abkit 0.1.0__tar.gz → 0.1.2__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (31) hide show
  1. {abkit-0.1.0/src/abkit.egg-info → abkit-0.1.2}/PKG-INFO +31 -19
  2. {abkit-0.1.0 → abkit-0.1.2}/README.md +29 -17
  3. {abkit-0.1.0 → abkit-0.1.2}/pyproject.toml +2 -2
  4. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/__init__.py +2 -2
  5. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/cli.py +3 -1
  6. {abkit-0.1.0 → abkit-0.1.2/src/abkit.egg-info}/PKG-INFO +31 -19
  7. {abkit-0.1.0 → abkit-0.1.2}/LICENSE +0 -0
  8. {abkit-0.1.0 → abkit-0.1.2}/setup.cfg +0 -0
  9. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/audit.py +0 -0
  10. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/bootstrap.py +0 -0
  11. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/demo.py +0 -0
  12. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/io.py +0 -0
  13. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/probes.py +0 -0
  14. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/py.typed +0 -0
  15. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/report.py +0 -0
  16. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/result.py +0 -0
  17. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/schema.py +0 -0
  18. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/stats.py +0 -0
  19. {abkit-0.1.0 → abkit-0.1.2}/src/abkit/synthetic.py +0 -0
  20. {abkit-0.1.0 → abkit-0.1.2}/src/abkit.egg-info/SOURCES.txt +0 -0
  21. {abkit-0.1.0 → abkit-0.1.2}/src/abkit.egg-info/dependency_links.txt +0 -0
  22. {abkit-0.1.0 → abkit-0.1.2}/src/abkit.egg-info/entry_points.txt +0 -0
  23. {abkit-0.1.0 → abkit-0.1.2}/src/abkit.egg-info/requires.txt +0 -0
  24. {abkit-0.1.0 → abkit-0.1.2}/src/abkit.egg-info/top_level.txt +0 -0
  25. {abkit-0.1.0 → abkit-0.1.2}/tests/test_bootstrap.py +0 -0
  26. {abkit-0.1.0 → abkit-0.1.2}/tests/test_cli.py +0 -0
  27. {abkit-0.1.0 → abkit-0.1.2}/tests/test_probes.py +0 -0
  28. {abkit-0.1.0 → abkit-0.1.2}/tests/test_result.py +0 -0
  29. {abkit-0.1.0 → abkit-0.1.2}/tests/test_schema.py +0 -0
  30. {abkit-0.1.0 → abkit-0.1.2}/tests/test_stats.py +0 -0
  31. {abkit-0.1.0 → abkit-0.1.2}/tests/test_synthetic.py +0 -0
{abkit-0.1.0/src/abkit.egg-info → abkit-0.1.2}/PKG-INFO RENAMED
@@ -1,7 +1,7 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: abkit
3
- Version: 0.1.0
4
- Summary: Audit an A/B-test readout before you ship the decision
3
+ Version: 0.1.2
4
+ Summary: Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts
5
5
  Author: Hadi Mohammadi
6
6
  License: MIT
7
7
  Project-URL: Homepage, https://github.com/mohammadi-hadi/abkit
@@ -28,11 +28,19 @@ Requires-Dist: scipy>=1.12; extra == "dev"
28
28
  Requires-Dist: statsmodels>=0.14; extra == "dev"
29
29
  Dynamic: license-file
30
30
 
31
- # abkit
31
+ <div align="center">
32
32
 
33
- Audit an A/B-test readout before you ship the decision.
33
+ # abkit
34
34
 
35
35
  [![ci](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml)
36
+ [![PyPI](https://img.shields.io/pypi/v/abkit)](https://pypi.org/project/abkit/)
37
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21809478.svg)](https://doi.org/10.5281/zenodo.21809478)
38
+ [![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
39
+ [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/abkit/blob/main/LICENSE)
40
+
41
+ *Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts.*
42
+
43
+ </div>
36
44
 
37
45
  An experiment readout can say "B wins, p = 0.03, ship it" while the traffic
38
46
  split was broken, three whale customers carried the whole effect, the analyst
@@ -67,22 +75,22 @@ is indistinguishable from a real win — p < 0.05, decent sample size — but th
67
75
  schedule it followed has a 19% false-positive rate, not 5%. The audit flags
68
76
  the process, which is the only place that defect lives.
69
77
 
70
- ![peeking](results/figures/peeking.png)
78
+ ![peeking](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/peeking.png)
71
79
 
72
- ![exaggeration](results/figures/exaggeration.png)
80
+ ![exaggeration](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/exaggeration.png)
73
81
 
74
- `make demo` regenerates the table, the [full report](results/report.md) and
82
+ `make demo` regenerates the table, the [full report](https://github.com/mohammadi-hadi/abkit/blob/main/results/report.md) and
75
83
  the figures from fixed seeds; CI rebuilds them from pinned dependencies and
76
84
  fails if a committed number differs from what the code produces.
77
85
 
78
86
  ## 4,873 real experiments, audited
79
87
 
80
- [`examples/upworthy/`](examples/upworthy/) runs abkit's aggregate checks over
88
+ [`examples/upworthy/`](https://github.com/mohammadi-hadi/abkit/tree/main/examples/upworthy/) runs abkit's aggregate checks over
81
89
  the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038/s41597-021-00934-7)
82
90
  — 4,873 real headline A/B tests from 2013–2015. Committed results:
83
- [examples/upworthy/results/upworthy.md](examples/upworthy/results/upworthy.md).
91
+ [examples/upworthy/results/upworthy.md](https://github.com/mohammadi-hadi/abkit/blob/main/examples/upworthy/results/upworthy.md).
84
92
 
85
- - **16% of real tests fail the sample-ratio check** at p < 0.001 against the
93
+ - **15% of real tests fail the sample-ratio check** at p < 0.001 against the
86
94
  platform's equal-allocation design — and not because packages were added
87
95
  mid-test; the rate is, if anything, higher in tests whose variants all
88
96
  launched together. Those readouts carried unknown exposure bias.
@@ -96,7 +104,7 @@ the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038
96
104
  ## Install
97
105
 
98
106
  ```
99
- pip install git+https://github.com/mohammadi-hadi/abkit
107
+ pip install abkit
100
108
  ```
101
109
 
102
110
  Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
@@ -184,7 +192,7 @@ experiment.
184
192
  ±0.10 — and flag only when the whole 95% bootstrap CI leaves it. Exact
185
193
  procedures use the field's conventional thresholds: SRM at p < 0.001,
186
194
  contamination at zero tolerance. All constants live in one place
187
- ([`probes.py`](src/abkit/probes.py)) and are easy to disagree with.
195
+ ([`probes.py`](https://github.com/mohammadi-hadi/abkit/blob/main/src/abkit/probes.py)) and are easy to disagree with.
188
196
  - **The stopping rule is audited, not the final p.** A peeked experiment's
189
197
  final test statistic looks ordinary; what is broken is the procedure that
190
198
  produced it. The peeking probe Monte-Carlos the analyst's actual look
@@ -237,22 +245,26 @@ The real-data case study uses the Upworthy Research Archive
237
245
  abkit's contribution is packaging the checks as one auditable tool with
238
246
  uncertainty on every number and validation by implantation.
239
247
 
240
- Companion projects: [judgekit](https://github.com/mohammadi-hadi/judgekit)
241
- applies the same audit-before-you-trust pattern to LLM judges;
242
- [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
243
- measures what outcome-only judges miss on agent trajectories.
248
+ ## Part of evalstack
249
+
250
+ [evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
251
+ eleven evaluation tools: what each one measures, what it found on real data,
252
+ and the two chains that run end to end.
244
253
 
245
254
  ## Citation
246
255
 
247
- If abkit is useful in your work, please cite it (see
248
- [CITATION.cff](CITATION.cff)):
256
+ Releases are archived on Zenodo. Cite the concept DOI
257
+ [10.5281/zenodo.21809478](https://doi.org/10.5281/zenodo.21809478), which always
258
+ resolves to the latest version; structured metadata is in
259
+ [CITATION.cff](https://github.com/mohammadi-hadi/abkit/blob/main/CITATION.cff).
249
260
 
250
261
  ```bibtex
251
262
  @software{mohammadi_abkit,
252
263
  author = {Mohammadi, Hadi},
253
264
  title = {abkit: audit an A/B-test readout before you ship the decision},
254
265
  url = {https://github.com/mohammadi-hadi/abkit},
255
- version = {0.1.0},
266
+ doi = {10.5281/zenodo.21809478},
267
+ version = {0.1.2},
256
268
  year = {2026}
257
269
  }
258
270
  ```
{abkit-0.1.0 → abkit-0.1.2}/README.md RENAMED
@@ -1,8 +1,16 @@
1
- # abkit
1
+ <div align="center">
2
2
 
3
- Audit an A/B-test readout before you ship the decision.
3
+ # abkit
4
4
 
5
5
  [![ci](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml)
6
+ [![PyPI](https://img.shields.io/pypi/v/abkit)](https://pypi.org/project/abkit/)
7
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21809478.svg)](https://doi.org/10.5281/zenodo.21809478)
8
+ [![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
9
+ [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/abkit/blob/main/LICENSE)
10
+
11
+ *Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts.*
12
+
13
+ </div>
6
14
 
7
15
  An experiment readout can say "B wins, p = 0.03, ship it" while the traffic
8
16
  split was broken, three whale customers carried the whole effect, the analyst
@@ -37,22 +45,22 @@ is indistinguishable from a real win — p < 0.05, decent sample size — but th
37
45
  schedule it followed has a 19% false-positive rate, not 5%. The audit flags
38
46
  the process, which is the only place that defect lives.
39
47
 
40
- ![peeking](results/figures/peeking.png)
48
+ ![peeking](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/peeking.png)
41
49
 
42
- ![exaggeration](results/figures/exaggeration.png)
50
+ ![exaggeration](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/exaggeration.png)
43
51
 
44
- `make demo` regenerates the table, the [full report](results/report.md) and
52
+ `make demo` regenerates the table, the [full report](https://github.com/mohammadi-hadi/abkit/blob/main/results/report.md) and
45
53
  the figures from fixed seeds; CI rebuilds them from pinned dependencies and
46
54
  fails if a committed number differs from what the code produces.
47
55
 
48
56
  ## 4,873 real experiments, audited
49
57
 
50
- [`examples/upworthy/`](examples/upworthy/) runs abkit's aggregate checks over
58
+ [`examples/upworthy/`](https://github.com/mohammadi-hadi/abkit/tree/main/examples/upworthy/) runs abkit's aggregate checks over
51
59
  the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038/s41597-021-00934-7)
52
60
  — 4,873 real headline A/B tests from 2013–2015. Committed results:
53
- [examples/upworthy/results/upworthy.md](examples/upworthy/results/upworthy.md).
61
+ [examples/upworthy/results/upworthy.md](https://github.com/mohammadi-hadi/abkit/blob/main/examples/upworthy/results/upworthy.md).
54
62
 
55
- - **16% of real tests fail the sample-ratio check** at p < 0.001 against the
63
+ - **15% of real tests fail the sample-ratio check** at p < 0.001 against the
56
64
  platform's equal-allocation design — and not because packages were added
57
65
  mid-test; the rate is, if anything, higher in tests whose variants all
58
66
  launched together. Those readouts carried unknown exposure bias.
@@ -66,7 +74,7 @@ the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038
66
74
  ## Install
67
75
 
68
76
  ```
69
- pip install git+https://github.com/mohammadi-hadi/abkit
77
+ pip install abkit
70
78
  ```
71
79
 
72
80
  Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
@@ -154,7 +162,7 @@ experiment.
154
162
  ±0.10 — and flag only when the whole 95% bootstrap CI leaves it. Exact
155
163
  procedures use the field's conventional thresholds: SRM at p < 0.001,
156
164
  contamination at zero tolerance. All constants live in one place
157
- ([`probes.py`](src/abkit/probes.py)) and are easy to disagree with.
165
+ ([`probes.py`](https://github.com/mohammadi-hadi/abkit/blob/main/src/abkit/probes.py)) and are easy to disagree with.
158
166
  - **The stopping rule is audited, not the final p.** A peeked experiment's
159
167
  final test statistic looks ordinary; what is broken is the procedure that
160
168
  produced it. The peeking probe Monte-Carlos the analyst's actual look
@@ -207,22 +215,26 @@ The real-data case study uses the Upworthy Research Archive
207
215
  abkit's contribution is packaging the checks as one auditable tool with
208
216
  uncertainty on every number and validation by implantation.
209
217
 
210
- Companion projects: [judgekit](https://github.com/mohammadi-hadi/judgekit)
211
- applies the same audit-before-you-trust pattern to LLM judges;
212
- [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
213
- measures what outcome-only judges miss on agent trajectories.
218
+ ## Part of evalstack
219
+
220
+ [evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
221
+ eleven evaluation tools: what each one measures, what it found on real data,
222
+ and the two chains that run end to end.
214
223
 
215
224
  ## Citation
216
225
 
217
- If abkit is useful in your work, please cite it (see
218
- [CITATION.cff](CITATION.cff)):
226
+ Releases are archived on Zenodo. Cite the concept DOI
227
+ [10.5281/zenodo.21809478](https://doi.org/10.5281/zenodo.21809478), which always
228
+ resolves to the latest version; structured metadata is in
229
+ [CITATION.cff](https://github.com/mohammadi-hadi/abkit/blob/main/CITATION.cff).
219
230
 
220
231
  ```bibtex
221
232
  @software{mohammadi_abkit,
222
233
  author = {Mohammadi, Hadi},
223
234
  title = {abkit: audit an A/B-test readout before you ship the decision},
224
235
  url = {https://github.com/mohammadi-hadi/abkit},
225
- version = {0.1.0},
236
+ doi = {10.5281/zenodo.21809478},
237
+ version = {0.1.2},
226
238
  year = {2026}
227
239
  }
228
240
  ```
{abkit-0.1.0 → abkit-0.1.2}/pyproject.toml RENAMED
@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
4
4
 
5
5
  [project]
6
6
  name = "abkit"
7
- version = "0.1.0"
8
- description = "Audit an A/B-test readout before you ship the decision"
7
+ version = "0.1.2"
8
+ description = "Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts"
9
9
  readme = "README.md"
10
10
  license = {text = "MIT"}
11
11
  authors = [{name = "Hadi Mohammadi"}]
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/__init__.py RENAMED
@@ -1,11 +1,11 @@
1
- """abkit: audit an A/B-test readout before you ship the decision."""
1
+ """abkit: sample-ratio, peeking, multiple-testing and winner's-curse checks."""
2
2
 
3
3
  from .audit import run_audit
4
4
  from .io import dump_design, dump_units, load_design, load_units
5
5
  from .result import AuditResult, ProbeResult, SkippedProbe
6
6
  from .schema import Design, Look, Unit
7
7
 
8
- __version__ = "0.1.0"
8
+ __version__ = "0.1.2"
9
9
 
10
10
  __all__ = [
11
11
  "AuditResult",
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/cli.py RENAMED
@@ -14,7 +14,9 @@ from .report import inject_readme, write_report
14
14
 
15
15
  def main(argv: list[str] | None = None) -> int:
16
16
  parser = argparse.ArgumentParser(
17
- prog="abkit", description="Audit an A/B-test readout before you ship the decision."
17
+ prog="abkit",
18
+ description="Sample-ratio, peeking, multiple-testing and winner's-curse checks "
19
+ "for experiment readouts.",
18
20
  )
19
21
  parser.add_argument("--version", action="version", version=f"abkit {__version__}")
20
22
  sub = parser.add_subparsers(dest="command", required=True)
{abkit-0.1.0 → abkit-0.1.2/src/abkit.egg-info}/PKG-INFO RENAMED
@@ -1,7 +1,7 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: abkit
3
- Version: 0.1.0
4
- Summary: Audit an A/B-test readout before you ship the decision
3
+ Version: 0.1.2
4
+ Summary: Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts
5
5
  Author: Hadi Mohammadi
6
6
  License: MIT
7
7
  Project-URL: Homepage, https://github.com/mohammadi-hadi/abkit
@@ -28,11 +28,19 @@ Requires-Dist: scipy>=1.12; extra == "dev"
28
28
  Requires-Dist: statsmodels>=0.14; extra == "dev"
29
29
  Dynamic: license-file
30
30
 
31
- # abkit
31
+ <div align="center">
32
32
 
33
- Audit an A/B-test readout before you ship the decision.
33
+ # abkit
34
34
 
35
35
  [![ci](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml/badge.svg)](https://github.com/mohammadi-hadi/abkit/actions/workflows/ci.yml)
36
+ [![PyPI](https://img.shields.io/pypi/v/abkit)](https://pypi.org/project/abkit/)
37
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.21809478.svg)](https://doi.org/10.5281/zenodo.21809478)
38
+ [![Python](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
39
+ [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mohammadi-hadi/abkit/blob/main/LICENSE)
40
+
41
+ *Sample-ratio, peeking, multiple-testing and winner's-curse checks for experiment readouts.*
42
+
43
+ </div>
36
44
 
37
45
  An experiment readout can say "B wins, p = 0.03, ship it" while the traffic
38
46
  split was broken, three whale customers carried the whole effect, the analyst
@@ -67,22 +75,22 @@ is indistinguishable from a real win — p < 0.05, decent sample size — but th
67
75
  schedule it followed has a 19% false-positive rate, not 5%. The audit flags
68
76
  the process, which is the only place that defect lives.
69
77
 
70
- ![peeking](results/figures/peeking.png)
78
+ ![peeking](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/peeking.png)
71
79
 
72
- ![exaggeration](results/figures/exaggeration.png)
80
+ ![exaggeration](https://raw.githubusercontent.com/mohammadi-hadi/abkit/main/results/figures/exaggeration.png)
73
81
 
74
- `make demo` regenerates the table, the [full report](results/report.md) and
82
+ `make demo` regenerates the table, the [full report](https://github.com/mohammadi-hadi/abkit/blob/main/results/report.md) and
75
83
  the figures from fixed seeds; CI rebuilds them from pinned dependencies and
76
84
  fails if a committed number differs from what the code produces.
77
85
 
78
86
  ## 4,873 real experiments, audited
79
87
 
80
- [`examples/upworthy/`](examples/upworthy/) runs abkit's aggregate checks over
88
+ [`examples/upworthy/`](https://github.com/mohammadi-hadi/abkit/tree/main/examples/upworthy/) runs abkit's aggregate checks over
81
89
  the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038/s41597-021-00934-7)
82
90
  — 4,873 real headline A/B tests from 2013–2015. Committed results:
83
- [examples/upworthy/results/upworthy.md](examples/upworthy/results/upworthy.md).
91
+ [examples/upworthy/results/upworthy.md](https://github.com/mohammadi-hadi/abkit/blob/main/examples/upworthy/results/upworthy.md).
84
92
 
85
- - **16% of real tests fail the sample-ratio check** at p < 0.001 against the
93
+ - **15% of real tests fail the sample-ratio check** at p < 0.001 against the
86
94
  platform's equal-allocation design — and not because packages were added
87
95
  mid-test; the rate is, if anything, higher in tests whose variants all
88
96
  launched together. Those readouts carried unknown exposure bias.
@@ -96,7 +104,7 @@ the exploratory split of the [Upworthy Research Archive](https://doi.org/10.1038
96
104
  ## Install
97
105
 
98
106
  ```
99
- pip install git+https://github.com/mohammadi-hadi/abkit
107
+ pip install abkit
100
108
  ```
101
109
 
102
110
  Python 3.11+. Runtime dependencies: numpy, pydantic, matplotlib.
@@ -184,7 +192,7 @@ experiment.
184
192
  ±0.10 — and flag only when the whole 95% bootstrap CI leaves it. Exact
185
193
  procedures use the field's conventional thresholds: SRM at p < 0.001,
186
194
  contamination at zero tolerance. All constants live in one place
187
- ([`probes.py`](src/abkit/probes.py)) and are easy to disagree with.
195
+ ([`probes.py`](https://github.com/mohammadi-hadi/abkit/blob/main/src/abkit/probes.py)) and are easy to disagree with.
188
196
  - **The stopping rule is audited, not the final p.** A peeked experiment's
189
197
  final test statistic looks ordinary; what is broken is the procedure that
190
198
  produced it. The peeking probe Monte-Carlos the analyst's actual look
@@ -237,22 +245,26 @@ The real-data case study uses the Upworthy Research Archive
237
245
  abkit's contribution is packaging the checks as one auditable tool with
238
246
  uncertainty on every number and validation by implantation.
239
247
 
240
- Companion projects: [judgekit](https://github.com/mohammadi-hadi/judgekit)
241
- applies the same audit-before-you-trust pattern to LLM judges;
242
- [trajectory-judge](https://github.com/mohammadi-hadi/trajectory-judge)
243
- measures what outcome-only judges miss on agent trajectories.
248
+ ## Part of evalstack
249
+
250
+ [evalstack](https://github.com/mohammadi-hadi/evalstack) is the map of these
251
+ eleven evaluation tools: what each one measures, what it found on real data,
252
+ and the two chains that run end to end.
244
253
 
245
254
  ## Citation
246
255
 
247
- If abkit is useful in your work, please cite it (see
248
- [CITATION.cff](CITATION.cff)):
256
+ Releases are archived on Zenodo. Cite the concept DOI
257
+ [10.5281/zenodo.21809478](https://doi.org/10.5281/zenodo.21809478), which always
258
+ resolves to the latest version; structured metadata is in
259
+ [CITATION.cff](https://github.com/mohammadi-hadi/abkit/blob/main/CITATION.cff).
249
260
 
250
261
  ```bibtex
251
262
  @software{mohammadi_abkit,
252
263
  author = {Mohammadi, Hadi},
253
264
  title = {abkit: audit an A/B-test readout before you ship the decision},
254
265
  url = {https://github.com/mohammadi-hadi/abkit},
255
- version = {0.1.0},
266
+ doi = {10.5281/zenodo.21809478},
267
+ version = {0.1.2},
256
268
  year = {2026}
257
269
  }
258
270
  ```
{abkit-0.1.0 → abkit-0.1.2}/LICENSE RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/setup.cfg RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/audit.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/bootstrap.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/demo.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/io.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/probes.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/py.typed RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/report.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/result.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/schema.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/stats.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/src/abkit/synthetic.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_bootstrap.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_cli.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_probes.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_result.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_schema.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_stats.py RENAMED
File without changes
{abkit-0.1.0 → abkit-0.1.2}/tests/test_synthetic.py RENAMED
File without changes