent evidence
@@ -230,10 +239,27 @@ the LLM-as-a-judge survey ([Gu et al., 2024](https://arxiv.org/abs/2411.15594)).
 judgekit's contribution is packaging the checks as one auditable tool with
 uncertainty on every number and validation by implantation.
 
-Companion projects: [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
-measures what outcome-only judges miss on agent trajectories;
-[EvalMORAAL](https://github.com/mohammadi-hadi/EvalMORAAL) applies
-chain-of-thought judging to moral alignment across 20 models.
+## Part of evalstack
+
+[evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
+eleven evaluation tools: what each one measures, what it found on real data,
+and the two chains that run end to end.
+
+## Citation
+
+If judgekit is useful in your research, please cite it (see
+[CITATION.cff](https://github.com/mohammadi-hadi/judgekit/blob/main/CITATION.cff)):
+
+```bibtex
+@software{mohammadi_judgekit,
+  author  = {Mohammadi, Hadi},
+  title   = {judgekit: audit an LLM judge before you trust it},
+  url     = {https://github.com/mohammadi-hadi/judgekit},
+  doi     = {10.5281/zenodo.21802868},
+  version = {0.2.2},
+  year    = {2026}
+}
+```
 
 ## License