eval-builder 0.1.3__tar.gz → 0.1.4__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 (120) hide show
  1. {eval_builder-0.1.3 → eval_builder-0.1.4}/CHANGELOG.md +6 -0
  2. {eval_builder-0.1.3 → eval_builder-0.1.4}/PKG-INFO +2 -2
  3. {eval_builder-0.1.3 → eval_builder-0.1.4}/pyproject.toml +2 -2
  4. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/__init__.py +1 -1
  5. {eval_builder-0.1.3 → eval_builder-0.1.4}/uv.lock +2 -2
  6. {eval_builder-0.1.3 → eval_builder-0.1.4}/.github/workflows/ci.yml +0 -0
  7. {eval_builder-0.1.3 → eval_builder-0.1.4}/.github/workflows/release.yml +0 -0
  8. {eval_builder-0.1.3 → eval_builder-0.1.4}/.gitignore +0 -0
  9. {eval_builder-0.1.3 → eval_builder-0.1.4}/AGENTS.md +0 -0
  10. {eval_builder-0.1.3 → eval_builder-0.1.4}/CONTRIBUTING.md +0 -0
  11. {eval_builder-0.1.3 → eval_builder-0.1.4}/LICENSE +0 -0
  12. {eval_builder-0.1.3 → eval_builder-0.1.4}/README.md +0 -0
  13. {eval_builder-0.1.3 → eval_builder-0.1.4}/SECURITY.md +0 -0
  14. {eval_builder-0.1.3 → eval_builder-0.1.4}/docs/demo.gif +0 -0
  15. {eval_builder-0.1.3 → eval_builder-0.1.4}/docs/demo.tape +0 -0
  16. {eval_builder-0.1.3 → eval_builder-0.1.4}/docs/label-sheet.png +0 -0
  17. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/judges/control_judges.py +0 -0
  18. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/judges/ollama_judge.py +0 -0
  19. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/README.md +0 -0
  20. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/expected_behaviors.yaml +0 -0
  21. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/fill_suite.py +0 -0
  22. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/cases.yaml +0 -0
  23. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/judge_check.json +0 -0
  24. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/judge_run_log.json +0 -0
  25. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/judgments.jsonl +0 -0
  26. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/labels.jsonl +0 -0
  27. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/report.json +0 -0
  28. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/report.md +0 -0
  29. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/judges/rubric.yaml +0 -0
  30. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/prepare.py +0 -0
  31. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/run.sh +0 -0
  32. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/cases.yaml +0 -0
  33. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/deepeval/dataset.json +0 -0
  34. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/deepeval/test_eval_builder.py +0 -0
  35. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/inspect/dataset.jsonl +0 -0
  36. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/inspect/task.py +0 -0
  37. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/jsonl/cases.jsonl +0 -0
  38. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/manifest.json +0 -0
  39. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/exports/promptfoo/promptfooconfig.yaml +0 -0
  40. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/ingest.json +0 -0
  41. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/judge_check.json +0 -0
  42. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/judge_run_log.json +0 -0
  43. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/judgments.jsonl +0 -0
  44. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/report.json +0 -0
  45. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/report.md +0 -0
  46. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/rubric.yaml +0 -0
  47. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/suite/selection.json +0 -0
  48. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/mt-bench/verify_exports.sh +0 -0
  49. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/README.md +0 -0
  50. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/decisions.json +0 -0
  51. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/drive_sheet.py +0 -0
  52. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/cases.yaml +0 -0
  53. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/deepeval/dataset.json +0 -0
  54. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/deepeval/test_eval_builder.py +0 -0
  55. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/inspect/dataset.jsonl +0 -0
  56. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/inspect/task.py +0 -0
  57. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/jsonl/cases.jsonl +0 -0
  58. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/manifest.json +0 -0
  59. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/exports/promptfoo/promptfooconfig.yaml +0 -0
  60. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/ingest.json +0 -0
  61. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/judge_check.json +0 -0
  62. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/judge_run_log.json +0 -0
  63. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/judgments.jsonl +0 -0
  64. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/label_plan.json +0 -0
  65. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/label_sheet.csv +0 -0
  66. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/label_sheet.html +0 -0
  67. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/labels.jsonl +0 -0
  68. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/report.json +0 -0
  69. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/report.md +0 -0
  70. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/rubric.yaml +0 -0
  71. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/selection.json +0 -0
  72. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/evalset/traces.jsonl +0 -0
  73. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/fill.py +0 -0
  74. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/run.sh +0 -0
  75. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/sheet-run/drive_log.txt +0 -0
  76. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/sheet-run/labels.jsonl +0 -0
  77. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/sheet-run/sheet-export.png +0 -0
  78. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/sheet-run/sheet-first-case.png +0 -0
  79. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample-labeling/sheet-run/sheet-phone.png +0 -0
  80. {eval_builder-0.1.3 → eval_builder-0.1.4}/examples/sample_logs.jsonl +0 -0
  81. {eval_builder-0.1.3 → eval_builder-0.1.4}/skills/eval-builder/SKILL.md +0 -0
  82. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/balance.py +0 -0
  83. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/cli.py +0 -0
  84. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/draft.py +0 -0
  85. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/export.py +0 -0
  86. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/ingest/__init__.py +0 -0
  87. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/ingest/formats.py +0 -0
  88. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/io.py +0 -0
  89. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/judge/__init__.py +0 -0
  90. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/judge/check.py +0 -0
  91. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/judge/plan.py +0 -0
  92. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/judge/run.py +0 -0
  93. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/judge/stats.py +0 -0
  94. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/label.py +0 -0
  95. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/label_sheet.py +0 -0
  96. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/mcp_server.py +0 -0
  97. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/redact.py +0 -0
  98. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/report.py +0 -0
  99. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/schema.py +0 -0
  100. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/select.py +0 -0
  101. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/setup_agents.py +0 -0
  102. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/status.py +0 -0
  103. {eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/workspace.py +0 -0
  104. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/conftest.py +0 -0
  105. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/fixtures/anthropic_messages.json +0 -0
  106. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/fixtures/generic.jsonl +0 -0
  107. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/fixtures/langfuse_export.json +0 -0
  108. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/fixtures/openai_chat.jsonl +0 -0
  109. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/fixtures/otel_genai.json +0 -0
  110. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_cli_mcp.py +0 -0
  111. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_draft.py +0 -0
  112. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_export.py +0 -0
  113. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_ingest.py +0 -0
  114. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_judge_check.py +0 -0
  115. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_judge_run.py +0 -0
  116. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_judge_stats.py +0 -0
  117. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_label.py +0 -0
  118. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_report.py +0 -0
  119. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_select.py +0 -0
  120. {eval_builder-0.1.3 → eval_builder-0.1.4}/tests/test_setup.py +0 -0
{eval_builder-0.1.3 → eval_builder-0.1.4}/CHANGELOG.md RENAMED
@@ -1,5 +1,11 @@
1
1
  # Changelog
2
2
 
3
+ ## 0.1.4
4
+
5
+ ### Fixed
6
+
7
+ - Require mcp 1.30 or newer. The old floor (`mcp>=1.2`) allowed releases that no longer import with current pydantic; the MCP server's tests pass on 1.30.0 and 2.3.0.
8
+
3
9
  ## 0.1.3
4
10
 
5
11
  ### Fixed
{eval_builder-0.1.3 → eval_builder-0.1.4}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: eval-builder
3
- Version: 0.1.3
3
+ Version: 0.1.4
4
4
  Summary: Turn real LLM app logs into an eval suite and measure which LLM judges you can trust.
5
5
  Author: Abel Yagubyan
6
6
  License-Expression: MIT
@@ -10,7 +10,7 @@ Classifier: Operating System :: OS Independent
10
10
  Classifier: Programming Language :: Python :: 3
11
11
  Classifier: Topic :: Software Development :: Testing
12
12
  Requires-Python: >=3.11
13
- Requires-Dist: mcp>=1.2
13
+ Requires-Dist: mcp>=1.30
14
14
  Requires-Dist: numpy>=1.24
15
15
  Requires-Dist: pyyaml>=6.0
16
16
  Requires-Dist: scikit-learn>=1.3
{eval_builder-0.1.3 → eval_builder-0.1.4}/pyproject.toml RENAMED
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "eval-builder"
3
- version = "0.1.3"
3
+ version = "0.1.4"
4
4
  description = "Turn real LLM app logs into an eval suite and measure which LLM judges you can trust."
5
5
  readme = "README.md"
6
6
  license = "MIT"
@@ -17,7 +17,7 @@ dependencies = [
17
17
  "numpy>=1.24",
18
18
  "scikit-learn>=1.3",
19
19
  "pyyaml>=6.0",
20
- "mcp>=1.2",
20
+ "mcp>=1.30",
21
21
  "scipy>=1.10",
22
22
  ]
23
23
 
{eval_builder-0.1.3 → eval_builder-0.1.4}/src/eval_builder/__init__.py RENAMED
@@ -1,3 +1,3 @@
1
1
  """eval-builder: turn real LLM app logs into an eval suite and check which judges you can trust."""
2
2
 
3
- __version__ = "0.1.3"
3
+ __version__ = "0.1.4"
{eval_builder-0.1.3 → eval_builder-0.1.4}/uv.lock RENAMED
@@ -299,7 +299,7 @@ wheels = [
299
299
 
300
300
  [[package]]
301
301
  name = "eval-builder"
302
- version = "0.1.3"
302
+ version = "0.1.4"
303
303
  source = { editable = "." }
304
304
  dependencies = [
305
305
  { name = "mcp" },
@@ -321,7 +321,7 @@ dev = [
321
321
 
322
322
  [package.metadata]
323
323
  requires-dist = [
324
- { name = "mcp", specifier = ">=1.2" },
324
+ { name = "mcp", specifier = ">=1.30" },
325
325
  { name = "numpy", specifier = ">=1.24" },
326
326
  { name = "pyyaml", specifier = ">=6.0" },
327
327
  { name = "scikit-learn", specifier = ">=1.3" },
{eval_builder-0.1.3 → eval_builder-0.1.4}/.gitignore RENAMED
File without changes
{eval_builder-0.1.3 → eval_builder-0.1.4}/AGENTS.md RENAMED
File without changes
{eval_builder-0.1.3 → eval_builder-0.1.4}/LICENSE RENAMED
File without changes
{eval_builder-0.1.3 → eval_builder-0.1.4}/README.md RENAMED
File without changes
{eval_builder-0.1.3 → eval_builder-0.1.4}/SECURITY.md RENAMED
File without changes
{eval_builder-0.1.3 → eval_builder-0.1.4}/docs/demo.gif RENAMED
File without changes