turambar928 commited on
Commit
01d645f
·
verified ·
1 Parent(s): 49ceb7e

Upload folder using huggingface_hub (part 336)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. 20260904-005134-science-environment-e187c8/task.toml +45 -0
  2. 20260904-005134-science-environment-e187c8/tests/Dockerfile +6 -0
  3. 20260904-005134-science-environment-e187c8/tests/evaluator_runner.py +92 -0
  4. 20260904-005134-science-environment-e187c8/tests/reference/__pycache__/evaluator.cpython-312.pyc +0 -0
  5. 20260904-005134-science-environment-e187c8/tests/reference/evaluator.py +114 -0
  6. 20260904-005134-science-environment-e187c8/tests/reference/ground_truth/clean_series.csv +19 -0
  7. 20260904-005134-science-environment-e187c8/tests/reference/ground_truth/decadal_statistics.csv +4 -0
  8. 20260904-005134-science-environment-e187c8/tests/reference/ground_truth/summary.json +23 -0
  9. 20260904-005134-science-environment-e187c8/tests/reference/ground_truth/threshold_years.csv +19 -0
  10. 20260904-005134-science-environment-e187c8/tests/reference/internal/workflow_adapters_runtime.py +679 -0
  11. 20260904-005134-science-environment-e187c8/tests/reference/internal/workflow_runtime.py +5 -0
  12. 20260904-005134-science-environment-e187c8/tests/reference/reference_manifest.json +34 -0
  13. 20260904-005134-science-environment-e187c8/tests/reference/reference_solution.py +20 -0
  14. 20260904-005134-science-environment-e187c8/tests/test.sh +29 -0
  15. 20260904-005149-science-environment-cee684/.DS_Store +3 -0
  16. 20260904-005149-science-environment-cee684/environment/Dockerfile +12 -0
  17. 20260904-005149-science-environment-cee684/environment/public_data/README.md +11 -0
  18. 20260904-005149-science-environment-cee684/environment/public_data/input/README.md +6 -0
  19. 20260904-005149-science-environment-cee684/environment/public_data/input/analysis_spec.json +76 -0
  20. 20260904-005149-science-environment-cee684/environment/public_data/input/source_context_windows.jsonl +0 -0
  21. 20260904-005149-science-environment-cee684/environment/public_data/input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv +0 -0
  22. 20260904-005149-science-environment-cee684/environment/public_data/main.py +180 -0
  23. 20260904-005149-science-environment-cee684/environment/public_data/software/validate_outputs.py +41 -0
  24. 20260904-005149-science-environment-cee684/environment/public_data/task_card.json +92 -0
  25. 20260904-005149-science-environment-cee684/instruction.md +33 -0
  26. 20260904-005149-science-environment-cee684/solution/internal/workflow_adapters_runtime.py +679 -0
  27. 20260904-005149-science-environment-cee684/solution/internal/workflow_runtime.py +5 -0
  28. 20260904-005149-science-environment-cee684/solution/reference_solution.py +20 -0
  29. 20260904-005149-science-environment-cee684/solution/solve.sh +7 -0
  30. 20260904-005149-science-environment-cee684/task.toml +45 -0
  31. 20260904-005149-science-environment-cee684/tests/Dockerfile +6 -0
  32. 20260904-005149-science-environment-cee684/tests/evaluator_runner.py +92 -0
  33. 20260904-005149-science-environment-cee684/tests/reference/__pycache__/evaluator.cpython-312.pyc +0 -0
  34. 20260904-005149-science-environment-cee684/tests/reference/evaluator.py +114 -0
  35. 20260904-005149-science-environment-cee684/tests/reference/ground_truth/clean_series.csv +20 -0
  36. 20260904-005149-science-environment-cee684/tests/reference/ground_truth/decadal_statistics.csv +3 -0
  37. 20260904-005149-science-environment-cee684/tests/reference/ground_truth/summary.json +23 -0
  38. 20260904-005149-science-environment-cee684/tests/reference/ground_truth/threshold_years.csv +18 -0
  39. 20260904-005149-science-environment-cee684/tests/reference/internal/workflow_adapters_runtime.py +679 -0
  40. 20260904-005149-science-environment-cee684/tests/reference/internal/workflow_runtime.py +5 -0
  41. 20260904-005149-science-environment-cee684/tests/reference/reference_manifest.json +34 -0
  42. 20260904-005149-science-environment-cee684/tests/reference/reference_solution.py +20 -0
  43. 20260904-005149-science-environment-cee684/tests/test.sh +29 -0
  44. 20260904-005151-science-environment-7e595f/.DS_Store +3 -0
  45. 20260904-005151-science-environment-7e595f/environment/Dockerfile +12 -0
  46. 20260904-005151-science-environment-7e595f/environment/public_data/README.md +11 -0
  47. 20260904-005151-science-environment-7e595f/environment/public_data/input/README.md +6 -0
  48. 20260904-005151-science-environment-7e595f/environment/public_data/input/analysis_spec.json +76 -0
  49. 20260904-005151-science-environment-7e595f/environment/public_data/input/source_context_windows.jsonl +0 -0
  50. 20260904-005151-science-environment-7e595f/environment/public_data/input/usgs_fdsn_eq_caribbean_1987_1989_m5p5_depth_mean_quota_repair_0779.csv +0 -0
20260904-005134-science-environment-e187c8/task.toml ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version = "1.3"
2
+ artifacts = ["/workspace/output/summary.json", "/workspace/output/decadal_statistics.csv", "/workspace/output/threshold_years.csv", "/workspace/output/clean_series.csv"]
3
+
4
+ [task]
5
+ name = "ale-cli/20260904-005134-science-environment-e187c8"
6
+ description = "Audit the real USGS FDSN Andes subduction zone earthquake catalog for 2008-2025 (events with magnitude >= 5.5) and produce reproducible annual annual maximum magnitude statistics, decadal aggregates, threshold-year checks, and a cleaned local series."
7
+ keywords = ["agents-last-exam", "ale-cli", "csv-annual-series-audit", "deterministic-evaluator", "linux-cli", "science-environment"]
8
+
9
+ [metadata]
10
+ benchmark = "ale_cli"
11
+ source_format = "ale_cli_official_style"
12
+ converter = "ale_cli_to_harbor"
13
+ source_task_key = "20260904-005134_science_environment_e187c8"
14
+ task_id = "science_environment/usgs_andes_annual_earthquake_maximum_audit"
15
+ domain = "science_environment"
16
+ task = "usgs_andes_annual_earthquake_maximum_audit"
17
+ variant = "base"
18
+ oracle_available = true
19
+ scoring_method = "deterministic_hidden_evaluator"
20
+ required_outputs = ["summary.json", "decadal_statistics.csv", "threshold_years.csv", "clean_series.csv"]
21
+
22
+ [agent]
23
+ timeout_sec = 1800
24
+
25
+ [verifier]
26
+ timeout_sec = 600
27
+ environment_mode = "separate"
28
+
29
+ [verifier.environment]
30
+ build_timeout_sec = 900.0
31
+ network_mode = "public"
32
+ os = "linux"
33
+ cpus = 1
34
+ memory_mb = 2048
35
+ storage_mb = 4096
36
+ gpus = 0
37
+
38
+ [environment]
39
+ build_timeout_sec = 1200.0
40
+ network_mode = "public"
41
+ os = "linux"
42
+ cpus = 2
43
+ memory_mb = 4096
44
+ storage_mb = 10240
45
+ gpus = 0
20260904-005134-science-environment-e187c8/tests/Dockerfile ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ FROM python:3.12-slim
2
+
3
+ WORKDIR /workspace
4
+
5
+ COPY . /tests
6
+ RUN chmod +x /tests/test.sh
20260904-005134-science-environment-e187c8/tests/evaluator_runner.py ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ from __future__ import annotations
3
+
4
+ import argparse
5
+ import importlib.util
6
+ import json
7
+ from pathlib import Path
8
+ from typing import Any
9
+
10
+
11
+ def read_json(path: Path) -> Any:
12
+ return json.loads(path.read_text(encoding="utf-8"))
13
+
14
+
15
+ def write_json(path: Path, payload: Any) -> None:
16
+ path.parent.mkdir(parents=True, exist_ok=True)
17
+ path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
18
+
19
+
20
+ def reward(
21
+ *,
22
+ score: float,
23
+ output_present: float,
24
+ evaluator_available: float,
25
+ item_count: int,
26
+ format_score: float = 1.0,
27
+ ) -> dict[str, float]:
28
+ clipped = max(0.0, min(1.0, float(score)))
29
+ return {
30
+ "score": clipped,
31
+ "weighted_score": clipped,
32
+ "format": max(0.0, min(1.0, float(format_score))),
33
+ "output_present": max(0.0, min(1.0, float(output_present))),
34
+ "evaluator_available": max(0.0, min(1.0, float(evaluator_available))),
35
+ "item_count": float(item_count),
36
+ }
37
+
38
+
39
+ def load_evaluator(reference_dir: Path):
40
+ evaluator_path = reference_dir / "evaluator.py"
41
+ spec = importlib.util.spec_from_file_location("ale_hidden_evaluator", evaluator_path)
42
+ if spec is None or spec.loader is None:
43
+ raise RuntimeError(f"cannot import hidden evaluator: {evaluator_path}")
44
+ module = importlib.util.module_from_spec(spec)
45
+ spec.loader.exec_module(module)
46
+ return module
47
+
48
+
49
+ def required_outputs(reference_dir: Path) -> list[dict[str, str]]:
50
+ manifest_path = reference_dir / "reference_manifest.json"
51
+ if not manifest_path.is_file():
52
+ return []
53
+ manifest = read_json(manifest_path)
54
+ rows = manifest.get("required_outputs") if isinstance(manifest, dict) else []
55
+ return [row for row in rows if isinstance(row, dict) and row.get("path")]
56
+
57
+
58
+ def main() -> int:
59
+ parser = argparse.ArgumentParser(description="Run a hidden ALE CLI evaluator in Harbor.")
60
+ parser.add_argument("--output-dir", type=Path, required=True)
61
+ parser.add_argument("--reference-dir", type=Path, required=True)
62
+ parser.add_argument("--reward-json", type=Path, required=True)
63
+ parser.add_argument("--details-json", type=Path, required=True)
64
+ args = parser.parse_args()
65
+
66
+ required = required_outputs(args.reference_dir)
67
+ present = [row for row in required if (args.output_dir / str(row["path"])).is_file()]
68
+ item_count = len(required)
69
+ if not required:
70
+ write_json(args.reward_json, reward(score=0.0, output_present=0.0, evaluator_available=0.0, item_count=0, format_score=0.0))
71
+ write_json(args.details_json, {"ok": False, "error": "reference_manifest.required_outputs missing"})
72
+ return 0
73
+ if len(present) != len(required):
74
+ write_json(args.reward_json, reward(score=0.0, output_present=0.0, evaluator_available=1.0, item_count=item_count))
75
+ write_json(args.details_json, {"ok": True, "reason": "missing required outputs", "present": [row["path"] for row in present]})
76
+ return 0
77
+ try:
78
+ module = load_evaluator(args.reference_dir)
79
+ result = module.evaluate(args.output_dir, args.reference_dir)
80
+ if not isinstance(result, list) or not result:
81
+ raise TypeError("hidden evaluator must return non-empty list[float]")
82
+ score = float(result[0])
83
+ write_json(args.reward_json, reward(score=score, output_present=1.0, evaluator_available=1.0, item_count=item_count))
84
+ write_json(args.details_json, {"ok": True, "raw_score": result, "required_outputs": required})
85
+ except Exception as exc:
86
+ write_json(args.reward_json, reward(score=0.0, output_present=1.0, evaluator_available=0.0, item_count=item_count))
87
+ write_json(args.details_json, {"ok": False, "error": f"{type(exc).__name__}: {exc}", "required_outputs": required})
88
+ return 0
89
+
90
+
91
+ if __name__ == "__main__":
92
+ raise SystemExit(main())
20260904-005134-science-environment-e187c8/tests/reference/__pycache__/evaluator.cpython-312.pyc ADDED
Binary file (6.22 kB). View file
 
20260904-005134-science-environment-e187c8/tests/reference/evaluator.py ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import csv
4
+ import json
5
+ import math
6
+ from pathlib import Path
7
+ from typing import Any
8
+
9
+
10
+ def _load_json(path: Path) -> Any:
11
+ with path.open("r", encoding="utf-8") as handle:
12
+ return json.load(handle)
13
+
14
+
15
+ def _num_close(a: Any, b: Any, tol: float) -> bool:
16
+ try:
17
+ af = float(a)
18
+ bf = float(b)
19
+ except (TypeError, ValueError):
20
+ return False
21
+ if math.isnan(af) and math.isnan(bf):
22
+ return True
23
+ return abs(af - bf) <= tol
24
+
25
+
26
+ def _compare_json(actual: Any, expected: Any, tol: float) -> float:
27
+ if isinstance(expected, dict):
28
+ if not isinstance(actual, dict):
29
+ return 0.0
30
+ if not expected:
31
+ return 1.0
32
+ scores = []
33
+ for key, exp_value in expected.items():
34
+ if key not in actual:
35
+ scores.append(0.0)
36
+ else:
37
+ scores.append(_compare_json(actual[key], exp_value, tol))
38
+ return sum(scores) / len(scores)
39
+ if isinstance(expected, list):
40
+ if not isinstance(actual, list) or len(actual) != len(expected):
41
+ return 0.0
42
+ if not expected:
43
+ return 1.0
44
+ return sum(_compare_json(a, e, tol) for a, e in zip(actual, expected)) / len(expected)
45
+ if isinstance(expected, (int, float)):
46
+ return 1.0 if _num_close(actual, expected, tol) else 0.0
47
+ return 1.0 if actual == expected else 0.0
48
+
49
+
50
+ def _read_csv(path: Path) -> list[dict[str, str]]:
51
+ with path.open("r", encoding="utf-8", newline="") as handle:
52
+ return list(csv.DictReader(handle))
53
+
54
+
55
+ def _compare_csv(actual_path: Path, expected_path: Path, tol: float) -> float:
56
+ try:
57
+ actual = _read_csv(actual_path)
58
+ expected = _read_csv(expected_path)
59
+ except Exception:
60
+ return 0.0
61
+ if len(actual) != len(expected):
62
+ return 0.0
63
+ if not expected:
64
+ return 1.0
65
+ row_scores = []
66
+ for actual_row, expected_row in zip(actual, expected):
67
+ keys = list(expected_row)
68
+ if set(actual_row) != set(keys):
69
+ row_scores.append(0.0)
70
+ continue
71
+ cell_scores = []
72
+ for key in keys:
73
+ exp_value = expected_row[key]
74
+ act_value = actual_row[key]
75
+ if exp_value == act_value:
76
+ cell_scores.append(1.0)
77
+ elif _num_close(act_value, exp_value, tol):
78
+ cell_scores.append(1.0)
79
+ else:
80
+ cell_scores.append(0.0)
81
+ row_scores.append(sum(cell_scores) / len(cell_scores) if cell_scores else 1.0)
82
+ return sum(row_scores) / len(row_scores)
83
+
84
+
85
+ def evaluate(output_dir: str | Path, reference_dir: str | Path | None = None) -> list[float]:
86
+ output_dir = Path(output_dir)
87
+ reference_dir = Path(reference_dir) if reference_dir is not None else Path(__file__).resolve().parent
88
+ manifest = _load_json(reference_dir / "reference_manifest.json")
89
+ tolerance = float(manifest.get("numeric_tolerance", 1e-9))
90
+ required = manifest.get("required_outputs") or []
91
+ if not required:
92
+ return [0.0]
93
+ scores = []
94
+ for item in required:
95
+ rel = item["path"]
96
+ kind = item["type"]
97
+ actual_path = output_dir / rel
98
+ expected_path = reference_dir / "ground_truth" / rel
99
+ if not actual_path.is_file() or not expected_path.is_file():
100
+ scores.append(0.0)
101
+ continue
102
+ try:
103
+ if kind == "json":
104
+ scores.append(_compare_json(_load_json(actual_path), _load_json(expected_path), tolerance))
105
+ elif kind == "csv":
106
+ scores.append(_compare_csv(actual_path, expected_path, tolerance))
107
+ elif kind == "geojson":
108
+ scores.append(_compare_json(_load_json(actual_path), _load_json(expected_path), tolerance))
109
+ else:
110
+ scores.append(1.0 if actual_path.read_bytes() == expected_path.read_bytes() else 0.0)
111
+ except Exception:
112
+ scores.append(0.0)
113
+ score = max(0.0, min(1.0, sum(scores) / len(scores)))
114
+ return [score]
20260904-005134-science-environment-e187c8/tests/reference/ground_truth/clean_series.csv ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Year,Departure from Average
2
+ 2008,6.400000
3
+ 2009,6.500000
4
+ 2010,8.800000
5
+ 2011,7.200000
6
+ 2012,7.300000
7
+ 2013,7.100000
8
+ 2014,8.200000
9
+ 2015,8.300000
10
+ 2016,7.800000
11
+ 2017,6.900000
12
+ 2018,7.100000
13
+ 2019,8.000000
14
+ 2020,6.800000
15
+ 2021,7.500000
16
+ 2022,7.200000
17
+ 2023,6.800000
18
+ 2024,7.400000
19
+ 2025,6.400000
20260904-005134-science-environment-e187c8/tests/reference/ground_truth/decadal_statistics.csv ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ decade_start,start_year,end_year,count,mean_departure,median_departure,min_departure,max_departure,sample_std_departure
2
+ 2000,2008,2009,2,6.450000,6.450000,6.400000,6.500000,0.070711
3
+ 2010,2010,2019,10,7.670000,7.550000,6.900000,8.800000,0.639531
4
+ 2020,2020,2025,6,7.016667,7.000000,6.400000,7.500000,0.421505
20260904-005134-science-environment-e187c8/tests/reference/ground_truth/summary.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "series_name": "USGS FDSN Andes subduction zone 2008-2025 earthquake annual maximum magnitude (min mag 4.5)",
3
+ "row_count": 18,
4
+ "year_range": {
5
+ "start": 2008,
6
+ "end": 2025
7
+ },
8
+ "value": {
9
+ "min": 6.4,
10
+ "min_year": 2008,
11
+ "max": 8.8,
12
+ "max_year": 2010,
13
+ "mean": 7.316667,
14
+ "median": 7.2,
15
+ "sample_std": 0.680182
16
+ },
17
+ "threshold": 6.0,
18
+ "years_at_or_above_threshold": 18,
19
+ "first_threshold_year": 2008,
20
+ "last_threshold_year": 2025,
21
+ "year_aggregation": "max",
22
+ "source_row_count": 9821
23
+ }
20260904-005134-science-environment-e187c8/tests/reference/ground_truth/threshold_years.csv ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ rank,year,departure,threshold
2
+ 1,2008,6.400000,6.000000
3
+ 2,2009,6.500000,6.000000
4
+ 3,2010,8.800000,6.000000
5
+ 4,2011,7.200000,6.000000
6
+ 5,2012,7.300000,6.000000
7
+ 6,2013,7.100000,6.000000
8
+ 7,2014,8.200000,6.000000
9
+ 8,2015,8.300000,6.000000
10
+ 9,2016,7.800000,6.000000
11
+ 10,2017,6.900000,6.000000
12
+ 11,2018,7.100000,6.000000
13
+ 12,2019,8.000000,6.000000
14
+ 13,2020,6.800000,6.000000
15
+ 14,2021,7.500000,6.000000
16
+ 15,2022,7.200000,6.000000
17
+ 16,2023,6.800000,6.000000
18
+ 17,2024,7.400000,6.000000
19
+ 18,2025,6.400000,6.000000
20260904-005134-science-environment-e187c8/tests/reference/internal/workflow_adapters_runtime.py ADDED
@@ -0,0 +1,679 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import csv
4
+ import datetime as dt
5
+ import html
6
+ import json
7
+ import math
8
+ import re
9
+ import shutil
10
+ import statistics
11
+ from pathlib import Path
12
+ from typing import Any
13
+
14
+
15
+ def _read_json(path: Path) -> Any:
16
+ with path.open("r", encoding="utf-8") as handle:
17
+ return json.load(handle)
18
+
19
+
20
+ def _write_json(path: Path, payload: Any) -> None:
21
+ path.parent.mkdir(parents=True, exist_ok=True)
22
+ with path.open("w", encoding="utf-8") as handle:
23
+ json.dump(payload, handle, ensure_ascii=False, indent=2)
24
+ handle.write("\n")
25
+
26
+ def _stable_region(place: str) -> str:
27
+ if not place:
28
+ return "unknown"
29
+ if "," in place:
30
+ return place.split(",")[-1].strip() or place.strip()
31
+ parts = place.split()
32
+ return " ".join(parts[-2:]) if len(parts) >= 2 else place.strip()
33
+
34
+
35
+ def _mag_bin(value: float | None) -> str:
36
+ if value is None or math.isnan(value):
37
+ return "unknown"
38
+ lower = math.floor(value)
39
+ upper = lower + 1
40
+ return f"{lower:.0f}-{upper:.0f}"
41
+
42
+
43
+ def _depth_bin(value: float | None) -> str:
44
+ if value is None or math.isnan(value):
45
+ return "unknown"
46
+ if value < 0:
47
+ return "<0"
48
+ if value < 10:
49
+ return "0-10"
50
+ if value < 30:
51
+ return "10-30"
52
+ if value < 70:
53
+ return "30-70"
54
+ if value < 300:
55
+ return "70-300"
56
+ return "300+"
57
+
58
+
59
+ def _clean_markup_text(raw: str) -> str:
60
+ text = re.sub(r"(?is)<(script|style|noscript)[^>]*>.*?</\1>", " ", raw)
61
+ text = re.sub(r"(?is)<br\s*/?>", "\n", text)
62
+ text = re.sub(r"(?is)</(p|div|li|tr|h[1-6]|pre)>", "\n", text)
63
+ text = re.sub(r"(?is)<[^>]+>", " ", text)
64
+ text = html.unescape(text)
65
+ text = re.sub(r"[ \t]+", " ", text)
66
+ text = re.sub(r"\n\s+", "\n", text)
67
+ return text.strip()
68
+
69
+
70
+ def _heading_slug(value: str) -> str:
71
+ slug = re.sub(r"[^a-z0-9]+", "_", value.lower()).strip("_")
72
+ return slug[:80] or "section"
73
+
74
+
75
+ def _extract_headings(raw: str) -> list[dict[str, str]]:
76
+ headings: list[dict[str, str]] = []
77
+ for match in re.finditer(r"(?is)<h([1-6])[^>]*>(.*?)</h\1>", raw):
78
+ text = _clean_markup_text(match.group(2))
79
+ if text:
80
+ headings.append({"level": f"h{match.group(1)}", "text": text})
81
+ if headings:
82
+ return headings
83
+ for line in raw.splitlines():
84
+ match = re.match(r"^(#{1,6})\s+(.+?)\s*$", line)
85
+ if match:
86
+ headings.append({"level": f"h{len(match.group(1))}", "text": match.group(2).strip()})
87
+ return headings
88
+
89
+
90
+ def _extract_code_blocks(raw: str) -> list[str]:
91
+ blocks: list[str] = []
92
+ for match in re.finditer(r"(?is)<pre[^>]*>(.*?)</pre>", raw):
93
+ text = _clean_markup_text(match.group(1))
94
+ if text:
95
+ blocks.append(text)
96
+ for match in re.finditer(r"(?is)<code[^>]*>(.*?)</code>", raw):
97
+ text = _clean_markup_text(match.group(1))
98
+ if text and text not in blocks:
99
+ blocks.append(text)
100
+ for match in re.finditer(r"(?s)```(?:[A-Za-z0-9_-]+)?\n(.*?)```", raw):
101
+ text = html.unescape(match.group(1).strip())
102
+ if text and text not in blocks:
103
+ blocks.append(text)
104
+ return blocks
105
+
106
+
107
+ def _window_rows(clean_text: str, *, width: int = 40, stride: int = 20) -> list[dict[str, Any]]:
108
+ words = re.findall(r"\S+", clean_text)
109
+ rows: list[dict[str, Any]] = []
110
+ if not words:
111
+ return rows
112
+ start = 0
113
+ while start < len(words):
114
+ end = min(start + width, len(words))
115
+ rows.append({
116
+ "window_id": len(rows) + 1,
117
+ "start_word": start,
118
+ "end_word": end,
119
+ "text": " ".join(words[start:end]),
120
+ })
121
+ if end == len(words):
122
+ break
123
+ start += stride
124
+ return rows
125
+
126
+
127
+ def _term_count(text: str, term: str) -> int:
128
+ term = term.strip()
129
+ if not term:
130
+ return 0
131
+ return len(re.findall(re.escape(term), text, flags=re.IGNORECASE))
132
+
133
+
134
+ def _first_window(windows: list[dict[str, Any]], term: str) -> str:
135
+ term_lower = term.lower()
136
+ for row in windows:
137
+ if term_lower in str(row.get("text") or "").lower():
138
+ return str(row.get("window_id") or "")
139
+ return ""
140
+
141
+
142
+ def _write_tutorial_public_inputs(input_dir: Path, source_path: Path, spec_payload: dict[str, Any]) -> None:
143
+ raw = source_path.read_text(encoding="utf-8", errors="replace")
144
+ # Tutorial examples may contain a maintainer's local Windows path. Keep
145
+ # the command semantics while preventing host-specific paths from entering
146
+ # the public task artifact.
147
+ sanitized_raw = re.sub(r"(?i)\b[A-Z]:\\(?:[^\\\r\n<>\"']+\\?)*[^\\\r\n<>\"']*", "./local.db", raw)
148
+ clean_text = _clean_markup_text(sanitized_raw)
149
+ windows = _window_rows(clean_text)
150
+ if len(windows) < 100:
151
+ raise ValueError(f"official tutorial source produced only {len(windows)} text windows; need at least 100")
152
+ required_terms = [str(item) for item in spec_payload.get("required_terms") or []]
153
+ code_terms = [str(item) for item in spec_payload.get("code_terms") or []]
154
+ input_dir.mkdir(parents=True, exist_ok=True)
155
+ destination = input_dir / str(spec_payload["source_file"])
156
+ # Always materialize the sanitized copy, including when the downloader
157
+ # already placed the source at the destination path.
158
+ destination.write_text(sanitized_raw, encoding="utf-8")
159
+ with (input_dir / str(spec_payload["input_file"])).open("w", encoding="utf-8", newline="") as handle:
160
+ fieldnames = ["window_id", "start_word", "end_word", "text", "contains_code_term", "matched_required_terms"]
161
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
162
+ writer.writeheader()
163
+ for row in windows:
164
+ text = str(row["text"])
165
+ matched = [term for term in required_terms if term.lower() in text.lower()]
166
+ contains_code = any(term.lower() in text.lower() for term in code_terms)
167
+ writer.writerow({
168
+ "window_id": row["window_id"],
169
+ "start_word": row["start_word"],
170
+ "end_word": row["end_word"],
171
+ "text": text,
172
+ "contains_code_term": "true" if contains_code else "false",
173
+ "matched_required_terms": "|".join(matched),
174
+ })
175
+ metadata = {
176
+ "software_name": spec_payload.get("software_name"),
177
+ "tutorial_url": spec_payload.get("tutorial_url"),
178
+ "source_urls": spec_payload.get("source_urls"),
179
+ "source_file": spec_payload.get("source_file"),
180
+ "window_file": spec_payload.get("input_file"),
181
+ "window_rule": spec_payload.get("window_rule"),
182
+ "required_terms": spec_payload.get("required_terms"),
183
+ "code_terms": spec_payload.get("code_terms"),
184
+ "note": "Provenance and public extraction rules only; scored aggregate values are intentionally not stored here.",
185
+ }
186
+ _write_json(input_dir / "tutorial_metadata.json", metadata)
187
+
188
+
189
+ def _source_context_text(path: Path) -> str:
190
+ try:
191
+ raw = path.read_text(encoding="utf-8", errors="replace")
192
+ except Exception:
193
+ return f"{path.name}\nsha256={_sha256_file(path)}\nsize_bytes={path.stat().st_size}\n"
194
+ if path.suffix.lower() in {".html", ".htm", ".md", ".txt"}:
195
+ return _clean_markup_text(raw)
196
+ return re.sub(r"\s+", " ", raw).strip()
197
+
198
+
199
+ def _ensure_public_input_footprint(input_dir: Path, source_paths: list[Path]) -> str:
200
+ current = _tree_file_bytes(input_dir)
201
+ if current >= MIN_PUBLIC_TASK_DATA_BYTES:
202
+ return f"public input footprint is {current} bytes; no context expansion needed"
203
+ context_path = input_dir / "source_context_windows.jsonl"
204
+ base = current - (context_path.stat().st_size if context_path.is_file() else 0)
205
+ context_path.unlink(missing_ok=True)
206
+ rows: list[dict[str, Any]] = []
207
+ for source in source_paths:
208
+ if not source.is_file() or source.resolve() == context_path.resolve():
209
+ continue
210
+ text = _source_context_text(source)
211
+ if not text:
212
+ continue
213
+ start = 0
214
+ width = 1800
215
+ stride = 900
216
+ while start < len(text):
217
+ end = min(len(text), start + width)
218
+ rows.append({
219
+ "source_file": source.relative_to(input_dir).as_posix() if source.is_relative_to(input_dir) else source.name,
220
+ "source_sha256": _sha256_file(source),
221
+ "start_char": start,
222
+ "end_char": end,
223
+ "text": text[start:end],
224
+ })
225
+ if end == len(text):
226
+ break
227
+ start += stride
228
+ if not rows:
229
+ raise ValueError("public input footprint expansion has no public source text")
230
+ input_dir.mkdir(parents=True, exist_ok=True)
231
+ rows_written = 0
232
+ target_extra = MIN_PUBLIC_TASK_DATA_BYTES - max(0, base)
233
+ with context_path.open("w", encoding="utf-8") as handle:
234
+ cycle = 0
235
+ while handle.tell() < target_extra or rows_written < 100:
236
+ row = dict(rows[rows_written % len(rows)])
237
+ row["window_id"] = rows_written + 1
238
+ row["cycle"] = cycle
239
+ handle.write(json.dumps(row, ensure_ascii=False) + "\n")
240
+ rows_written += 1
241
+ if rows_written % len(rows) == 0:
242
+ cycle += 1
243
+ handle.flush()
244
+ return f"wrote input/source_context_windows.jsonl with {rows_written} public-source windows and {context_path.stat().st_size} bytes"
245
+
246
+
247
+ def _write_software_manifest(software_dir: Path, spec_payload: dict[str, Any]) -> None:
248
+ if str(spec_payload.get("workflow_kind") or "") != "software_tutorial_doc_audit":
249
+ return
250
+ software_dir.mkdir(parents=True, exist_ok=True)
251
+ software_name = str(spec_payload.get("software_name") or "").strip()
252
+ executables = [str(item).strip() for item in spec_payload.get("executables") or [] if str(item).strip()]
253
+ system_packages = [str(item).strip() for item in spec_payload.get("required_system_packages") or [] if str(item).strip()]
254
+ source_urls = [str(item).strip() for item in spec_payload.get("source_urls") or [] if str(item).strip()]
255
+ _write_json(software_dir / "software_manifest.json", {
256
+ "software_name": software_name,
257
+ "executables": executables,
258
+ "required_system_packages": system_packages,
259
+ "official_sources": source_urls,
260
+ "solve_time_network": "disabled",
261
+ "python_wrapper_only": False,
262
+ "note": "This task audits real official software documentation/source staged offline; Python helpers only validate outputs.",
263
+ })
264
+ with (software_dir / "command_examples.csv").open("w", encoding="utf-8", newline="") as handle:
265
+ writer = csv.DictWriter(handle, fieldnames=["term", "kind", "software_name"])
266
+ writer.writeheader()
267
+ for term in spec_payload.get("code_terms") or []:
268
+ writer.writerow({"term": str(term), "kind": "code_term", "software_name": software_name})
269
+ for exe in executables:
270
+ writer.writerow({"term": exe, "kind": "executable", "software_name": software_name})
271
+
272
+
273
+ def software_tutorial_outputs(input_dir: Path, output_dir: Path) -> None:
274
+ spec = _read_json(input_dir / "analysis_spec.json")
275
+ source_file = str(spec["source_file"])
276
+ source_text = (input_dir / source_file).read_text(encoding="utf-8", errors="replace")
277
+ clean_text = _clean_markup_text(source_text)
278
+ headings = _extract_headings(source_text)
279
+ code_blocks = _extract_code_blocks(source_text)
280
+ window_rows: list[dict[str, str]] = []
281
+ with (input_dir / str(spec["input_file"])).open("r", encoding="utf-8", newline="") as handle:
282
+ window_rows = list(csv.DictReader(handle))
283
+ output_dir.mkdir(parents=True, exist_ok=True)
284
+
285
+ with (output_dir / "section_outline.csv").open("w", encoding="utf-8", newline="") as handle:
286
+ fieldnames = ["rank", "heading_level", "heading_text", "slug"]
287
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
288
+ writer.writeheader()
289
+ for rank, heading in enumerate(headings, start=1):
290
+ writer.writerow({
291
+ "rank": rank,
292
+ "heading_level": heading["level"],
293
+ "heading_text": heading["text"],
294
+ "slug": _heading_slug(heading["text"]),
295
+ })
296
+
297
+ all_terms: list[tuple[str, str]] = []
298
+ for term in spec.get("required_terms") or []:
299
+ all_terms.append((str(term), "required"))
300
+ for term in spec.get("code_terms") or []:
301
+ if str(term) not in {item[0] for item in all_terms}:
302
+ all_terms.append((str(term), "code"))
303
+ code_text = "\n".join(code_blocks)
304
+ term_rows: list[dict[str, Any]] = []
305
+ for term, category in all_terms:
306
+ term_rows.append({
307
+ "term": term,
308
+ "category": category,
309
+ "count": _term_count(clean_text, term),
310
+ "first_window": _first_window(window_rows, term),
311
+ "appears_in_code": "true" if term.lower() in code_text.lower() else "false",
312
+ })
313
+ term_rows.sort(key=lambda row: (str(row["category"]), str(row["term"]).lower()))
314
+ with (output_dir / "keyword_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
315
+ fieldnames = ["term", "category", "count", "first_window", "appears_in_code"]
316
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
317
+ writer.writeheader()
318
+ for row in term_rows:
319
+ writer.writerow(row)
320
+
321
+ code_terms = [str(item).lower() for item in spec.get("code_terms") or []]
322
+ with (output_dir / "code_block_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
323
+ fieldnames = ["block_index", "line_count", "char_count", "contains_code_term", "first_line"]
324
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
325
+ writer.writeheader()
326
+ for index, block in enumerate(code_blocks, start=1):
327
+ lines = [line for line in block.splitlines() if line.strip()]
328
+ contains = any(term and term in block.lower() for term in code_terms)
329
+ writer.writerow({
330
+ "block_index": index,
331
+ "line_count": len(lines),
332
+ "char_count": len(block),
333
+ "contains_code_term": "true" if contains else "false",
334
+ "first_line": lines[0][:200] if lines else "",
335
+ })
336
+
337
+ keyword_total = sum(int(row["count"]) for row in term_rows)
338
+ required_found = sum(1 for row in term_rows if row["category"] == "required" and int(row["count"]) > 0)
339
+ top_keyword = None
340
+ if term_rows:
341
+ top = max(term_rows, key=lambda row: (int(row["count"]), str(row["term"]).lower()))
342
+ top_keyword = str(top["term"]) if int(top["count"]) > 0 else None
343
+ summary = {
344
+ "software_name": str(spec.get("software_name") or ""),
345
+ "source_file": source_file,
346
+ "window_count": len(window_rows),
347
+ "heading_count": len(headings),
348
+ "code_block_count": len(code_blocks),
349
+ "keyword_total": keyword_total,
350
+ "unique_required_terms_found": required_found,
351
+ "top_keyword": top_keyword,
352
+ "source_character_count": len(clean_text),
353
+ }
354
+ _write_json(output_dir / "summary.json", summary)
355
+
356
+
357
+ def geojson_outputs(input_dir: Path, output_dir: Path) -> None:
358
+ payload = _read_json(input_dir / "events.geojson")
359
+ features = payload.get("features") if isinstance(payload, dict) else []
360
+ rows: list[dict[str, Any]] = []
361
+ for feature in features:
362
+ if not isinstance(feature, dict):
363
+ continue
364
+ props = feature.get("properties") if isinstance(feature.get("properties"), dict) else {}
365
+ geom = feature.get("geometry") if isinstance(feature.get("geometry"), dict) else {}
366
+ coords = geom.get("coordinates") if isinstance(geom.get("coordinates"), list) else []
367
+ mag = props.get("mag")
368
+ try:
369
+ mag = float(mag) if mag is not None else None
370
+ except (TypeError, ValueError):
371
+ mag = None
372
+ depth = None
373
+ if len(coords) >= 3:
374
+ try:
375
+ depth = float(coords[2])
376
+ except (TypeError, ValueError):
377
+ depth = None
378
+ time_ms = props.get("time")
379
+ date_utc = "unknown"
380
+ if isinstance(time_ms, (int, float)):
381
+ date_utc = dt.datetime.utcfromtimestamp(time_ms / 1000).date().isoformat()
382
+ rows.append({
383
+ "id": feature.get("id") or props.get("code") or "",
384
+ "mag": mag,
385
+ "depth": depth,
386
+ "place": str(props.get("place") or ""),
387
+ "region": _stable_region(str(props.get("place") or "")),
388
+ "date_utc": date_utc,
389
+ "feature": feature,
390
+ })
391
+ output_dir.mkdir(parents=True, exist_ok=True)
392
+ mags = [row["mag"] for row in rows if row["mag"] is not None]
393
+ depths = [row["depth"] for row in rows if row["depth"] is not None]
394
+ summary = {
395
+ "event_count": len(rows),
396
+ "magnitude": {
397
+ "count": len(mags),
398
+ "min": round(min(mags), 6) if mags else None,
399
+ "max": round(max(mags), 6) if mags else None,
400
+ "mean": round(statistics.fmean(mags), 6) if mags else None,
401
+ "median": round(statistics.median(mags), 6) if mags else None,
402
+ },
403
+ "depth_km": {
404
+ "count": len(depths),
405
+ "min": round(min(depths), 6) if depths else None,
406
+ "max": round(max(depths), 6) if depths else None,
407
+ "mean": round(statistics.fmean(depths), 6) if depths else None,
408
+ },
409
+ "time_range_utc": {
410
+ "start_date": min((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
411
+ "end_date": max((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
412
+ },
413
+ "quality_flags": {
414
+ "missing_magnitude": sum(1 for row in rows if row["mag"] is None),
415
+ "missing_depth": sum(1 for row in rows if row["depth"] is None),
416
+ "unique_regions": len({row["region"] for row in rows}),
417
+ },
418
+ }
419
+ _write_json(output_dir / "summary.json", summary)
420
+
421
+ grouped: dict[tuple[str, str], list[dict[str, Any]]] = {}
422
+ for row in rows:
423
+ grouped.setdefault((_mag_bin(row["mag"]), _depth_bin(row["depth"])), []).append(row)
424
+ with (output_dir / "magnitude_depth_bins.csv").open("w", encoding="utf-8", newline="") as handle:
425
+ fieldnames = ["mag_bin", "depth_bin", "event_count", "mean_magnitude", "max_magnitude"]
426
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
427
+ writer.writeheader()
428
+ for (mag_bin, depth_bin), group in sorted(grouped.items()):
429
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
430
+ writer.writerow({
431
+ "mag_bin": mag_bin,
432
+ "depth_bin": depth_bin,
433
+ "event_count": len(group),
434
+ "mean_magnitude": f"{statistics.fmean(group_mags):.6f}" if group_mags else "",
435
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
436
+ })
437
+
438
+ daily: dict[tuple[str, str], list[dict[str, Any]]] = {}
439
+ for row in rows:
440
+ daily.setdefault((row["region"], row["date_utc"]), []).append(row)
441
+ with (output_dir / "region_daily_counts.csv").open("w", encoding="utf-8", newline="") as handle:
442
+ fieldnames = ["region", "date_utc", "event_count", "max_magnitude"]
443
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
444
+ writer.writeheader()
445
+ for (region, date_utc), group in sorted(daily.items()):
446
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
447
+ writer.writerow({
448
+ "region": region,
449
+ "date_utc": date_utc,
450
+ "event_count": len(group),
451
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
452
+ })
453
+
454
+ filtered = [row["feature"] for row in rows if row["mag"] is not None and row["mag"] >= 4.5]
455
+ _write_json(output_dir / "filtered_events.geojson", {"type": "FeatureCollection", "features": filtered})
456
+
457
+
458
+ def csv_time_series_outputs(input_dir: Path, output_dir: Path) -> None:
459
+ spec = _read_json(input_dir / "analysis_spec.json")
460
+ csv_name = spec["input_file"]
461
+ date_column = spec["date_column"]
462
+ value_column = spec["value_column"]
463
+ rows: list[dict[str, Any]] = []
464
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
465
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
466
+ reader = csv.DictReader(lines)
467
+ for item in reader:
468
+ raw_date = item.get(date_column, "")
469
+ try:
470
+ day = dt.date.fromisoformat(raw_date)
471
+ except ValueError:
472
+ if raw_date.isdigit() and len(raw_date) == 4:
473
+ day = dt.date(int(raw_date), 1, 1)
474
+ else:
475
+ continue
476
+ raw_value = item.get(value_column, "")
477
+ try:
478
+ value = float(raw_value)
479
+ except (TypeError, ValueError):
480
+ value = None
481
+ rows.append({"date": day, "value": value, "raw": item})
482
+ rows.sort(key=lambda item: item["date"])
483
+ valid = [row for row in rows if row["value"] is not None]
484
+ output_dir.mkdir(parents=True, exist_ok=True)
485
+ monthly: dict[str, list[float]] = {}
486
+ for row in valid:
487
+ key = row["date"].strftime("%Y-%m")
488
+ monthly.setdefault(key, []).append(row["value"])
489
+ with (output_dir / "monthly_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
490
+ fieldnames = ["month", "count", "mean", "median", "min", "max"]
491
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
492
+ writer.writeheader()
493
+ for month, values in sorted(monthly.items()):
494
+ writer.writerow({
495
+ "month": month,
496
+ "count": len(values),
497
+ "mean": f"{statistics.fmean(values):.6f}",
498
+ "median": f"{statistics.median(values):.6f}",
499
+ "min": f"{min(values):.6f}",
500
+ "max": f"{max(values):.6f}",
501
+ })
502
+ existing_dates = {row["date"] for row in rows}
503
+ missing_days: list[dt.date] = []
504
+ if rows:
505
+ day = rows[0]["date"]
506
+ while day <= rows[-1]["date"]:
507
+ if day.weekday() < 5 and day not in existing_dates:
508
+ missing_days.append(day)
509
+ day += dt.timedelta(days=1)
510
+ with (output_dir / "missing_business_days.csv").open("w", encoding="utf-8", newline="") as handle:
511
+ writer = csv.DictWriter(handle, fieldnames=["date"])
512
+ writer.writeheader()
513
+ for day in missing_days:
514
+ writer.writerow({"date": day.isoformat()})
515
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
516
+ writer = csv.DictWriter(handle, fieldnames=["date", "value"])
517
+ writer.writeheader()
518
+ for row in valid:
519
+ writer.writerow({"date": row["date"].isoformat(), "value": f"{row['value']:.6f}"})
520
+ values = [row["value"] for row in valid]
521
+ summary = {
522
+ "series_name": spec.get("series_name", value_column),
523
+ "row_count": len(rows),
524
+ "valid_value_count": len(valid),
525
+ "missing_value_count": len(rows) - len(valid),
526
+ "missing_business_day_count": len(missing_days),
527
+ "date_range": {
528
+ "start": rows[0]["date"].isoformat() if rows else None,
529
+ "end": rows[-1]["date"].isoformat() if rows else None,
530
+ },
531
+ "value": {
532
+ "min": round(min(values), 6) if values else None,
533
+ "max": round(max(values), 6) if values else None,
534
+ "mean": round(statistics.fmean(values), 6) if values else None,
535
+ "median": round(statistics.median(values), 6) if values else None,
536
+ },
537
+ }
538
+ _write_json(output_dir / "summary.json", summary)
539
+
540
+
541
+ def _aggregate_annual_rows(rows: list[dict[str, Any]], aggregation: str) -> list[dict[str, Any]]:
542
+ if not aggregation:
543
+ return rows
544
+ grouped: dict[int, list[float]] = {}
545
+ for row in rows:
546
+ grouped.setdefault(int(row["year"]), []).append(float(row["value"]))
547
+ aggregated: list[dict[str, Any]] = []
548
+ for year, values in sorted(grouped.items()):
549
+ if aggregation == "mean":
550
+ value = statistics.fmean(values)
551
+ elif aggregation == "median":
552
+ value = statistics.median(values)
553
+ elif aggregation == "max":
554
+ value = max(values)
555
+ elif aggregation == "min":
556
+ value = min(values)
557
+ elif aggregation == "count":
558
+ value = float(len(values))
559
+ else:
560
+ raise ValueError(f"unsupported year_aggregation for csv_annual_series_audit: {aggregation}")
561
+ aggregated.append({"year": year, "value": value})
562
+ return aggregated
563
+
564
+
565
+ def csv_annual_series_outputs(input_dir: Path, output_dir: Path) -> None:
566
+ spec = _read_json(input_dir / "analysis_spec.json")
567
+ csv_name = spec["input_file"]
568
+ year_column = spec["year_column"]
569
+ value_column = spec["value_column"]
570
+ # Optional: derive the year from an ISO-8601 timestamp column (e.g. USGS FDSN
571
+ # `time` = "2022-06-29T19:17:58.929Z") instead of parsing an integer year.
572
+ # When unset, behaviour is unchanged (integer year column).
573
+ year_from_timestamp = bool(spec.get("year_from_timestamp", False))
574
+ year_aggregation = str(spec.get("year_aggregation") or "").strip().lower()
575
+ raw_rows: list[dict[str, Any]] = []
576
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
577
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
578
+ reader = csv.DictReader(lines)
579
+ for item in reader:
580
+ raw_year = str(item.get(year_column, "")).strip()
581
+ try:
582
+ if year_from_timestamp:
583
+ # Take the leading 4-digit year of an ISO timestamp/date.
584
+ match = re.match(r"\s*(\d{4})", raw_year)
585
+ if not match:
586
+ continue
587
+ year = int(match.group(1))
588
+ else:
589
+ year = int(raw_year)
590
+ value = float(str(item.get(value_column, "")).strip())
591
+ except (TypeError, ValueError):
592
+ continue
593
+ raw_rows.append({"year": year, "value": value})
594
+ rows = _aggregate_annual_rows(raw_rows, year_aggregation)
595
+ rows.sort(key=lambda item: item["year"])
596
+ output_dir.mkdir(parents=True, exist_ok=True)
597
+ decades: dict[str, list[float]] = {}
598
+ for row in rows:
599
+ decade_start = (row["year"] // 10) * 10
600
+ decades.setdefault(f"{decade_start}s", []).append(row["value"])
601
+ with (output_dir / "decadal_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
602
+ fieldnames = [
603
+ "decade_start",
604
+ "start_year",
605
+ "end_year",
606
+ "count",
607
+ "mean_departure",
608
+ "median_departure",
609
+ "min_departure",
610
+ "max_departure",
611
+ "sample_std_departure",
612
+ ]
613
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
614
+ writer.writeheader()
615
+ for decade, values in sorted(decades.items()):
616
+ decade_start = int(decade.rstrip("s"))
617
+ years = [row["year"] for row in rows if decade_start <= row["year"] <= decade_start + 9]
618
+ writer.writerow({
619
+ "decade_start": decade_start,
620
+ "start_year": min(years),
621
+ "end_year": max(years),
622
+ "count": len(values),
623
+ "mean_departure": f"{statistics.fmean(values):.6f}",
624
+ "median_departure": f"{statistics.median(values):.6f}",
625
+ "min_departure": f"{min(values):.6f}",
626
+ "max_departure": f"{max(values):.6f}",
627
+ "sample_std_departure": f"{statistics.stdev(values):.6f}" if len(values) >= 2 else "",
628
+ })
629
+ threshold = float(spec.get("threshold", 1.0))
630
+ with (output_dir / "threshold_years.csv").open("w", encoding="utf-8", newline="") as handle:
631
+ fieldnames = ["rank", "year", "departure", "threshold"]
632
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
633
+ writer.writeheader()
634
+ rank = 1
635
+ for row in rows:
636
+ if row["value"] >= threshold:
637
+ writer.writerow({
638
+ "rank": rank,
639
+ "year": row["year"],
640
+ "departure": f"{row['value']:.6f}",
641
+ "threshold": f"{threshold:.6f}",
642
+ })
643
+ rank += 1
644
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
645
+ writer = csv.DictWriter(handle, fieldnames=["Year", "Departure from Average"])
646
+ writer.writeheader()
647
+ for row in rows:
648
+ writer.writerow({"Year": row["year"], "Departure from Average": f"{row['value']:.6f}"})
649
+ values = [row["value"] for row in rows]
650
+ threshold_rows = [row for row in rows if row["value"] >= threshold]
651
+ min_row = min(rows, key=lambda row: row["value"]) if rows else None
652
+ max_row = max(rows, key=lambda row: row["value"]) if rows else None
653
+ summary = {
654
+ "series_name": spec.get("series_name", value_column),
655
+ "row_count": len(rows),
656
+ "year_range": {
657
+ "start": rows[0]["year"] if rows else None,
658
+ "end": rows[-1]["year"] if rows else None,
659
+ },
660
+ "value": {
661
+ "min": round(min(values), 6) if values else None,
662
+ "min_year": min_row["year"] if min_row else None,
663
+ "max": round(max(values), 6) if values else None,
664
+ "max_year": max_row["year"] if max_row else None,
665
+ "mean": round(statistics.fmean(values), 6) if values else None,
666
+ "median": round(statistics.median(values), 6) if values else None,
667
+ "sample_std": round(statistics.stdev(values), 6) if len(values) >= 2 else None,
668
+ },
669
+ "threshold": threshold,
670
+ "years_at_or_above_threshold": len(threshold_rows),
671
+ "first_threshold_year": threshold_rows[0]["year"] if threshold_rows else None,
672
+ "last_threshold_year": threshold_rows[-1]["year"] if threshold_rows else None,
673
+ }
674
+ if year_aggregation:
675
+ summary["year_aggregation"] = year_aggregation
676
+ summary["source_row_count"] = len(raw_rows)
677
+ _write_json(output_dir / "summary.json", summary)
678
+
679
+
20260904-005134-science-environment-e187c8/tests/reference/internal/workflow_runtime.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ from pathlib import Path
2
+ from workflow_adapters_runtime import csv_annual_series_outputs
3
+
4
+ def run_reference(input_dir: Path, output_dir: Path) -> None:
5
+ csv_annual_series_outputs(input_dir, output_dir)
20260904-005134-science-environment-e187c8/tests/reference/reference_manifest.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "workflow_kind": "csv_annual_series_audit",
3
+ "task_id": "usgs_fdsn_eq_andes_2016_2018_m5p5_mag_max_quota_repair_0777",
4
+ "domain": "science_environment",
5
+ "task": "usgs_andes_annual_earthquake_maximum_audit",
6
+ "variant": "base",
7
+ "created_from": "standard reference workflow",
8
+ "required_outputs": [
9
+ {
10
+ "path": "summary.json",
11
+ "type": "json"
12
+ },
13
+ {
14
+ "path": "decadal_statistics.csv",
15
+ "type": "csv"
16
+ },
17
+ {
18
+ "path": "threshold_years.csv",
19
+ "type": "csv"
20
+ },
21
+ {
22
+ "path": "clean_series.csv",
23
+ "type": "csv"
24
+ }
25
+ ],
26
+ "numeric_tolerance": 1e-09,
27
+ "source_prior_sha256": "e1ddddb3c34d2aaff002330320271be8f8316fbb6b963eeafc9638717cac5944",
28
+ "gold_output_hashes": {
29
+ "summary.json": "43e7644764e3a762c2a9bf218389e2de5530a08a8232ebc25f8eb943b8fb0f7a",
30
+ "decadal_statistics.csv": "3ced27dd512a0068d76b5fe5e11829834f3185f97e9425d71d77edd9675e0cf2",
31
+ "threshold_years.csv": "84e6fcc6c2facbc4ad772a3779979d14c1af8d9954186cebef848670e962c78e",
32
+ "clean_series.csv": "121b8ad4cb14071d6199d3556649ed4de87f5068810eb660a58c166c9e4c3cfc"
33
+ }
34
+ }
20260904-005134-science-environment-e187c8/tests/reference/reference_solution.py ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import argparse
4
+ import sys
5
+ from pathlib import Path
6
+
7
+
8
+ def main() -> int:
9
+ parser = argparse.ArgumentParser()
10
+ parser.add_argument("--input", required=True)
11
+ parser.add_argument("--output", required=True)
12
+ args = parser.parse_args()
13
+ sys.path.insert(0, str(Path(__file__).resolve().parent / "internal"))
14
+ from workflow_runtime import run_reference
15
+ run_reference(Path(args.input), Path(args.output))
16
+ return 0
17
+
18
+
19
+ if __name__ == "__main__":
20
+ raise SystemExit(main())
20260904-005134-science-environment-e187c8/tests/test.sh ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ set -u
3
+
4
+ TESTS_DIR="${HARBOR_TESTS_DIR:-/tests}"
5
+ WORKSPACE_DIR="${HARBOR_WORKSPACE_DIR:-/workspace}"
6
+ LOGS_DIR="${HARBOR_LOGS_DIR:-/logs/verifier}"
7
+
8
+ mkdir -p "${LOGS_DIR}" "${WORKSPACE_DIR}/output"
9
+
10
+ python "${TESTS_DIR}/evaluator_runner.py" \
11
+ --output-dir "${WORKSPACE_DIR}/output" \
12
+ --reference-dir "${TESTS_DIR}/reference" \
13
+ --reward-json "${LOGS_DIR}/reward.json" \
14
+ --details-json "${LOGS_DIR}/ale_evaluator_details.json"
15
+
16
+ if [ ! -f "${LOGS_DIR}/reward.json" ]; then
17
+ echo '{"score":0.0,"format":0.0,"output_present":0.0,"evaluator_available":0.0,"item_count":0.0}' > "${LOGS_DIR}/reward.json"
18
+ fi
19
+
20
+ python - <<'PY' "${LOGS_DIR}/reward.json" "${LOGS_DIR}/reward.txt"
21
+ import json
22
+ import sys
23
+ from pathlib import Path
24
+ reward = json.loads(Path(sys.argv[1]).read_text(encoding="utf-8"))
25
+ score = float(reward.get("score", 0.0))
26
+ Path(sys.argv[2]).write_text(f"{score}\n", encoding="utf-8")
27
+ PY
28
+
29
+ exit 0
20260904-005149-science-environment-cee684/.DS_Store ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:43563f84f4a31a498b303d984b47ef20cb1e8a4ca879b1b64027e16d70f7267c
3
+ size 6148
20260904-005149-science-environment-cee684/environment/Dockerfile ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ FROM python:3.12-slim
2
+
3
+ WORKDIR /workspace
4
+
5
+ COPY public_data/input /workspace/input
6
+ COPY public_data/software /workspace/software
7
+ COPY public_data/task_card.json /workspace/task_card.json
8
+ COPY public_data/main.py /workspace/main.py
9
+ COPY public_data/README.md /workspace/README.md
10
+
11
+ RUN mkdir -p /workspace/output /workspace/report /logs/agent /logs/artifacts \
12
+ && chmod -R 777 /workspace/output /workspace/report /logs
20260904-005149-science-environment-cee684/environment/public_data/README.md ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Public Task Inputs
2
+
3
+ Files in this directory are copied into the workspace:
4
+
5
+ - `input/` -> `/workspace/input/`
6
+ - `software/` -> `/workspace/software/`
7
+ - `task_card.json` -> `/workspace/task_card.json`
8
+ - `main.py` -> `/workspace/main.py`
9
+
10
+ Hidden reference data and evaluator files are not present in the agent workspace.
11
+ Final deliverables must be written under `/workspace/output/`.
20260904-005149-science-environment-cee684/environment/public_data/input/README.md ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ This directory contains the public staged input for this task.
2
+ input/analysis_spec.json is the authoritative public contract for input columns, grouping/filtering rules, required output files, and output schemas.
3
+ Primary data file: input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv.
4
+ Workflow kind: csv_annual_series_audit.
5
+ Required output files: output/summary.json, output/decadal_statistics.csv, output/threshold_years.csv, output/clean_series.csv.
6
+ Solve the task from the staged local files only; do not fetch fresh remote data during solve time.
20260904-005149-science-environment-cee684/environment/public_data/input/analysis_spec.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input_file": "usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv",
3
+ "workflow_kind": "csv_annual_series_audit",
4
+ "year_column": "time",
5
+ "value_column": "depth",
6
+ "series_name": "USGS FDSN Andes subduction zone 1970-1988-1970-1988 earthquake annual mean focal depth (km) (min mag 4.5)",
7
+ "threshold": 70.0,
8
+ "year_from_timestamp": true,
9
+ "comment_prefix": "#",
10
+ "required_outputs": [
11
+ {
12
+ "path": "summary.json",
13
+ "type": "json"
14
+ },
15
+ {
16
+ "path": "decadal_statistics.csv",
17
+ "type": "csv"
18
+ },
19
+ {
20
+ "path": "threshold_years.csv",
21
+ "type": "csv"
22
+ },
23
+ {
24
+ "path": "clean_series.csv",
25
+ "type": "csv"
26
+ }
27
+ ],
28
+ "output_schemas": {
29
+ "summary.json": {
30
+ "series_name": "string",
31
+ "row_count": "integer valid annual rows",
32
+ "year_range": {
33
+ "start": "integer|null",
34
+ "end": "integer|null"
35
+ },
36
+ "value": {
37
+ "min": "number|null",
38
+ "min_year": "integer|null",
39
+ "max": "number|null",
40
+ "max_year": "integer|null",
41
+ "mean": "number|null",
42
+ "median": "number|null",
43
+ "sample_std": "number|null"
44
+ },
45
+ "threshold": "number",
46
+ "years_at_or_above_threshold": "integer",
47
+ "first_threshold_year": "integer|null",
48
+ "last_threshold_year": "integer|null",
49
+ "year_aggregation": "string annual aggregation applied to raw rows",
50
+ "source_row_count": "integer raw rows before annual aggregation"
51
+ },
52
+ "decadal_statistics.csv": [
53
+ "decade_start",
54
+ "start_year",
55
+ "end_year",
56
+ "count",
57
+ "mean_departure",
58
+ "median_departure",
59
+ "min_departure",
60
+ "max_departure",
61
+ "sample_std_departure"
62
+ ],
63
+ "threshold_years.csv": [
64
+ "rank",
65
+ "year",
66
+ "departure",
67
+ "threshold"
68
+ ],
69
+ "clean_series.csv": [
70
+ "Year",
71
+ "Departure from Average"
72
+ ]
73
+ },
74
+ "notes": "First group raw CSV rows by year and aggregate the value column with mean; then compute reproducible annual-series audit outputs. Do not fetch new data.",
75
+ "year_aggregation": "mean"
76
+ }
20260904-005149-science-environment-cee684/environment/public_data/input/source_context_windows.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
20260904-005149-science-environment-cee684/environment/public_data/input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv ADDED
The diff for this file is too large to render. See raw diff
 
20260904-005149-science-environment-cee684/environment/public_data/main.py ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import importlib.util
4
+ import json
5
+ import os
6
+ import shlex
7
+ import shutil
8
+ from pathlib import Path
9
+ from typing import Any
10
+
11
+
12
+ class Task:
13
+ def __init__(self, task_id: str, description: str, metadata: dict[str, Any]):
14
+ self.task_id = task_id
15
+ self.description = description
16
+ self.prompt = description
17
+ self.metadata = metadata
18
+ self.computer = {"setup_config": {"os_type": "linux"}}
19
+
20
+
21
+ def _task_dir() -> Path:
22
+ return Path(__file__).resolve().parent
23
+
24
+
25
+ def _load_card() -> dict[str, Any]:
26
+ path = _task_dir() / "task_card.json"
27
+ if not path.is_file():
28
+ return {
29
+ "taskId": os.environ.get("ALE_TASK_ID", "science_environment/usgs_andes_annual_focal_depth_audit"),
30
+ "taskPrompt": "Task card has not been rendered yet; this fallback exists for stage-03 hook smoke only.",
31
+ "metadata": {
32
+ "domain": "science_environment",
33
+ "task": "usgs_andes_annual_focal_depth_audit",
34
+ "variant": "base",
35
+ "domain_name": "science_environment",
36
+ "task_name": "usgs_andes_annual_focal_depth_audit",
37
+ "variant_name": "base",
38
+ "requires_task_data": True,
39
+ "task_dir": "/media/user/data/agenthle/science_environment/usgs_andes_annual_focal_depth_audit/base",
40
+ "input_dir": "/media/user/data/agenthle/science_environment/usgs_andes_annual_focal_depth_audit/base/input",
41
+ "software_dir": "/media/user/data/agenthle/science_environment/usgs_andes_annual_focal_depth_audit/base/software",
42
+ "remote_output_dir": "/media/user/data/agenthle/science_environment/usgs_andes_annual_focal_depth_audit/base/output",
43
+ "reference_dir": "/media/user/data/agenthle/science_environment/usgs_andes_annual_focal_depth_audit/base/reference",
44
+ },
45
+ }
46
+ with path.open("r", encoding="utf-8") as handle:
47
+ return json.load(handle)
48
+
49
+
50
+ def load() -> list[Task]:
51
+ card = _load_card()
52
+ metadata = dict(card.get("metadata") or {})
53
+ if not metadata:
54
+ domain, _, task = str(card["taskId"]).partition("/")
55
+ variant = "base"
56
+ task_dir = f"/media/user/data/agenthle/{domain}/{task}/{variant}"
57
+ metadata.update({
58
+ "domain": domain,
59
+ "task": task,
60
+ "variant": variant,
61
+ "domain_name": domain,
62
+ "task_name": task,
63
+ "variant_name": variant,
64
+ "requires_task_data": True,
65
+ "task_dir": task_dir,
66
+ "input_dir": f"{task_dir}/input",
67
+ "software_dir": f"{task_dir}/software",
68
+ "remote_output_dir": f"{task_dir}/output",
69
+ "reference_dir": f"{task_dir}/reference",
70
+ })
71
+ return [Task(task_id=card["taskId"], description=card["taskPrompt"], metadata=metadata)]
72
+
73
+
74
+ def _metadata(cfg: Any = None) -> dict[str, Any]:
75
+ if cfg is None:
76
+ return {}
77
+ value = getattr(cfg, "metadata", None)
78
+ return value if isinstance(value, dict) else {}
79
+
80
+
81
+ def _remote_path(metadata: dict[str, Any], key: str, fallback: str) -> str:
82
+ value = str(metadata.get(key) or "").strip()
83
+ return value or fallback
84
+
85
+
86
+ async def _remote_run(session: Any, command: str, *, check: bool = False) -> Any:
87
+ try:
88
+ return await session.run_command(command, check=check)
89
+ except TypeError:
90
+ return await session.run_command(command)
91
+
92
+
93
+ async def _remote_start(cfg: Any, session: Any) -> dict[str, Any]:
94
+ metadata = _metadata(cfg)
95
+ output_dir = _remote_path(metadata, "remote_output_dir", "output")
96
+ input_dir = _remote_path(metadata, "input_dir", "input")
97
+ software_dir = _remote_path(metadata, "software_dir", "software")
98
+ command = (
99
+ "set -eu\n"
100
+ f"mkdir -p {shlex.quote(output_dir)}\n"
101
+ f"find {shlex.quote(output_dir)} -mindepth 1 -maxdepth 1 -exec rm -rf -- {{}} +\n"
102
+ )
103
+ await _remote_run(session, command, check=True)
104
+ return {"ok": True, "input_dir": input_dir, "software_dir": software_dir, "output_dir": output_dir}
105
+
106
+
107
+ def start(cfg: Any = None, session: Any = None) -> dict[str, Any]:
108
+ if session is not None:
109
+ return _remote_start(cfg, session)
110
+ del cfg
111
+ output_dir = Path(os.environ.get("ALE_OUTPUT_DIR", "output"))
112
+ output_dir.mkdir(parents=True, exist_ok=True)
113
+ for child in output_dir.iterdir():
114
+ if child.is_dir():
115
+ shutil.rmtree(child)
116
+ else:
117
+ child.unlink()
118
+ input_dir = Path(os.environ.get("ALE_INPUT_DIR", "input"))
119
+ software_dir = Path(os.environ.get("ALE_SOFTWARE_DIR", "software"))
120
+ return {
121
+ "ok": True,
122
+ "input_dir": str(input_dir),
123
+ "software_dir": str(software_dir),
124
+ "output_dir": str(output_dir),
125
+ }
126
+
127
+
128
+ async def _remote_evaluate(cfg: Any, session: Any) -> list[float]:
129
+ metadata = _metadata(cfg)
130
+ output_dir = _remote_path(metadata, "remote_output_dir", "output")
131
+ reference_dir = _remote_path(metadata, "reference_dir", "reference")
132
+ script = f"""
133
+ import importlib.util
134
+ import json
135
+ import pathlib
136
+
137
+ output_dir = pathlib.Path({output_dir!r})
138
+ reference_dir = pathlib.Path({reference_dir!r})
139
+ evaluator_path = reference_dir / "evaluator.py"
140
+ if not evaluator_path.is_file():
141
+ raise FileNotFoundError(f"hidden evaluator is missing: {{evaluator_path}}")
142
+ spec = importlib.util.spec_from_file_location("hidden_ale_evaluator", evaluator_path)
143
+ if spec is None or spec.loader is None:
144
+ raise RuntimeError(f"cannot import hidden evaluator: {{evaluator_path}}")
145
+ module = importlib.util.module_from_spec(spec)
146
+ spec.loader.exec_module(module)
147
+ result = module.evaluate(output_dir, reference_dir)
148
+ print(json.dumps(result))
149
+ """
150
+ command = "python3 - <<'PY'\n" + script + "\nPY"
151
+ result = await _remote_run(session, command, check=True)
152
+ stdout = getattr(result, "stdout", "") or ""
153
+ for line in reversed([item.strip() for item in stdout.splitlines() if item.strip()]):
154
+ try:
155
+ payload = json.loads(line)
156
+ if isinstance(payload, list):
157
+ return [float(max(0.0, min(1.0, value))) for value in payload]
158
+ except Exception:
159
+ continue
160
+ raise RuntimeError("hidden evaluator did not emit a JSON score list")
161
+
162
+
163
+ def evaluate(cfg: Any = None, session: Any = None) -> list[float]:
164
+ if session is not None:
165
+ return _remote_evaluate(cfg, session)
166
+ del cfg
167
+ output_dir = Path(os.environ.get("ALE_OUTPUT_DIR", "output"))
168
+ reference_dir = Path(os.environ.get("ALE_REFERENCE_DIR", "reference"))
169
+ evaluator_path = reference_dir / "evaluator.py"
170
+ if not evaluator_path.is_file():
171
+ raise FileNotFoundError(f"hidden evaluator is missing: {evaluator_path}")
172
+ spec = importlib.util.spec_from_file_location("hidden_ale_evaluator", evaluator_path)
173
+ if spec is None or spec.loader is None:
174
+ raise RuntimeError(f"cannot import hidden evaluator: {evaluator_path}")
175
+ module = importlib.util.module_from_spec(spec)
176
+ spec.loader.exec_module(module)
177
+ result = module.evaluate(output_dir, reference_dir)
178
+ if not isinstance(result, list):
179
+ raise TypeError("hidden evaluator must return list[float]")
180
+ return [float(max(0.0, min(1.0, value))) for value in result]
20260904-005149-science-environment-cee684/environment/public_data/software/validate_outputs.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ from __future__ import annotations
3
+
4
+ import json
5
+ import os
6
+ import sys
7
+ from pathlib import Path
8
+
9
+
10
+ def required_outputs(input_dir: Path) -> list[str]:
11
+ spec_path = input_dir / "analysis_spec.json"
12
+ if spec_path.is_file():
13
+ payload = json.loads(spec_path.read_text(encoding="utf-8"))
14
+ required = payload.get("required_outputs") if isinstance(payload, dict) else None
15
+ if isinstance(required, list):
16
+ rows = [str(item.get("path") or "") for item in required if isinstance(item, dict)]
17
+ rows = [item for item in rows if item]
18
+ if rows:
19
+ return rows
20
+ return [
21
+ "summary.json",
22
+ "magnitude_depth_bins.csv",
23
+ "region_daily_counts.csv",
24
+ "filtered_events.geojson",
25
+ ]
26
+
27
+
28
+ def main() -> int:
29
+ output_dir = Path(sys.argv[1]) if len(sys.argv) > 1 else Path("output")
30
+ input_dir = Path(sys.argv[2]) if len(sys.argv) > 2 else Path(os.environ.get("ALE_INPUT_DIR", "input"))
31
+ required = required_outputs(input_dir)
32
+ missing = [name for name in required if not (output_dir / name).is_file()]
33
+ if missing:
34
+ print(json.dumps({"ok": False, "missing": missing}, indent=2))
35
+ return 1
36
+ print(json.dumps({"ok": True, "required": required}, indent=2))
37
+ return 0
38
+
39
+
40
+ if __name__ == "__main__":
41
+ raise SystemExit(main())
20260904-005149-science-environment-cee684/environment/public_data/task_card.json ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "taskId": "science_environment/usgs_andes_annual_focal_depth_audit",
3
+ "title": "USGS Fdsn Eq Andes 1970 1970 M4p5 Depth Mean Quota Repair 0778",
4
+ "summary": "Audit the real USGS FDSN Andes subduction zone earthquake catalog for 1970-1988-1970-1988 (events with magnitude >= 4.5) and produce reproducible annual annual mean focal depth (km) statistics, decadal aggregates, threshold-year checks, and a cleaned local series.",
5
+ "category": "science_environment",
6
+ "taxonomy": {
7
+ "domain_id": "science_environment",
8
+ "domain_code": "science_environment",
9
+ "subdomain_id": "science_environment.data_analysis",
10
+ "subdomain_code": "data_analysis",
11
+ "subdomain_name": "Data Analysis",
12
+ "classifier": {
13
+ "source": "deterministic task metadata",
14
+ "rationale": "CLI-only task derived from a public, reproducible workflow seed."
15
+ }
16
+ },
17
+ "taskPrompt": "Use a Linux command-line workflow to complete this local data-audit task. Read only the staged files in input/ and software/, write final deliverables under output/, and do not use network access, credentials, GUI tools, browser automation, Windows-only tooling, or mobile apps. The authoritative public contract is input/analysis_spec.json; if any wording elsewhere is less specific, follow analysis_spec.json for file names, columns, filters, grouping rules, rounding rules, and output schemas. The workflow kind is csv_annual_series_audit. The primary staged data file is input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv. Produce exactly these required deliverables: output/summary.json, output/decadal_statistics.csv, output/threshold_years.csv, output/clean_series.csv. The output schemas are:\n- output/summary.json: JSON keys and nested shape {\"first_threshold_year\": \"integer|null\", \"last_threshold_year\": \"integer|null\", \"row_count\": \"integer valid annual rows\", \"series_name\": \"string\", \"source_row_count\": \"integer raw rows before annual aggregation\", \"threshold\": \"number\", \"value\": {\"max\": \"number|null\", \"max_year\": \"integer|null\", \"mean\": \"number|null\", \"median\": \"number|null\", \"min\": \"number|null\", \"min_year\": \"integer|null\", \"sample_std\": \"number|null\"}, \"year_aggregation\": \"string annual aggregation applied to raw rows\", \"year_range\": {\"end\": \"integer|null\", \"start\": \"integer|null\"}, \"years_at_or_above_threshold\": \"integer\"}\n- output/decadal_statistics.csv: CSV header exactly decade_start, start_year, end_year, count, mean_departure, median_departure, min_departure, max_departure, sample_std_departure\n- output/threshold_years.csv: CSV header exactly rank, year, departure, threshold\n- output/clean_series.csv: CSV header exactly Year, Departure from Average\nCompute all requested values from the staged data rather than hard-coding constants. Use deterministic ordering, UTF-8 text, explicit CSV headers, valid JSON/GeoJSON where applicable, and stable numeric formatting. You may run software/validate_outputs.py as a public sanity check after writing outputs, but the deliverables in output/ are what will be evaluated.",
18
+ "agentMustDo": [
19
+ "Read input/analysis_spec.json first and treat it as the authoritative public specification.",
20
+ "Read input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv and compute results from the staged local data.",
21
+ "Create output/summary.json as json exactly as specified by analysis_spec.json.",
22
+ "Create output/decadal_statistics.csv as csv exactly as specified by analysis_spec.json.",
23
+ "Create output/threshold_years.csv as csv exactly as specified by analysis_spec.json.",
24
+ "Create output/clean_series.csv as csv exactly as specified by analysis_spec.json.",
25
+ "Place every final deliverable directly under output/ with the exact required filename.",
26
+ "Use only local staged files and installed Linux CLI tooling; do not download fresh data."
27
+ ],
28
+ "inputFiles": [
29
+ {
30
+ "name": "README.md",
31
+ "format": "Markdown",
32
+ "path": "input/README.md",
33
+ "description": "Agent-visible staged input file for this task."
34
+ },
35
+ {
36
+ "name": "analysis_spec.json",
37
+ "format": "JSON",
38
+ "path": "input/analysis_spec.json",
39
+ "description": "Agent-visible staged input file for this task."
40
+ },
41
+ {
42
+ "name": "source_context_windows.jsonl",
43
+ "format": "JSONL",
44
+ "path": "input/source_context_windows.jsonl",
45
+ "description": "Agent-visible staged input file for this task."
46
+ },
47
+ {
48
+ "name": "usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv",
49
+ "format": "CSV",
50
+ "path": "input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv",
51
+ "description": "Agent-visible staged input file for this task."
52
+ }
53
+ ],
54
+ "referenceFiles": [
55
+ {
56
+ "name": "summary.json",
57
+ "format": "JSON",
58
+ "path": "output/summary.json",
59
+ "description": "Required agent output file evaluated against hidden reference data."
60
+ },
61
+ {
62
+ "name": "decadal_statistics.csv",
63
+ "format": "CSV",
64
+ "path": "output/decadal_statistics.csv",
65
+ "description": "Required agent output file evaluated against hidden reference data."
66
+ },
67
+ {
68
+ "name": "threshold_years.csv",
69
+ "format": "CSV",
70
+ "path": "output/threshold_years.csv",
71
+ "description": "Required agent output file evaluated against hidden reference data."
72
+ },
73
+ {
74
+ "name": "clean_series.csv",
75
+ "format": "CSV",
76
+ "path": "output/clean_series.csv",
77
+ "description": "Required agent output file evaluated against hidden reference data."
78
+ }
79
+ ],
80
+ "software": [
81
+ "Python 3.11",
82
+ "standard Unix shell",
83
+ "software/validate_outputs.py"
84
+ ],
85
+ "requiredSystemPackages": [],
86
+ "evaluation": "Final required files under output/ are compared against standard-reference outputs generated by the deterministic workflow for this task.",
87
+ "vm": {
88
+ "machineType": "c4-standard-4",
89
+ "snapshot": "cpu-free-ubuntu",
90
+ "timeout": 7200
91
+ }
92
+ }
20260904-005149-science-environment-cee684/instruction.md ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # USGS Fdsn Eq Andes 1970 1970 M4p5 Depth Mean Quota Repair 0778
2
+
3
+ Audit the real USGS FDSN Andes subduction zone earthquake catalog for 1970-1988-1970-1988 (events with magnitude >= 4.5) and produce reproducible annual annual mean focal depth (km) statistics, decadal aggregates, threshold-year checks, and a cleaned local series.
4
+
5
+ ## Task
6
+
7
+ Use a Linux command-line workflow to complete this local data-audit task. Read only the staged files in input/ and software/, write final deliverables under output/, and do not use network access, credentials, GUI tools, browser automation, Windows-only tooling, or mobile apps. The authoritative public contract is input/analysis_spec.json; if any wording elsewhere is less specific, follow analysis_spec.json for file names, columns, filters, grouping rules, rounding rules, and output schemas. The workflow kind is csv_annual_series_audit. The primary staged data file is input/usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778.csv. Produce exactly these required deliverables: output/summary.json, output/decadal_statistics.csv, output/threshold_years.csv, output/clean_series.csv. The output schemas are:
8
+ - output/summary.json: JSON keys and nested shape {"first_threshold_year": "integer|null", "last_threshold_year": "integer|null", "row_count": "integer valid annual rows", "series_name": "string", "source_row_count": "integer raw rows before annual aggregation", "threshold": "number", "value": {"max": "number|null", "max_year": "integer|null", "mean": "number|null", "median": "number|null", "min": "number|null", "min_year": "integer|null", "sample_std": "number|null"}, "year_aggregation": "string annual aggregation applied to raw rows", "year_range": {"end": "integer|null", "start": "integer|null"}, "years_at_or_above_threshold": "integer"}
9
+ - output/decadal_statistics.csv: CSV header exactly decade_start, start_year, end_year, count, mean_departure, median_departure, min_departure, max_departure, sample_std_departure
10
+ - output/threshold_years.csv: CSV header exactly rank, year, departure, threshold
11
+ - output/clean_series.csv: CSV header exactly Year, Departure from Average
12
+ Compute all requested values from the staged data rather than hard-coding constants. Use deterministic ordering, UTF-8 text, explicit CSV headers, valid JSON/GeoJSON where applicable, and stable numeric formatting. You may run software/validate_outputs.py as a public sanity check after writing outputs, but the deliverables in output/ are what will be evaluated.
13
+
14
+ ## Workspace
15
+
16
+ You are working in `/workspace`.
17
+
18
+ - Public input files are under `/workspace/input/`.
19
+ - Public helper software and validation scripts are under `/workspace/software/`.
20
+ - The public task card is available at `/workspace/task_card.json`.
21
+ - The public task driver is available at `/workspace/main.py` for reference only.
22
+ - Do not use hidden reference data, network downloads, credentials, GUI tools, or browser automation.
23
+
24
+ ## Required Outputs
25
+
26
+ Create every final deliverable exactly at these paths:
27
+
28
+ - `/workspace/output/summary.json` (json)
29
+ - `/workspace/output/decadal_statistics.csv` (csv)
30
+ - `/workspace/output/threshold_years.csv` (csv)
31
+ - `/workspace/output/clean_series.csv` (csv)
32
+
33
+ Follow `/workspace/input/analysis_spec.json` whenever it is more specific than prose. You may run public validation helpers, but the files under `/workspace/output/` are what will be evaluated.
20260904-005149-science-environment-cee684/solution/internal/workflow_adapters_runtime.py ADDED
@@ -0,0 +1,679 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import csv
4
+ import datetime as dt
5
+ import html
6
+ import json
7
+ import math
8
+ import re
9
+ import shutil
10
+ import statistics
11
+ from pathlib import Path
12
+ from typing import Any
13
+
14
+
15
+ def _read_json(path: Path) -> Any:
16
+ with path.open("r", encoding="utf-8") as handle:
17
+ return json.load(handle)
18
+
19
+
20
+ def _write_json(path: Path, payload: Any) -> None:
21
+ path.parent.mkdir(parents=True, exist_ok=True)
22
+ with path.open("w", encoding="utf-8") as handle:
23
+ json.dump(payload, handle, ensure_ascii=False, indent=2)
24
+ handle.write("\n")
25
+
26
+ def _stable_region(place: str) -> str:
27
+ if not place:
28
+ return "unknown"
29
+ if "," in place:
30
+ return place.split(",")[-1].strip() or place.strip()
31
+ parts = place.split()
32
+ return " ".join(parts[-2:]) if len(parts) >= 2 else place.strip()
33
+
34
+
35
+ def _mag_bin(value: float | None) -> str:
36
+ if value is None or math.isnan(value):
37
+ return "unknown"
38
+ lower = math.floor(value)
39
+ upper = lower + 1
40
+ return f"{lower:.0f}-{upper:.0f}"
41
+
42
+
43
+ def _depth_bin(value: float | None) -> str:
44
+ if value is None or math.isnan(value):
45
+ return "unknown"
46
+ if value < 0:
47
+ return "<0"
48
+ if value < 10:
49
+ return "0-10"
50
+ if value < 30:
51
+ return "10-30"
52
+ if value < 70:
53
+ return "30-70"
54
+ if value < 300:
55
+ return "70-300"
56
+ return "300+"
57
+
58
+
59
+ def _clean_markup_text(raw: str) -> str:
60
+ text = re.sub(r"(?is)<(script|style|noscript)[^>]*>.*?</\1>", " ", raw)
61
+ text = re.sub(r"(?is)<br\s*/?>", "\n", text)
62
+ text = re.sub(r"(?is)</(p|div|li|tr|h[1-6]|pre)>", "\n", text)
63
+ text = re.sub(r"(?is)<[^>]+>", " ", text)
64
+ text = html.unescape(text)
65
+ text = re.sub(r"[ \t]+", " ", text)
66
+ text = re.sub(r"\n\s+", "\n", text)
67
+ return text.strip()
68
+
69
+
70
+ def _heading_slug(value: str) -> str:
71
+ slug = re.sub(r"[^a-z0-9]+", "_", value.lower()).strip("_")
72
+ return slug[:80] or "section"
73
+
74
+
75
+ def _extract_headings(raw: str) -> list[dict[str, str]]:
76
+ headings: list[dict[str, str]] = []
77
+ for match in re.finditer(r"(?is)<h([1-6])[^>]*>(.*?)</h\1>", raw):
78
+ text = _clean_markup_text(match.group(2))
79
+ if text:
80
+ headings.append({"level": f"h{match.group(1)}", "text": text})
81
+ if headings:
82
+ return headings
83
+ for line in raw.splitlines():
84
+ match = re.match(r"^(#{1,6})\s+(.+?)\s*$", line)
85
+ if match:
86
+ headings.append({"level": f"h{len(match.group(1))}", "text": match.group(2).strip()})
87
+ return headings
88
+
89
+
90
+ def _extract_code_blocks(raw: str) -> list[str]:
91
+ blocks: list[str] = []
92
+ for match in re.finditer(r"(?is)<pre[^>]*>(.*?)</pre>", raw):
93
+ text = _clean_markup_text(match.group(1))
94
+ if text:
95
+ blocks.append(text)
96
+ for match in re.finditer(r"(?is)<code[^>]*>(.*?)</code>", raw):
97
+ text = _clean_markup_text(match.group(1))
98
+ if text and text not in blocks:
99
+ blocks.append(text)
100
+ for match in re.finditer(r"(?s)```(?:[A-Za-z0-9_-]+)?\n(.*?)```", raw):
101
+ text = html.unescape(match.group(1).strip())
102
+ if text and text not in blocks:
103
+ blocks.append(text)
104
+ return blocks
105
+
106
+
107
+ def _window_rows(clean_text: str, *, width: int = 40, stride: int = 20) -> list[dict[str, Any]]:
108
+ words = re.findall(r"\S+", clean_text)
109
+ rows: list[dict[str, Any]] = []
110
+ if not words:
111
+ return rows
112
+ start = 0
113
+ while start < len(words):
114
+ end = min(start + width, len(words))
115
+ rows.append({
116
+ "window_id": len(rows) + 1,
117
+ "start_word": start,
118
+ "end_word": end,
119
+ "text": " ".join(words[start:end]),
120
+ })
121
+ if end == len(words):
122
+ break
123
+ start += stride
124
+ return rows
125
+
126
+
127
+ def _term_count(text: str, term: str) -> int:
128
+ term = term.strip()
129
+ if not term:
130
+ return 0
131
+ return len(re.findall(re.escape(term), text, flags=re.IGNORECASE))
132
+
133
+
134
+ def _first_window(windows: list[dict[str, Any]], term: str) -> str:
135
+ term_lower = term.lower()
136
+ for row in windows:
137
+ if term_lower in str(row.get("text") or "").lower():
138
+ return str(row.get("window_id") or "")
139
+ return ""
140
+
141
+
142
+ def _write_tutorial_public_inputs(input_dir: Path, source_path: Path, spec_payload: dict[str, Any]) -> None:
143
+ raw = source_path.read_text(encoding="utf-8", errors="replace")
144
+ # Tutorial examples may contain a maintainer's local Windows path. Keep
145
+ # the command semantics while preventing host-specific paths from entering
146
+ # the public task artifact.
147
+ sanitized_raw = re.sub(r"(?i)\b[A-Z]:\\(?:[^\\\r\n<>\"']+\\?)*[^\\\r\n<>\"']*", "./local.db", raw)
148
+ clean_text = _clean_markup_text(sanitized_raw)
149
+ windows = _window_rows(clean_text)
150
+ if len(windows) < 100:
151
+ raise ValueError(f"official tutorial source produced only {len(windows)} text windows; need at least 100")
152
+ required_terms = [str(item) for item in spec_payload.get("required_terms") or []]
153
+ code_terms = [str(item) for item in spec_payload.get("code_terms") or []]
154
+ input_dir.mkdir(parents=True, exist_ok=True)
155
+ destination = input_dir / str(spec_payload["source_file"])
156
+ # Always materialize the sanitized copy, including when the downloader
157
+ # already placed the source at the destination path.
158
+ destination.write_text(sanitized_raw, encoding="utf-8")
159
+ with (input_dir / str(spec_payload["input_file"])).open("w", encoding="utf-8", newline="") as handle:
160
+ fieldnames = ["window_id", "start_word", "end_word", "text", "contains_code_term", "matched_required_terms"]
161
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
162
+ writer.writeheader()
163
+ for row in windows:
164
+ text = str(row["text"])
165
+ matched = [term for term in required_terms if term.lower() in text.lower()]
166
+ contains_code = any(term.lower() in text.lower() for term in code_terms)
167
+ writer.writerow({
168
+ "window_id": row["window_id"],
169
+ "start_word": row["start_word"],
170
+ "end_word": row["end_word"],
171
+ "text": text,
172
+ "contains_code_term": "true" if contains_code else "false",
173
+ "matched_required_terms": "|".join(matched),
174
+ })
175
+ metadata = {
176
+ "software_name": spec_payload.get("software_name"),
177
+ "tutorial_url": spec_payload.get("tutorial_url"),
178
+ "source_urls": spec_payload.get("source_urls"),
179
+ "source_file": spec_payload.get("source_file"),
180
+ "window_file": spec_payload.get("input_file"),
181
+ "window_rule": spec_payload.get("window_rule"),
182
+ "required_terms": spec_payload.get("required_terms"),
183
+ "code_terms": spec_payload.get("code_terms"),
184
+ "note": "Provenance and public extraction rules only; scored aggregate values are intentionally not stored here.",
185
+ }
186
+ _write_json(input_dir / "tutorial_metadata.json", metadata)
187
+
188
+
189
+ def _source_context_text(path: Path) -> str:
190
+ try:
191
+ raw = path.read_text(encoding="utf-8", errors="replace")
192
+ except Exception:
193
+ return f"{path.name}\nsha256={_sha256_file(path)}\nsize_bytes={path.stat().st_size}\n"
194
+ if path.suffix.lower() in {".html", ".htm", ".md", ".txt"}:
195
+ return _clean_markup_text(raw)
196
+ return re.sub(r"\s+", " ", raw).strip()
197
+
198
+
199
+ def _ensure_public_input_footprint(input_dir: Path, source_paths: list[Path]) -> str:
200
+ current = _tree_file_bytes(input_dir)
201
+ if current >= MIN_PUBLIC_TASK_DATA_BYTES:
202
+ return f"public input footprint is {current} bytes; no context expansion needed"
203
+ context_path = input_dir / "source_context_windows.jsonl"
204
+ base = current - (context_path.stat().st_size if context_path.is_file() else 0)
205
+ context_path.unlink(missing_ok=True)
206
+ rows: list[dict[str, Any]] = []
207
+ for source in source_paths:
208
+ if not source.is_file() or source.resolve() == context_path.resolve():
209
+ continue
210
+ text = _source_context_text(source)
211
+ if not text:
212
+ continue
213
+ start = 0
214
+ width = 1800
215
+ stride = 900
216
+ while start < len(text):
217
+ end = min(len(text), start + width)
218
+ rows.append({
219
+ "source_file": source.relative_to(input_dir).as_posix() if source.is_relative_to(input_dir) else source.name,
220
+ "source_sha256": _sha256_file(source),
221
+ "start_char": start,
222
+ "end_char": end,
223
+ "text": text[start:end],
224
+ })
225
+ if end == len(text):
226
+ break
227
+ start += stride
228
+ if not rows:
229
+ raise ValueError("public input footprint expansion has no public source text")
230
+ input_dir.mkdir(parents=True, exist_ok=True)
231
+ rows_written = 0
232
+ target_extra = MIN_PUBLIC_TASK_DATA_BYTES - max(0, base)
233
+ with context_path.open("w", encoding="utf-8") as handle:
234
+ cycle = 0
235
+ while handle.tell() < target_extra or rows_written < 100:
236
+ row = dict(rows[rows_written % len(rows)])
237
+ row["window_id"] = rows_written + 1
238
+ row["cycle"] = cycle
239
+ handle.write(json.dumps(row, ensure_ascii=False) + "\n")
240
+ rows_written += 1
241
+ if rows_written % len(rows) == 0:
242
+ cycle += 1
243
+ handle.flush()
244
+ return f"wrote input/source_context_windows.jsonl with {rows_written} public-source windows and {context_path.stat().st_size} bytes"
245
+
246
+
247
+ def _write_software_manifest(software_dir: Path, spec_payload: dict[str, Any]) -> None:
248
+ if str(spec_payload.get("workflow_kind") or "") != "software_tutorial_doc_audit":
249
+ return
250
+ software_dir.mkdir(parents=True, exist_ok=True)
251
+ software_name = str(spec_payload.get("software_name") or "").strip()
252
+ executables = [str(item).strip() for item in spec_payload.get("executables") or [] if str(item).strip()]
253
+ system_packages = [str(item).strip() for item in spec_payload.get("required_system_packages") or [] if str(item).strip()]
254
+ source_urls = [str(item).strip() for item in spec_payload.get("source_urls") or [] if str(item).strip()]
255
+ _write_json(software_dir / "software_manifest.json", {
256
+ "software_name": software_name,
257
+ "executables": executables,
258
+ "required_system_packages": system_packages,
259
+ "official_sources": source_urls,
260
+ "solve_time_network": "disabled",
261
+ "python_wrapper_only": False,
262
+ "note": "This task audits real official software documentation/source staged offline; Python helpers only validate outputs.",
263
+ })
264
+ with (software_dir / "command_examples.csv").open("w", encoding="utf-8", newline="") as handle:
265
+ writer = csv.DictWriter(handle, fieldnames=["term", "kind", "software_name"])
266
+ writer.writeheader()
267
+ for term in spec_payload.get("code_terms") or []:
268
+ writer.writerow({"term": str(term), "kind": "code_term", "software_name": software_name})
269
+ for exe in executables:
270
+ writer.writerow({"term": exe, "kind": "executable", "software_name": software_name})
271
+
272
+
273
+ def software_tutorial_outputs(input_dir: Path, output_dir: Path) -> None:
274
+ spec = _read_json(input_dir / "analysis_spec.json")
275
+ source_file = str(spec["source_file"])
276
+ source_text = (input_dir / source_file).read_text(encoding="utf-8", errors="replace")
277
+ clean_text = _clean_markup_text(source_text)
278
+ headings = _extract_headings(source_text)
279
+ code_blocks = _extract_code_blocks(source_text)
280
+ window_rows: list[dict[str, str]] = []
281
+ with (input_dir / str(spec["input_file"])).open("r", encoding="utf-8", newline="") as handle:
282
+ window_rows = list(csv.DictReader(handle))
283
+ output_dir.mkdir(parents=True, exist_ok=True)
284
+
285
+ with (output_dir / "section_outline.csv").open("w", encoding="utf-8", newline="") as handle:
286
+ fieldnames = ["rank", "heading_level", "heading_text", "slug"]
287
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
288
+ writer.writeheader()
289
+ for rank, heading in enumerate(headings, start=1):
290
+ writer.writerow({
291
+ "rank": rank,
292
+ "heading_level": heading["level"],
293
+ "heading_text": heading["text"],
294
+ "slug": _heading_slug(heading["text"]),
295
+ })
296
+
297
+ all_terms: list[tuple[str, str]] = []
298
+ for term in spec.get("required_terms") or []:
299
+ all_terms.append((str(term), "required"))
300
+ for term in spec.get("code_terms") or []:
301
+ if str(term) not in {item[0] for item in all_terms}:
302
+ all_terms.append((str(term), "code"))
303
+ code_text = "\n".join(code_blocks)
304
+ term_rows: list[dict[str, Any]] = []
305
+ for term, category in all_terms:
306
+ term_rows.append({
307
+ "term": term,
308
+ "category": category,
309
+ "count": _term_count(clean_text, term),
310
+ "first_window": _first_window(window_rows, term),
311
+ "appears_in_code": "true" if term.lower() in code_text.lower() else "false",
312
+ })
313
+ term_rows.sort(key=lambda row: (str(row["category"]), str(row["term"]).lower()))
314
+ with (output_dir / "keyword_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
315
+ fieldnames = ["term", "category", "count", "first_window", "appears_in_code"]
316
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
317
+ writer.writeheader()
318
+ for row in term_rows:
319
+ writer.writerow(row)
320
+
321
+ code_terms = [str(item).lower() for item in spec.get("code_terms") or []]
322
+ with (output_dir / "code_block_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
323
+ fieldnames = ["block_index", "line_count", "char_count", "contains_code_term", "first_line"]
324
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
325
+ writer.writeheader()
326
+ for index, block in enumerate(code_blocks, start=1):
327
+ lines = [line for line in block.splitlines() if line.strip()]
328
+ contains = any(term and term in block.lower() for term in code_terms)
329
+ writer.writerow({
330
+ "block_index": index,
331
+ "line_count": len(lines),
332
+ "char_count": len(block),
333
+ "contains_code_term": "true" if contains else "false",
334
+ "first_line": lines[0][:200] if lines else "",
335
+ })
336
+
337
+ keyword_total = sum(int(row["count"]) for row in term_rows)
338
+ required_found = sum(1 for row in term_rows if row["category"] == "required" and int(row["count"]) > 0)
339
+ top_keyword = None
340
+ if term_rows:
341
+ top = max(term_rows, key=lambda row: (int(row["count"]), str(row["term"]).lower()))
342
+ top_keyword = str(top["term"]) if int(top["count"]) > 0 else None
343
+ summary = {
344
+ "software_name": str(spec.get("software_name") or ""),
345
+ "source_file": source_file,
346
+ "window_count": len(window_rows),
347
+ "heading_count": len(headings),
348
+ "code_block_count": len(code_blocks),
349
+ "keyword_total": keyword_total,
350
+ "unique_required_terms_found": required_found,
351
+ "top_keyword": top_keyword,
352
+ "source_character_count": len(clean_text),
353
+ }
354
+ _write_json(output_dir / "summary.json", summary)
355
+
356
+
357
+ def geojson_outputs(input_dir: Path, output_dir: Path) -> None:
358
+ payload = _read_json(input_dir / "events.geojson")
359
+ features = payload.get("features") if isinstance(payload, dict) else []
360
+ rows: list[dict[str, Any]] = []
361
+ for feature in features:
362
+ if not isinstance(feature, dict):
363
+ continue
364
+ props = feature.get("properties") if isinstance(feature.get("properties"), dict) else {}
365
+ geom = feature.get("geometry") if isinstance(feature.get("geometry"), dict) else {}
366
+ coords = geom.get("coordinates") if isinstance(geom.get("coordinates"), list) else []
367
+ mag = props.get("mag")
368
+ try:
369
+ mag = float(mag) if mag is not None else None
370
+ except (TypeError, ValueError):
371
+ mag = None
372
+ depth = None
373
+ if len(coords) >= 3:
374
+ try:
375
+ depth = float(coords[2])
376
+ except (TypeError, ValueError):
377
+ depth = None
378
+ time_ms = props.get("time")
379
+ date_utc = "unknown"
380
+ if isinstance(time_ms, (int, float)):
381
+ date_utc = dt.datetime.utcfromtimestamp(time_ms / 1000).date().isoformat()
382
+ rows.append({
383
+ "id": feature.get("id") or props.get("code") or "",
384
+ "mag": mag,
385
+ "depth": depth,
386
+ "place": str(props.get("place") or ""),
387
+ "region": _stable_region(str(props.get("place") or "")),
388
+ "date_utc": date_utc,
389
+ "feature": feature,
390
+ })
391
+ output_dir.mkdir(parents=True, exist_ok=True)
392
+ mags = [row["mag"] for row in rows if row["mag"] is not None]
393
+ depths = [row["depth"] for row in rows if row["depth"] is not None]
394
+ summary = {
395
+ "event_count": len(rows),
396
+ "magnitude": {
397
+ "count": len(mags),
398
+ "min": round(min(mags), 6) if mags else None,
399
+ "max": round(max(mags), 6) if mags else None,
400
+ "mean": round(statistics.fmean(mags), 6) if mags else None,
401
+ "median": round(statistics.median(mags), 6) if mags else None,
402
+ },
403
+ "depth_km": {
404
+ "count": len(depths),
405
+ "min": round(min(depths), 6) if depths else None,
406
+ "max": round(max(depths), 6) if depths else None,
407
+ "mean": round(statistics.fmean(depths), 6) if depths else None,
408
+ },
409
+ "time_range_utc": {
410
+ "start_date": min((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
411
+ "end_date": max((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
412
+ },
413
+ "quality_flags": {
414
+ "missing_magnitude": sum(1 for row in rows if row["mag"] is None),
415
+ "missing_depth": sum(1 for row in rows if row["depth"] is None),
416
+ "unique_regions": len({row["region"] for row in rows}),
417
+ },
418
+ }
419
+ _write_json(output_dir / "summary.json", summary)
420
+
421
+ grouped: dict[tuple[str, str], list[dict[str, Any]]] = {}
422
+ for row in rows:
423
+ grouped.setdefault((_mag_bin(row["mag"]), _depth_bin(row["depth"])), []).append(row)
424
+ with (output_dir / "magnitude_depth_bins.csv").open("w", encoding="utf-8", newline="") as handle:
425
+ fieldnames = ["mag_bin", "depth_bin", "event_count", "mean_magnitude", "max_magnitude"]
426
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
427
+ writer.writeheader()
428
+ for (mag_bin, depth_bin), group in sorted(grouped.items()):
429
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
430
+ writer.writerow({
431
+ "mag_bin": mag_bin,
432
+ "depth_bin": depth_bin,
433
+ "event_count": len(group),
434
+ "mean_magnitude": f"{statistics.fmean(group_mags):.6f}" if group_mags else "",
435
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
436
+ })
437
+
438
+ daily: dict[tuple[str, str], list[dict[str, Any]]] = {}
439
+ for row in rows:
440
+ daily.setdefault((row["region"], row["date_utc"]), []).append(row)
441
+ with (output_dir / "region_daily_counts.csv").open("w", encoding="utf-8", newline="") as handle:
442
+ fieldnames = ["region", "date_utc", "event_count", "max_magnitude"]
443
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
444
+ writer.writeheader()
445
+ for (region, date_utc), group in sorted(daily.items()):
446
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
447
+ writer.writerow({
448
+ "region": region,
449
+ "date_utc": date_utc,
450
+ "event_count": len(group),
451
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
452
+ })
453
+
454
+ filtered = [row["feature"] for row in rows if row["mag"] is not None and row["mag"] >= 4.5]
455
+ _write_json(output_dir / "filtered_events.geojson", {"type": "FeatureCollection", "features": filtered})
456
+
457
+
458
+ def csv_time_series_outputs(input_dir: Path, output_dir: Path) -> None:
459
+ spec = _read_json(input_dir / "analysis_spec.json")
460
+ csv_name = spec["input_file"]
461
+ date_column = spec["date_column"]
462
+ value_column = spec["value_column"]
463
+ rows: list[dict[str, Any]] = []
464
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
465
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
466
+ reader = csv.DictReader(lines)
467
+ for item in reader:
468
+ raw_date = item.get(date_column, "")
469
+ try:
470
+ day = dt.date.fromisoformat(raw_date)
471
+ except ValueError:
472
+ if raw_date.isdigit() and len(raw_date) == 4:
473
+ day = dt.date(int(raw_date), 1, 1)
474
+ else:
475
+ continue
476
+ raw_value = item.get(value_column, "")
477
+ try:
478
+ value = float(raw_value)
479
+ except (TypeError, ValueError):
480
+ value = None
481
+ rows.append({"date": day, "value": value, "raw": item})
482
+ rows.sort(key=lambda item: item["date"])
483
+ valid = [row for row in rows if row["value"] is not None]
484
+ output_dir.mkdir(parents=True, exist_ok=True)
485
+ monthly: dict[str, list[float]] = {}
486
+ for row in valid:
487
+ key = row["date"].strftime("%Y-%m")
488
+ monthly.setdefault(key, []).append(row["value"])
489
+ with (output_dir / "monthly_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
490
+ fieldnames = ["month", "count", "mean", "median", "min", "max"]
491
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
492
+ writer.writeheader()
493
+ for month, values in sorted(monthly.items()):
494
+ writer.writerow({
495
+ "month": month,
496
+ "count": len(values),
497
+ "mean": f"{statistics.fmean(values):.6f}",
498
+ "median": f"{statistics.median(values):.6f}",
499
+ "min": f"{min(values):.6f}",
500
+ "max": f"{max(values):.6f}",
501
+ })
502
+ existing_dates = {row["date"] for row in rows}
503
+ missing_days: list[dt.date] = []
504
+ if rows:
505
+ day = rows[0]["date"]
506
+ while day <= rows[-1]["date"]:
507
+ if day.weekday() < 5 and day not in existing_dates:
508
+ missing_days.append(day)
509
+ day += dt.timedelta(days=1)
510
+ with (output_dir / "missing_business_days.csv").open("w", encoding="utf-8", newline="") as handle:
511
+ writer = csv.DictWriter(handle, fieldnames=["date"])
512
+ writer.writeheader()
513
+ for day in missing_days:
514
+ writer.writerow({"date": day.isoformat()})
515
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
516
+ writer = csv.DictWriter(handle, fieldnames=["date", "value"])
517
+ writer.writeheader()
518
+ for row in valid:
519
+ writer.writerow({"date": row["date"].isoformat(), "value": f"{row['value']:.6f}"})
520
+ values = [row["value"] for row in valid]
521
+ summary = {
522
+ "series_name": spec.get("series_name", value_column),
523
+ "row_count": len(rows),
524
+ "valid_value_count": len(valid),
525
+ "missing_value_count": len(rows) - len(valid),
526
+ "missing_business_day_count": len(missing_days),
527
+ "date_range": {
528
+ "start": rows[0]["date"].isoformat() if rows else None,
529
+ "end": rows[-1]["date"].isoformat() if rows else None,
530
+ },
531
+ "value": {
532
+ "min": round(min(values), 6) if values else None,
533
+ "max": round(max(values), 6) if values else None,
534
+ "mean": round(statistics.fmean(values), 6) if values else None,
535
+ "median": round(statistics.median(values), 6) if values else None,
536
+ },
537
+ }
538
+ _write_json(output_dir / "summary.json", summary)
539
+
540
+
541
+ def _aggregate_annual_rows(rows: list[dict[str, Any]], aggregation: str) -> list[dict[str, Any]]:
542
+ if not aggregation:
543
+ return rows
544
+ grouped: dict[int, list[float]] = {}
545
+ for row in rows:
546
+ grouped.setdefault(int(row["year"]), []).append(float(row["value"]))
547
+ aggregated: list[dict[str, Any]] = []
548
+ for year, values in sorted(grouped.items()):
549
+ if aggregation == "mean":
550
+ value = statistics.fmean(values)
551
+ elif aggregation == "median":
552
+ value = statistics.median(values)
553
+ elif aggregation == "max":
554
+ value = max(values)
555
+ elif aggregation == "min":
556
+ value = min(values)
557
+ elif aggregation == "count":
558
+ value = float(len(values))
559
+ else:
560
+ raise ValueError(f"unsupported year_aggregation for csv_annual_series_audit: {aggregation}")
561
+ aggregated.append({"year": year, "value": value})
562
+ return aggregated
563
+
564
+
565
+ def csv_annual_series_outputs(input_dir: Path, output_dir: Path) -> None:
566
+ spec = _read_json(input_dir / "analysis_spec.json")
567
+ csv_name = spec["input_file"]
568
+ year_column = spec["year_column"]
569
+ value_column = spec["value_column"]
570
+ # Optional: derive the year from an ISO-8601 timestamp column (e.g. USGS FDSN
571
+ # `time` = "2022-06-29T19:17:58.929Z") instead of parsing an integer year.
572
+ # When unset, behaviour is unchanged (integer year column).
573
+ year_from_timestamp = bool(spec.get("year_from_timestamp", False))
574
+ year_aggregation = str(spec.get("year_aggregation") or "").strip().lower()
575
+ raw_rows: list[dict[str, Any]] = []
576
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
577
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
578
+ reader = csv.DictReader(lines)
579
+ for item in reader:
580
+ raw_year = str(item.get(year_column, "")).strip()
581
+ try:
582
+ if year_from_timestamp:
583
+ # Take the leading 4-digit year of an ISO timestamp/date.
584
+ match = re.match(r"\s*(\d{4})", raw_year)
585
+ if not match:
586
+ continue
587
+ year = int(match.group(1))
588
+ else:
589
+ year = int(raw_year)
590
+ value = float(str(item.get(value_column, "")).strip())
591
+ except (TypeError, ValueError):
592
+ continue
593
+ raw_rows.append({"year": year, "value": value})
594
+ rows = _aggregate_annual_rows(raw_rows, year_aggregation)
595
+ rows.sort(key=lambda item: item["year"])
596
+ output_dir.mkdir(parents=True, exist_ok=True)
597
+ decades: dict[str, list[float]] = {}
598
+ for row in rows:
599
+ decade_start = (row["year"] // 10) * 10
600
+ decades.setdefault(f"{decade_start}s", []).append(row["value"])
601
+ with (output_dir / "decadal_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
602
+ fieldnames = [
603
+ "decade_start",
604
+ "start_year",
605
+ "end_year",
606
+ "count",
607
+ "mean_departure",
608
+ "median_departure",
609
+ "min_departure",
610
+ "max_departure",
611
+ "sample_std_departure",
612
+ ]
613
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
614
+ writer.writeheader()
615
+ for decade, values in sorted(decades.items()):
616
+ decade_start = int(decade.rstrip("s"))
617
+ years = [row["year"] for row in rows if decade_start <= row["year"] <= decade_start + 9]
618
+ writer.writerow({
619
+ "decade_start": decade_start,
620
+ "start_year": min(years),
621
+ "end_year": max(years),
622
+ "count": len(values),
623
+ "mean_departure": f"{statistics.fmean(values):.6f}",
624
+ "median_departure": f"{statistics.median(values):.6f}",
625
+ "min_departure": f"{min(values):.6f}",
626
+ "max_departure": f"{max(values):.6f}",
627
+ "sample_std_departure": f"{statistics.stdev(values):.6f}" if len(values) >= 2 else "",
628
+ })
629
+ threshold = float(spec.get("threshold", 1.0))
630
+ with (output_dir / "threshold_years.csv").open("w", encoding="utf-8", newline="") as handle:
631
+ fieldnames = ["rank", "year", "departure", "threshold"]
632
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
633
+ writer.writeheader()
634
+ rank = 1
635
+ for row in rows:
636
+ if row["value"] >= threshold:
637
+ writer.writerow({
638
+ "rank": rank,
639
+ "year": row["year"],
640
+ "departure": f"{row['value']:.6f}",
641
+ "threshold": f"{threshold:.6f}",
642
+ })
643
+ rank += 1
644
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
645
+ writer = csv.DictWriter(handle, fieldnames=["Year", "Departure from Average"])
646
+ writer.writeheader()
647
+ for row in rows:
648
+ writer.writerow({"Year": row["year"], "Departure from Average": f"{row['value']:.6f}"})
649
+ values = [row["value"] for row in rows]
650
+ threshold_rows = [row for row in rows if row["value"] >= threshold]
651
+ min_row = min(rows, key=lambda row: row["value"]) if rows else None
652
+ max_row = max(rows, key=lambda row: row["value"]) if rows else None
653
+ summary = {
654
+ "series_name": spec.get("series_name", value_column),
655
+ "row_count": len(rows),
656
+ "year_range": {
657
+ "start": rows[0]["year"] if rows else None,
658
+ "end": rows[-1]["year"] if rows else None,
659
+ },
660
+ "value": {
661
+ "min": round(min(values), 6) if values else None,
662
+ "min_year": min_row["year"] if min_row else None,
663
+ "max": round(max(values), 6) if values else None,
664
+ "max_year": max_row["year"] if max_row else None,
665
+ "mean": round(statistics.fmean(values), 6) if values else None,
666
+ "median": round(statistics.median(values), 6) if values else None,
667
+ "sample_std": round(statistics.stdev(values), 6) if len(values) >= 2 else None,
668
+ },
669
+ "threshold": threshold,
670
+ "years_at_or_above_threshold": len(threshold_rows),
671
+ "first_threshold_year": threshold_rows[0]["year"] if threshold_rows else None,
672
+ "last_threshold_year": threshold_rows[-1]["year"] if threshold_rows else None,
673
+ }
674
+ if year_aggregation:
675
+ summary["year_aggregation"] = year_aggregation
676
+ summary["source_row_count"] = len(raw_rows)
677
+ _write_json(output_dir / "summary.json", summary)
678
+
679
+
20260904-005149-science-environment-cee684/solution/internal/workflow_runtime.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ from pathlib import Path
2
+ from workflow_adapters_runtime import csv_annual_series_outputs
3
+
4
+ def run_reference(input_dir: Path, output_dir: Path) -> None:
5
+ csv_annual_series_outputs(input_dir, output_dir)
20260904-005149-science-environment-cee684/solution/reference_solution.py ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import argparse
4
+ import sys
5
+ from pathlib import Path
6
+
7
+
8
+ def main() -> int:
9
+ parser = argparse.ArgumentParser()
10
+ parser.add_argument("--input", required=True)
11
+ parser.add_argument("--output", required=True)
12
+ args = parser.parse_args()
13
+ sys.path.insert(0, str(Path(__file__).resolve().parent / "internal"))
14
+ from workflow_runtime import run_reference
15
+ run_reference(Path(args.input), Path(args.output))
16
+ return 0
17
+
18
+
19
+ if __name__ == "__main__":
20
+ raise SystemExit(main())
20260904-005149-science-environment-cee684/solution/solve.sh ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ set -euo pipefail
3
+ INPUT_DIR="${1:-/workspace/input}"
4
+ OUTPUT_DIR="${2:-/workspace/output}"
5
+ SOLUTION_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
6
+ mkdir -p "${OUTPUT_DIR}"
7
+ python "${SOLUTION_DIR}/reference_solution.py" --input "${INPUT_DIR}" --output "${OUTPUT_DIR}"
20260904-005149-science-environment-cee684/task.toml ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version = "1.3"
2
+ artifacts = ["/workspace/output/summary.json", "/workspace/output/decadal_statistics.csv", "/workspace/output/threshold_years.csv", "/workspace/output/clean_series.csv"]
3
+
4
+ [task]
5
+ name = "ale-cli/20260904-005149-science-environment-cee684"
6
+ description = "Audit the real USGS FDSN Andes subduction zone earthquake catalog for 1970-1988-1970-1988 (events with magnitude >= 4.5) and produce reproducible annual annual mean focal depth (km) statistics, decadal aggregates, threshold-year checks, and a cleaned local series."
7
+ keywords = ["agents-last-exam", "ale-cli", "csv-annual-series-audit", "deterministic-evaluator", "linux-cli", "science-environment"]
8
+
9
+ [metadata]
10
+ benchmark = "ale_cli"
11
+ source_format = "ale_cli_official_style"
12
+ converter = "ale_cli_to_harbor"
13
+ source_task_key = "20260904-005149_science_environment_cee684"
14
+ task_id = "science_environment/usgs_andes_annual_focal_depth_audit"
15
+ domain = "science_environment"
16
+ task = "usgs_andes_annual_focal_depth_audit"
17
+ variant = "base"
18
+ oracle_available = true
19
+ scoring_method = "deterministic_hidden_evaluator"
20
+ required_outputs = ["summary.json", "decadal_statistics.csv", "threshold_years.csv", "clean_series.csv"]
21
+
22
+ [agent]
23
+ timeout_sec = 1800
24
+
25
+ [verifier]
26
+ timeout_sec = 600
27
+ environment_mode = "separate"
28
+
29
+ [verifier.environment]
30
+ build_timeout_sec = 900.0
31
+ network_mode = "public"
32
+ os = "linux"
33
+ cpus = 1
34
+ memory_mb = 2048
35
+ storage_mb = 4096
36
+ gpus = 0
37
+
38
+ [environment]
39
+ build_timeout_sec = 1200.0
40
+ network_mode = "public"
41
+ os = "linux"
42
+ cpus = 2
43
+ memory_mb = 4096
44
+ storage_mb = 10240
45
+ gpus = 0
20260904-005149-science-environment-cee684/tests/Dockerfile ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ FROM python:3.12-slim
2
+
3
+ WORKDIR /workspace
4
+
5
+ COPY . /tests
6
+ RUN chmod +x /tests/test.sh
20260904-005149-science-environment-cee684/tests/evaluator_runner.py ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ from __future__ import annotations
3
+
4
+ import argparse
5
+ import importlib.util
6
+ import json
7
+ from pathlib import Path
8
+ from typing import Any
9
+
10
+
11
+ def read_json(path: Path) -> Any:
12
+ return json.loads(path.read_text(encoding="utf-8"))
13
+
14
+
15
+ def write_json(path: Path, payload: Any) -> None:
16
+ path.parent.mkdir(parents=True, exist_ok=True)
17
+ path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
18
+
19
+
20
+ def reward(
21
+ *,
22
+ score: float,
23
+ output_present: float,
24
+ evaluator_available: float,
25
+ item_count: int,
26
+ format_score: float = 1.0,
27
+ ) -> dict[str, float]:
28
+ clipped = max(0.0, min(1.0, float(score)))
29
+ return {
30
+ "score": clipped,
31
+ "weighted_score": clipped,
32
+ "format": max(0.0, min(1.0, float(format_score))),
33
+ "output_present": max(0.0, min(1.0, float(output_present))),
34
+ "evaluator_available": max(0.0, min(1.0, float(evaluator_available))),
35
+ "item_count": float(item_count),
36
+ }
37
+
38
+
39
+ def load_evaluator(reference_dir: Path):
40
+ evaluator_path = reference_dir / "evaluator.py"
41
+ spec = importlib.util.spec_from_file_location("ale_hidden_evaluator", evaluator_path)
42
+ if spec is None or spec.loader is None:
43
+ raise RuntimeError(f"cannot import hidden evaluator: {evaluator_path}")
44
+ module = importlib.util.module_from_spec(spec)
45
+ spec.loader.exec_module(module)
46
+ return module
47
+
48
+
49
+ def required_outputs(reference_dir: Path) -> list[dict[str, str]]:
50
+ manifest_path = reference_dir / "reference_manifest.json"
51
+ if not manifest_path.is_file():
52
+ return []
53
+ manifest = read_json(manifest_path)
54
+ rows = manifest.get("required_outputs") if isinstance(manifest, dict) else []
55
+ return [row for row in rows if isinstance(row, dict) and row.get("path")]
56
+
57
+
58
+ def main() -> int:
59
+ parser = argparse.ArgumentParser(description="Run a hidden ALE CLI evaluator in Harbor.")
60
+ parser.add_argument("--output-dir", type=Path, required=True)
61
+ parser.add_argument("--reference-dir", type=Path, required=True)
62
+ parser.add_argument("--reward-json", type=Path, required=True)
63
+ parser.add_argument("--details-json", type=Path, required=True)
64
+ args = parser.parse_args()
65
+
66
+ required = required_outputs(args.reference_dir)
67
+ present = [row for row in required if (args.output_dir / str(row["path"])).is_file()]
68
+ item_count = len(required)
69
+ if not required:
70
+ write_json(args.reward_json, reward(score=0.0, output_present=0.0, evaluator_available=0.0, item_count=0, format_score=0.0))
71
+ write_json(args.details_json, {"ok": False, "error": "reference_manifest.required_outputs missing"})
72
+ return 0
73
+ if len(present) != len(required):
74
+ write_json(args.reward_json, reward(score=0.0, output_present=0.0, evaluator_available=1.0, item_count=item_count))
75
+ write_json(args.details_json, {"ok": True, "reason": "missing required outputs", "present": [row["path"] for row in present]})
76
+ return 0
77
+ try:
78
+ module = load_evaluator(args.reference_dir)
79
+ result = module.evaluate(args.output_dir, args.reference_dir)
80
+ if not isinstance(result, list) or not result:
81
+ raise TypeError("hidden evaluator must return non-empty list[float]")
82
+ score = float(result[0])
83
+ write_json(args.reward_json, reward(score=score, output_present=1.0, evaluator_available=1.0, item_count=item_count))
84
+ write_json(args.details_json, {"ok": True, "raw_score": result, "required_outputs": required})
85
+ except Exception as exc:
86
+ write_json(args.reward_json, reward(score=0.0, output_present=1.0, evaluator_available=0.0, item_count=item_count))
87
+ write_json(args.details_json, {"ok": False, "error": f"{type(exc).__name__}: {exc}", "required_outputs": required})
88
+ return 0
89
+
90
+
91
+ if __name__ == "__main__":
92
+ raise SystemExit(main())
20260904-005149-science-environment-cee684/tests/reference/__pycache__/evaluator.cpython-312.pyc ADDED
Binary file (6.22 kB). View file
 
20260904-005149-science-environment-cee684/tests/reference/evaluator.py ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import csv
4
+ import json
5
+ import math
6
+ from pathlib import Path
7
+ from typing import Any
8
+
9
+
10
+ def _load_json(path: Path) -> Any:
11
+ with path.open("r", encoding="utf-8") as handle:
12
+ return json.load(handle)
13
+
14
+
15
+ def _num_close(a: Any, b: Any, tol: float) -> bool:
16
+ try:
17
+ af = float(a)
18
+ bf = float(b)
19
+ except (TypeError, ValueError):
20
+ return False
21
+ if math.isnan(af) and math.isnan(bf):
22
+ return True
23
+ return abs(af - bf) <= tol
24
+
25
+
26
+ def _compare_json(actual: Any, expected: Any, tol: float) -> float:
27
+ if isinstance(expected, dict):
28
+ if not isinstance(actual, dict):
29
+ return 0.0
30
+ if not expected:
31
+ return 1.0
32
+ scores = []
33
+ for key, exp_value in expected.items():
34
+ if key not in actual:
35
+ scores.append(0.0)
36
+ else:
37
+ scores.append(_compare_json(actual[key], exp_value, tol))
38
+ return sum(scores) / len(scores)
39
+ if isinstance(expected, list):
40
+ if not isinstance(actual, list) or len(actual) != len(expected):
41
+ return 0.0
42
+ if not expected:
43
+ return 1.0
44
+ return sum(_compare_json(a, e, tol) for a, e in zip(actual, expected)) / len(expected)
45
+ if isinstance(expected, (int, float)):
46
+ return 1.0 if _num_close(actual, expected, tol) else 0.0
47
+ return 1.0 if actual == expected else 0.0
48
+
49
+
50
+ def _read_csv(path: Path) -> list[dict[str, str]]:
51
+ with path.open("r", encoding="utf-8", newline="") as handle:
52
+ return list(csv.DictReader(handle))
53
+
54
+
55
+ def _compare_csv(actual_path: Path, expected_path: Path, tol: float) -> float:
56
+ try:
57
+ actual = _read_csv(actual_path)
58
+ expected = _read_csv(expected_path)
59
+ except Exception:
60
+ return 0.0
61
+ if len(actual) != len(expected):
62
+ return 0.0
63
+ if not expected:
64
+ return 1.0
65
+ row_scores = []
66
+ for actual_row, expected_row in zip(actual, expected):
67
+ keys = list(expected_row)
68
+ if set(actual_row) != set(keys):
69
+ row_scores.append(0.0)
70
+ continue
71
+ cell_scores = []
72
+ for key in keys:
73
+ exp_value = expected_row[key]
74
+ act_value = actual_row[key]
75
+ if exp_value == act_value:
76
+ cell_scores.append(1.0)
77
+ elif _num_close(act_value, exp_value, tol):
78
+ cell_scores.append(1.0)
79
+ else:
80
+ cell_scores.append(0.0)
81
+ row_scores.append(sum(cell_scores) / len(cell_scores) if cell_scores else 1.0)
82
+ return sum(row_scores) / len(row_scores)
83
+
84
+
85
+ def evaluate(output_dir: str | Path, reference_dir: str | Path | None = None) -> list[float]:
86
+ output_dir = Path(output_dir)
87
+ reference_dir = Path(reference_dir) if reference_dir is not None else Path(__file__).resolve().parent
88
+ manifest = _load_json(reference_dir / "reference_manifest.json")
89
+ tolerance = float(manifest.get("numeric_tolerance", 1e-9))
90
+ required = manifest.get("required_outputs") or []
91
+ if not required:
92
+ return [0.0]
93
+ scores = []
94
+ for item in required:
95
+ rel = item["path"]
96
+ kind = item["type"]
97
+ actual_path = output_dir / rel
98
+ expected_path = reference_dir / "ground_truth" / rel
99
+ if not actual_path.is_file() or not expected_path.is_file():
100
+ scores.append(0.0)
101
+ continue
102
+ try:
103
+ if kind == "json":
104
+ scores.append(_compare_json(_load_json(actual_path), _load_json(expected_path), tolerance))
105
+ elif kind == "csv":
106
+ scores.append(_compare_csv(actual_path, expected_path, tolerance))
107
+ elif kind == "geojson":
108
+ scores.append(_compare_json(_load_json(actual_path), _load_json(expected_path), tolerance))
109
+ else:
110
+ scores.append(1.0 if actual_path.read_bytes() == expected_path.read_bytes() else 0.0)
111
+ except Exception:
112
+ scores.append(0.0)
113
+ score = max(0.0, min(1.0, sum(scores) / len(scores)))
114
+ return [score]
20260904-005149-science-environment-cee684/tests/reference/ground_truth/clean_series.csv ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Year,Departure from Average
2
+ 1970,48.274699
3
+ 1971,58.431169
4
+ 1972,73.626829
5
+ 1973,70.027108
6
+ 1974,73.080247
7
+ 1975,77.405738
8
+ 1976,81.280822
9
+ 1977,73.045455
10
+ 1978,100.060870
11
+ 1979,77.753709
12
+ 1980,88.443662
13
+ 1981,85.402239
14
+ 1982,92.615226
15
+ 1983,97.826855
16
+ 1984,106.942248
17
+ 1985,73.270323
18
+ 1986,88.672381
19
+ 1987,81.799398
20
+ 1988,90.250000
20260904-005149-science-environment-cee684/tests/reference/ground_truth/decadal_statistics.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ decade_start,start_year,end_year,count,mean_departure,median_departure,min_departure,max_departure,sample_std_departure
2
+ 1970,1970,1979,10,73.298665,73.353538,48.274699,100.060870,13.641480
3
+ 1980,1980,1988,9,89.469148,88.672381,73.270323,106.942248,9.520512
20260904-005149-science-environment-cee684/tests/reference/ground_truth/summary.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "series_name": "USGS FDSN Andes subduction zone 1970-1988-1970-1988 earthquake annual mean focal depth (km) (min mag 4.5)",
3
+ "row_count": 19,
4
+ "year_range": {
5
+ "start": 1970,
6
+ "end": 1988
7
+ },
8
+ "value": {
9
+ "min": 48.274699,
10
+ "min_year": 1970,
11
+ "max": 106.942248,
12
+ "max_year": 1984,
13
+ "mean": 80.958367,
14
+ "median": 81.280822,
15
+ "sample_std": 14.217638
16
+ },
17
+ "threshold": 70.0,
18
+ "years_at_or_above_threshold": 17,
19
+ "first_threshold_year": 1972,
20
+ "last_threshold_year": 1988,
21
+ "year_aggregation": "mean",
22
+ "source_row_count": 5070
23
+ }
20260904-005149-science-environment-cee684/tests/reference/ground_truth/threshold_years.csv ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ rank,year,departure,threshold
2
+ 1,1972,73.626829,70.000000
3
+ 2,1973,70.027108,70.000000
4
+ 3,1974,73.080247,70.000000
5
+ 4,1975,77.405738,70.000000
6
+ 5,1976,81.280822,70.000000
7
+ 6,1977,73.045455,70.000000
8
+ 7,1978,100.060870,70.000000
9
+ 8,1979,77.753709,70.000000
10
+ 9,1980,88.443662,70.000000
11
+ 10,1981,85.402239,70.000000
12
+ 11,1982,92.615226,70.000000
13
+ 12,1983,97.826855,70.000000
14
+ 13,1984,106.942248,70.000000
15
+ 14,1985,73.270323,70.000000
16
+ 15,1986,88.672381,70.000000
17
+ 16,1987,81.799398,70.000000
18
+ 17,1988,90.250000,70.000000
20260904-005149-science-environment-cee684/tests/reference/internal/workflow_adapters_runtime.py ADDED
@@ -0,0 +1,679 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import csv
4
+ import datetime as dt
5
+ import html
6
+ import json
7
+ import math
8
+ import re
9
+ import shutil
10
+ import statistics
11
+ from pathlib import Path
12
+ from typing import Any
13
+
14
+
15
+ def _read_json(path: Path) -> Any:
16
+ with path.open("r", encoding="utf-8") as handle:
17
+ return json.load(handle)
18
+
19
+
20
+ def _write_json(path: Path, payload: Any) -> None:
21
+ path.parent.mkdir(parents=True, exist_ok=True)
22
+ with path.open("w", encoding="utf-8") as handle:
23
+ json.dump(payload, handle, ensure_ascii=False, indent=2)
24
+ handle.write("\n")
25
+
26
+ def _stable_region(place: str) -> str:
27
+ if not place:
28
+ return "unknown"
29
+ if "," in place:
30
+ return place.split(",")[-1].strip() or place.strip()
31
+ parts = place.split()
32
+ return " ".join(parts[-2:]) if len(parts) >= 2 else place.strip()
33
+
34
+
35
+ def _mag_bin(value: float | None) -> str:
36
+ if value is None or math.isnan(value):
37
+ return "unknown"
38
+ lower = math.floor(value)
39
+ upper = lower + 1
40
+ return f"{lower:.0f}-{upper:.0f}"
41
+
42
+
43
+ def _depth_bin(value: float | None) -> str:
44
+ if value is None or math.isnan(value):
45
+ return "unknown"
46
+ if value < 0:
47
+ return "<0"
48
+ if value < 10:
49
+ return "0-10"
50
+ if value < 30:
51
+ return "10-30"
52
+ if value < 70:
53
+ return "30-70"
54
+ if value < 300:
55
+ return "70-300"
56
+ return "300+"
57
+
58
+
59
+ def _clean_markup_text(raw: str) -> str:
60
+ text = re.sub(r"(?is)<(script|style|noscript)[^>]*>.*?</\1>", " ", raw)
61
+ text = re.sub(r"(?is)<br\s*/?>", "\n", text)
62
+ text = re.sub(r"(?is)</(p|div|li|tr|h[1-6]|pre)>", "\n", text)
63
+ text = re.sub(r"(?is)<[^>]+>", " ", text)
64
+ text = html.unescape(text)
65
+ text = re.sub(r"[ \t]+", " ", text)
66
+ text = re.sub(r"\n\s+", "\n", text)
67
+ return text.strip()
68
+
69
+
70
+ def _heading_slug(value: str) -> str:
71
+ slug = re.sub(r"[^a-z0-9]+", "_", value.lower()).strip("_")
72
+ return slug[:80] or "section"
73
+
74
+
75
+ def _extract_headings(raw: str) -> list[dict[str, str]]:
76
+ headings: list[dict[str, str]] = []
77
+ for match in re.finditer(r"(?is)<h([1-6])[^>]*>(.*?)</h\1>", raw):
78
+ text = _clean_markup_text(match.group(2))
79
+ if text:
80
+ headings.append({"level": f"h{match.group(1)}", "text": text})
81
+ if headings:
82
+ return headings
83
+ for line in raw.splitlines():
84
+ match = re.match(r"^(#{1,6})\s+(.+?)\s*$", line)
85
+ if match:
86
+ headings.append({"level": f"h{len(match.group(1))}", "text": match.group(2).strip()})
87
+ return headings
88
+
89
+
90
+ def _extract_code_blocks(raw: str) -> list[str]:
91
+ blocks: list[str] = []
92
+ for match in re.finditer(r"(?is)<pre[^>]*>(.*?)</pre>", raw):
93
+ text = _clean_markup_text(match.group(1))
94
+ if text:
95
+ blocks.append(text)
96
+ for match in re.finditer(r"(?is)<code[^>]*>(.*?)</code>", raw):
97
+ text = _clean_markup_text(match.group(1))
98
+ if text and text not in blocks:
99
+ blocks.append(text)
100
+ for match in re.finditer(r"(?s)```(?:[A-Za-z0-9_-]+)?\n(.*?)```", raw):
101
+ text = html.unescape(match.group(1).strip())
102
+ if text and text not in blocks:
103
+ blocks.append(text)
104
+ return blocks
105
+
106
+
107
+ def _window_rows(clean_text: str, *, width: int = 40, stride: int = 20) -> list[dict[str, Any]]:
108
+ words = re.findall(r"\S+", clean_text)
109
+ rows: list[dict[str, Any]] = []
110
+ if not words:
111
+ return rows
112
+ start = 0
113
+ while start < len(words):
114
+ end = min(start + width, len(words))
115
+ rows.append({
116
+ "window_id": len(rows) + 1,
117
+ "start_word": start,
118
+ "end_word": end,
119
+ "text": " ".join(words[start:end]),
120
+ })
121
+ if end == len(words):
122
+ break
123
+ start += stride
124
+ return rows
125
+
126
+
127
+ def _term_count(text: str, term: str) -> int:
128
+ term = term.strip()
129
+ if not term:
130
+ return 0
131
+ return len(re.findall(re.escape(term), text, flags=re.IGNORECASE))
132
+
133
+
134
+ def _first_window(windows: list[dict[str, Any]], term: str) -> str:
135
+ term_lower = term.lower()
136
+ for row in windows:
137
+ if term_lower in str(row.get("text") or "").lower():
138
+ return str(row.get("window_id") or "")
139
+ return ""
140
+
141
+
142
+ def _write_tutorial_public_inputs(input_dir: Path, source_path: Path, spec_payload: dict[str, Any]) -> None:
143
+ raw = source_path.read_text(encoding="utf-8", errors="replace")
144
+ # Tutorial examples may contain a maintainer's local Windows path. Keep
145
+ # the command semantics while preventing host-specific paths from entering
146
+ # the public task artifact.
147
+ sanitized_raw = re.sub(r"(?i)\b[A-Z]:\\(?:[^\\\r\n<>\"']+\\?)*[^\\\r\n<>\"']*", "./local.db", raw)
148
+ clean_text = _clean_markup_text(sanitized_raw)
149
+ windows = _window_rows(clean_text)
150
+ if len(windows) < 100:
151
+ raise ValueError(f"official tutorial source produced only {len(windows)} text windows; need at least 100")
152
+ required_terms = [str(item) for item in spec_payload.get("required_terms") or []]
153
+ code_terms = [str(item) for item in spec_payload.get("code_terms") or []]
154
+ input_dir.mkdir(parents=True, exist_ok=True)
155
+ destination = input_dir / str(spec_payload["source_file"])
156
+ # Always materialize the sanitized copy, including when the downloader
157
+ # already placed the source at the destination path.
158
+ destination.write_text(sanitized_raw, encoding="utf-8")
159
+ with (input_dir / str(spec_payload["input_file"])).open("w", encoding="utf-8", newline="") as handle:
160
+ fieldnames = ["window_id", "start_word", "end_word", "text", "contains_code_term", "matched_required_terms"]
161
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
162
+ writer.writeheader()
163
+ for row in windows:
164
+ text = str(row["text"])
165
+ matched = [term for term in required_terms if term.lower() in text.lower()]
166
+ contains_code = any(term.lower() in text.lower() for term in code_terms)
167
+ writer.writerow({
168
+ "window_id": row["window_id"],
169
+ "start_word": row["start_word"],
170
+ "end_word": row["end_word"],
171
+ "text": text,
172
+ "contains_code_term": "true" if contains_code else "false",
173
+ "matched_required_terms": "|".join(matched),
174
+ })
175
+ metadata = {
176
+ "software_name": spec_payload.get("software_name"),
177
+ "tutorial_url": spec_payload.get("tutorial_url"),
178
+ "source_urls": spec_payload.get("source_urls"),
179
+ "source_file": spec_payload.get("source_file"),
180
+ "window_file": spec_payload.get("input_file"),
181
+ "window_rule": spec_payload.get("window_rule"),
182
+ "required_terms": spec_payload.get("required_terms"),
183
+ "code_terms": spec_payload.get("code_terms"),
184
+ "note": "Provenance and public extraction rules only; scored aggregate values are intentionally not stored here.",
185
+ }
186
+ _write_json(input_dir / "tutorial_metadata.json", metadata)
187
+
188
+
189
+ def _source_context_text(path: Path) -> str:
190
+ try:
191
+ raw = path.read_text(encoding="utf-8", errors="replace")
192
+ except Exception:
193
+ return f"{path.name}\nsha256={_sha256_file(path)}\nsize_bytes={path.stat().st_size}\n"
194
+ if path.suffix.lower() in {".html", ".htm", ".md", ".txt"}:
195
+ return _clean_markup_text(raw)
196
+ return re.sub(r"\s+", " ", raw).strip()
197
+
198
+
199
+ def _ensure_public_input_footprint(input_dir: Path, source_paths: list[Path]) -> str:
200
+ current = _tree_file_bytes(input_dir)
201
+ if current >= MIN_PUBLIC_TASK_DATA_BYTES:
202
+ return f"public input footprint is {current} bytes; no context expansion needed"
203
+ context_path = input_dir / "source_context_windows.jsonl"
204
+ base = current - (context_path.stat().st_size if context_path.is_file() else 0)
205
+ context_path.unlink(missing_ok=True)
206
+ rows: list[dict[str, Any]] = []
207
+ for source in source_paths:
208
+ if not source.is_file() or source.resolve() == context_path.resolve():
209
+ continue
210
+ text = _source_context_text(source)
211
+ if not text:
212
+ continue
213
+ start = 0
214
+ width = 1800
215
+ stride = 900
216
+ while start < len(text):
217
+ end = min(len(text), start + width)
218
+ rows.append({
219
+ "source_file": source.relative_to(input_dir).as_posix() if source.is_relative_to(input_dir) else source.name,
220
+ "source_sha256": _sha256_file(source),
221
+ "start_char": start,
222
+ "end_char": end,
223
+ "text": text[start:end],
224
+ })
225
+ if end == len(text):
226
+ break
227
+ start += stride
228
+ if not rows:
229
+ raise ValueError("public input footprint expansion has no public source text")
230
+ input_dir.mkdir(parents=True, exist_ok=True)
231
+ rows_written = 0
232
+ target_extra = MIN_PUBLIC_TASK_DATA_BYTES - max(0, base)
233
+ with context_path.open("w", encoding="utf-8") as handle:
234
+ cycle = 0
235
+ while handle.tell() < target_extra or rows_written < 100:
236
+ row = dict(rows[rows_written % len(rows)])
237
+ row["window_id"] = rows_written + 1
238
+ row["cycle"] = cycle
239
+ handle.write(json.dumps(row, ensure_ascii=False) + "\n")
240
+ rows_written += 1
241
+ if rows_written % len(rows) == 0:
242
+ cycle += 1
243
+ handle.flush()
244
+ return f"wrote input/source_context_windows.jsonl with {rows_written} public-source windows and {context_path.stat().st_size} bytes"
245
+
246
+
247
+ def _write_software_manifest(software_dir: Path, spec_payload: dict[str, Any]) -> None:
248
+ if str(spec_payload.get("workflow_kind") or "") != "software_tutorial_doc_audit":
249
+ return
250
+ software_dir.mkdir(parents=True, exist_ok=True)
251
+ software_name = str(spec_payload.get("software_name") or "").strip()
252
+ executables = [str(item).strip() for item in spec_payload.get("executables") or [] if str(item).strip()]
253
+ system_packages = [str(item).strip() for item in spec_payload.get("required_system_packages") or [] if str(item).strip()]
254
+ source_urls = [str(item).strip() for item in spec_payload.get("source_urls") or [] if str(item).strip()]
255
+ _write_json(software_dir / "software_manifest.json", {
256
+ "software_name": software_name,
257
+ "executables": executables,
258
+ "required_system_packages": system_packages,
259
+ "official_sources": source_urls,
260
+ "solve_time_network": "disabled",
261
+ "python_wrapper_only": False,
262
+ "note": "This task audits real official software documentation/source staged offline; Python helpers only validate outputs.",
263
+ })
264
+ with (software_dir / "command_examples.csv").open("w", encoding="utf-8", newline="") as handle:
265
+ writer = csv.DictWriter(handle, fieldnames=["term", "kind", "software_name"])
266
+ writer.writeheader()
267
+ for term in spec_payload.get("code_terms") or []:
268
+ writer.writerow({"term": str(term), "kind": "code_term", "software_name": software_name})
269
+ for exe in executables:
270
+ writer.writerow({"term": exe, "kind": "executable", "software_name": software_name})
271
+
272
+
273
+ def software_tutorial_outputs(input_dir: Path, output_dir: Path) -> None:
274
+ spec = _read_json(input_dir / "analysis_spec.json")
275
+ source_file = str(spec["source_file"])
276
+ source_text = (input_dir / source_file).read_text(encoding="utf-8", errors="replace")
277
+ clean_text = _clean_markup_text(source_text)
278
+ headings = _extract_headings(source_text)
279
+ code_blocks = _extract_code_blocks(source_text)
280
+ window_rows: list[dict[str, str]] = []
281
+ with (input_dir / str(spec["input_file"])).open("r", encoding="utf-8", newline="") as handle:
282
+ window_rows = list(csv.DictReader(handle))
283
+ output_dir.mkdir(parents=True, exist_ok=True)
284
+
285
+ with (output_dir / "section_outline.csv").open("w", encoding="utf-8", newline="") as handle:
286
+ fieldnames = ["rank", "heading_level", "heading_text", "slug"]
287
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
288
+ writer.writeheader()
289
+ for rank, heading in enumerate(headings, start=1):
290
+ writer.writerow({
291
+ "rank": rank,
292
+ "heading_level": heading["level"],
293
+ "heading_text": heading["text"],
294
+ "slug": _heading_slug(heading["text"]),
295
+ })
296
+
297
+ all_terms: list[tuple[str, str]] = []
298
+ for term in spec.get("required_terms") or []:
299
+ all_terms.append((str(term), "required"))
300
+ for term in spec.get("code_terms") or []:
301
+ if str(term) not in {item[0] for item in all_terms}:
302
+ all_terms.append((str(term), "code"))
303
+ code_text = "\n".join(code_blocks)
304
+ term_rows: list[dict[str, Any]] = []
305
+ for term, category in all_terms:
306
+ term_rows.append({
307
+ "term": term,
308
+ "category": category,
309
+ "count": _term_count(clean_text, term),
310
+ "first_window": _first_window(window_rows, term),
311
+ "appears_in_code": "true" if term.lower() in code_text.lower() else "false",
312
+ })
313
+ term_rows.sort(key=lambda row: (str(row["category"]), str(row["term"]).lower()))
314
+ with (output_dir / "keyword_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
315
+ fieldnames = ["term", "category", "count", "first_window", "appears_in_code"]
316
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
317
+ writer.writeheader()
318
+ for row in term_rows:
319
+ writer.writerow(row)
320
+
321
+ code_terms = [str(item).lower() for item in spec.get("code_terms") or []]
322
+ with (output_dir / "code_block_inventory.csv").open("w", encoding="utf-8", newline="") as handle:
323
+ fieldnames = ["block_index", "line_count", "char_count", "contains_code_term", "first_line"]
324
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
325
+ writer.writeheader()
326
+ for index, block in enumerate(code_blocks, start=1):
327
+ lines = [line for line in block.splitlines() if line.strip()]
328
+ contains = any(term and term in block.lower() for term in code_terms)
329
+ writer.writerow({
330
+ "block_index": index,
331
+ "line_count": len(lines),
332
+ "char_count": len(block),
333
+ "contains_code_term": "true" if contains else "false",
334
+ "first_line": lines[0][:200] if lines else "",
335
+ })
336
+
337
+ keyword_total = sum(int(row["count"]) for row in term_rows)
338
+ required_found = sum(1 for row in term_rows if row["category"] == "required" and int(row["count"]) > 0)
339
+ top_keyword = None
340
+ if term_rows:
341
+ top = max(term_rows, key=lambda row: (int(row["count"]), str(row["term"]).lower()))
342
+ top_keyword = str(top["term"]) if int(top["count"]) > 0 else None
343
+ summary = {
344
+ "software_name": str(spec.get("software_name") or ""),
345
+ "source_file": source_file,
346
+ "window_count": len(window_rows),
347
+ "heading_count": len(headings),
348
+ "code_block_count": len(code_blocks),
349
+ "keyword_total": keyword_total,
350
+ "unique_required_terms_found": required_found,
351
+ "top_keyword": top_keyword,
352
+ "source_character_count": len(clean_text),
353
+ }
354
+ _write_json(output_dir / "summary.json", summary)
355
+
356
+
357
+ def geojson_outputs(input_dir: Path, output_dir: Path) -> None:
358
+ payload = _read_json(input_dir / "events.geojson")
359
+ features = payload.get("features") if isinstance(payload, dict) else []
360
+ rows: list[dict[str, Any]] = []
361
+ for feature in features:
362
+ if not isinstance(feature, dict):
363
+ continue
364
+ props = feature.get("properties") if isinstance(feature.get("properties"), dict) else {}
365
+ geom = feature.get("geometry") if isinstance(feature.get("geometry"), dict) else {}
366
+ coords = geom.get("coordinates") if isinstance(geom.get("coordinates"), list) else []
367
+ mag = props.get("mag")
368
+ try:
369
+ mag = float(mag) if mag is not None else None
370
+ except (TypeError, ValueError):
371
+ mag = None
372
+ depth = None
373
+ if len(coords) >= 3:
374
+ try:
375
+ depth = float(coords[2])
376
+ except (TypeError, ValueError):
377
+ depth = None
378
+ time_ms = props.get("time")
379
+ date_utc = "unknown"
380
+ if isinstance(time_ms, (int, float)):
381
+ date_utc = dt.datetime.utcfromtimestamp(time_ms / 1000).date().isoformat()
382
+ rows.append({
383
+ "id": feature.get("id") or props.get("code") or "",
384
+ "mag": mag,
385
+ "depth": depth,
386
+ "place": str(props.get("place") or ""),
387
+ "region": _stable_region(str(props.get("place") or "")),
388
+ "date_utc": date_utc,
389
+ "feature": feature,
390
+ })
391
+ output_dir.mkdir(parents=True, exist_ok=True)
392
+ mags = [row["mag"] for row in rows if row["mag"] is not None]
393
+ depths = [row["depth"] for row in rows if row["depth"] is not None]
394
+ summary = {
395
+ "event_count": len(rows),
396
+ "magnitude": {
397
+ "count": len(mags),
398
+ "min": round(min(mags), 6) if mags else None,
399
+ "max": round(max(mags), 6) if mags else None,
400
+ "mean": round(statistics.fmean(mags), 6) if mags else None,
401
+ "median": round(statistics.median(mags), 6) if mags else None,
402
+ },
403
+ "depth_km": {
404
+ "count": len(depths),
405
+ "min": round(min(depths), 6) if depths else None,
406
+ "max": round(max(depths), 6) if depths else None,
407
+ "mean": round(statistics.fmean(depths), 6) if depths else None,
408
+ },
409
+ "time_range_utc": {
410
+ "start_date": min((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
411
+ "end_date": max((row["date_utc"] for row in rows if row["date_utc"] != "unknown"), default=None),
412
+ },
413
+ "quality_flags": {
414
+ "missing_magnitude": sum(1 for row in rows if row["mag"] is None),
415
+ "missing_depth": sum(1 for row in rows if row["depth"] is None),
416
+ "unique_regions": len({row["region"] for row in rows}),
417
+ },
418
+ }
419
+ _write_json(output_dir / "summary.json", summary)
420
+
421
+ grouped: dict[tuple[str, str], list[dict[str, Any]]] = {}
422
+ for row in rows:
423
+ grouped.setdefault((_mag_bin(row["mag"]), _depth_bin(row["depth"])), []).append(row)
424
+ with (output_dir / "magnitude_depth_bins.csv").open("w", encoding="utf-8", newline="") as handle:
425
+ fieldnames = ["mag_bin", "depth_bin", "event_count", "mean_magnitude", "max_magnitude"]
426
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
427
+ writer.writeheader()
428
+ for (mag_bin, depth_bin), group in sorted(grouped.items()):
429
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
430
+ writer.writerow({
431
+ "mag_bin": mag_bin,
432
+ "depth_bin": depth_bin,
433
+ "event_count": len(group),
434
+ "mean_magnitude": f"{statistics.fmean(group_mags):.6f}" if group_mags else "",
435
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
436
+ })
437
+
438
+ daily: dict[tuple[str, str], list[dict[str, Any]]] = {}
439
+ for row in rows:
440
+ daily.setdefault((row["region"], row["date_utc"]), []).append(row)
441
+ with (output_dir / "region_daily_counts.csv").open("w", encoding="utf-8", newline="") as handle:
442
+ fieldnames = ["region", "date_utc", "event_count", "max_magnitude"]
443
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
444
+ writer.writeheader()
445
+ for (region, date_utc), group in sorted(daily.items()):
446
+ group_mags = [row["mag"] for row in group if row["mag"] is not None]
447
+ writer.writerow({
448
+ "region": region,
449
+ "date_utc": date_utc,
450
+ "event_count": len(group),
451
+ "max_magnitude": f"{max(group_mags):.6f}" if group_mags else "",
452
+ })
453
+
454
+ filtered = [row["feature"] for row in rows if row["mag"] is not None and row["mag"] >= 4.5]
455
+ _write_json(output_dir / "filtered_events.geojson", {"type": "FeatureCollection", "features": filtered})
456
+
457
+
458
+ def csv_time_series_outputs(input_dir: Path, output_dir: Path) -> None:
459
+ spec = _read_json(input_dir / "analysis_spec.json")
460
+ csv_name = spec["input_file"]
461
+ date_column = spec["date_column"]
462
+ value_column = spec["value_column"]
463
+ rows: list[dict[str, Any]] = []
464
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
465
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
466
+ reader = csv.DictReader(lines)
467
+ for item in reader:
468
+ raw_date = item.get(date_column, "")
469
+ try:
470
+ day = dt.date.fromisoformat(raw_date)
471
+ except ValueError:
472
+ if raw_date.isdigit() and len(raw_date) == 4:
473
+ day = dt.date(int(raw_date), 1, 1)
474
+ else:
475
+ continue
476
+ raw_value = item.get(value_column, "")
477
+ try:
478
+ value = float(raw_value)
479
+ except (TypeError, ValueError):
480
+ value = None
481
+ rows.append({"date": day, "value": value, "raw": item})
482
+ rows.sort(key=lambda item: item["date"])
483
+ valid = [row for row in rows if row["value"] is not None]
484
+ output_dir.mkdir(parents=True, exist_ok=True)
485
+ monthly: dict[str, list[float]] = {}
486
+ for row in valid:
487
+ key = row["date"].strftime("%Y-%m")
488
+ monthly.setdefault(key, []).append(row["value"])
489
+ with (output_dir / "monthly_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
490
+ fieldnames = ["month", "count", "mean", "median", "min", "max"]
491
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
492
+ writer.writeheader()
493
+ for month, values in sorted(monthly.items()):
494
+ writer.writerow({
495
+ "month": month,
496
+ "count": len(values),
497
+ "mean": f"{statistics.fmean(values):.6f}",
498
+ "median": f"{statistics.median(values):.6f}",
499
+ "min": f"{min(values):.6f}",
500
+ "max": f"{max(values):.6f}",
501
+ })
502
+ existing_dates = {row["date"] for row in rows}
503
+ missing_days: list[dt.date] = []
504
+ if rows:
505
+ day = rows[0]["date"]
506
+ while day <= rows[-1]["date"]:
507
+ if day.weekday() < 5 and day not in existing_dates:
508
+ missing_days.append(day)
509
+ day += dt.timedelta(days=1)
510
+ with (output_dir / "missing_business_days.csv").open("w", encoding="utf-8", newline="") as handle:
511
+ writer = csv.DictWriter(handle, fieldnames=["date"])
512
+ writer.writeheader()
513
+ for day in missing_days:
514
+ writer.writerow({"date": day.isoformat()})
515
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
516
+ writer = csv.DictWriter(handle, fieldnames=["date", "value"])
517
+ writer.writeheader()
518
+ for row in valid:
519
+ writer.writerow({"date": row["date"].isoformat(), "value": f"{row['value']:.6f}"})
520
+ values = [row["value"] for row in valid]
521
+ summary = {
522
+ "series_name": spec.get("series_name", value_column),
523
+ "row_count": len(rows),
524
+ "valid_value_count": len(valid),
525
+ "missing_value_count": len(rows) - len(valid),
526
+ "missing_business_day_count": len(missing_days),
527
+ "date_range": {
528
+ "start": rows[0]["date"].isoformat() if rows else None,
529
+ "end": rows[-1]["date"].isoformat() if rows else None,
530
+ },
531
+ "value": {
532
+ "min": round(min(values), 6) if values else None,
533
+ "max": round(max(values), 6) if values else None,
534
+ "mean": round(statistics.fmean(values), 6) if values else None,
535
+ "median": round(statistics.median(values), 6) if values else None,
536
+ },
537
+ }
538
+ _write_json(output_dir / "summary.json", summary)
539
+
540
+
541
+ def _aggregate_annual_rows(rows: list[dict[str, Any]], aggregation: str) -> list[dict[str, Any]]:
542
+ if not aggregation:
543
+ return rows
544
+ grouped: dict[int, list[float]] = {}
545
+ for row in rows:
546
+ grouped.setdefault(int(row["year"]), []).append(float(row["value"]))
547
+ aggregated: list[dict[str, Any]] = []
548
+ for year, values in sorted(grouped.items()):
549
+ if aggregation == "mean":
550
+ value = statistics.fmean(values)
551
+ elif aggregation == "median":
552
+ value = statistics.median(values)
553
+ elif aggregation == "max":
554
+ value = max(values)
555
+ elif aggregation == "min":
556
+ value = min(values)
557
+ elif aggregation == "count":
558
+ value = float(len(values))
559
+ else:
560
+ raise ValueError(f"unsupported year_aggregation for csv_annual_series_audit: {aggregation}")
561
+ aggregated.append({"year": year, "value": value})
562
+ return aggregated
563
+
564
+
565
+ def csv_annual_series_outputs(input_dir: Path, output_dir: Path) -> None:
566
+ spec = _read_json(input_dir / "analysis_spec.json")
567
+ csv_name = spec["input_file"]
568
+ year_column = spec["year_column"]
569
+ value_column = spec["value_column"]
570
+ # Optional: derive the year from an ISO-8601 timestamp column (e.g. USGS FDSN
571
+ # `time` = "2022-06-29T19:17:58.929Z") instead of parsing an integer year.
572
+ # When unset, behaviour is unchanged (integer year column).
573
+ year_from_timestamp = bool(spec.get("year_from_timestamp", False))
574
+ year_aggregation = str(spec.get("year_aggregation") or "").strip().lower()
575
+ raw_rows: list[dict[str, Any]] = []
576
+ with (input_dir / csv_name).open("r", encoding="utf-8", newline="") as raw_handle:
577
+ lines = [line for line in raw_handle if not line.lstrip().startswith("#")]
578
+ reader = csv.DictReader(lines)
579
+ for item in reader:
580
+ raw_year = str(item.get(year_column, "")).strip()
581
+ try:
582
+ if year_from_timestamp:
583
+ # Take the leading 4-digit year of an ISO timestamp/date.
584
+ match = re.match(r"\s*(\d{4})", raw_year)
585
+ if not match:
586
+ continue
587
+ year = int(match.group(1))
588
+ else:
589
+ year = int(raw_year)
590
+ value = float(str(item.get(value_column, "")).strip())
591
+ except (TypeError, ValueError):
592
+ continue
593
+ raw_rows.append({"year": year, "value": value})
594
+ rows = _aggregate_annual_rows(raw_rows, year_aggregation)
595
+ rows.sort(key=lambda item: item["year"])
596
+ output_dir.mkdir(parents=True, exist_ok=True)
597
+ decades: dict[str, list[float]] = {}
598
+ for row in rows:
599
+ decade_start = (row["year"] // 10) * 10
600
+ decades.setdefault(f"{decade_start}s", []).append(row["value"])
601
+ with (output_dir / "decadal_statistics.csv").open("w", encoding="utf-8", newline="") as handle:
602
+ fieldnames = [
603
+ "decade_start",
604
+ "start_year",
605
+ "end_year",
606
+ "count",
607
+ "mean_departure",
608
+ "median_departure",
609
+ "min_departure",
610
+ "max_departure",
611
+ "sample_std_departure",
612
+ ]
613
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
614
+ writer.writeheader()
615
+ for decade, values in sorted(decades.items()):
616
+ decade_start = int(decade.rstrip("s"))
617
+ years = [row["year"] for row in rows if decade_start <= row["year"] <= decade_start + 9]
618
+ writer.writerow({
619
+ "decade_start": decade_start,
620
+ "start_year": min(years),
621
+ "end_year": max(years),
622
+ "count": len(values),
623
+ "mean_departure": f"{statistics.fmean(values):.6f}",
624
+ "median_departure": f"{statistics.median(values):.6f}",
625
+ "min_departure": f"{min(values):.6f}",
626
+ "max_departure": f"{max(values):.6f}",
627
+ "sample_std_departure": f"{statistics.stdev(values):.6f}" if len(values) >= 2 else "",
628
+ })
629
+ threshold = float(spec.get("threshold", 1.0))
630
+ with (output_dir / "threshold_years.csv").open("w", encoding="utf-8", newline="") as handle:
631
+ fieldnames = ["rank", "year", "departure", "threshold"]
632
+ writer = csv.DictWriter(handle, fieldnames=fieldnames)
633
+ writer.writeheader()
634
+ rank = 1
635
+ for row in rows:
636
+ if row["value"] >= threshold:
637
+ writer.writerow({
638
+ "rank": rank,
639
+ "year": row["year"],
640
+ "departure": f"{row['value']:.6f}",
641
+ "threshold": f"{threshold:.6f}",
642
+ })
643
+ rank += 1
644
+ with (output_dir / "clean_series.csv").open("w", encoding="utf-8", newline="") as handle:
645
+ writer = csv.DictWriter(handle, fieldnames=["Year", "Departure from Average"])
646
+ writer.writeheader()
647
+ for row in rows:
648
+ writer.writerow({"Year": row["year"], "Departure from Average": f"{row['value']:.6f}"})
649
+ values = [row["value"] for row in rows]
650
+ threshold_rows = [row for row in rows if row["value"] >= threshold]
651
+ min_row = min(rows, key=lambda row: row["value"]) if rows else None
652
+ max_row = max(rows, key=lambda row: row["value"]) if rows else None
653
+ summary = {
654
+ "series_name": spec.get("series_name", value_column),
655
+ "row_count": len(rows),
656
+ "year_range": {
657
+ "start": rows[0]["year"] if rows else None,
658
+ "end": rows[-1]["year"] if rows else None,
659
+ },
660
+ "value": {
661
+ "min": round(min(values), 6) if values else None,
662
+ "min_year": min_row["year"] if min_row else None,
663
+ "max": round(max(values), 6) if values else None,
664
+ "max_year": max_row["year"] if max_row else None,
665
+ "mean": round(statistics.fmean(values), 6) if values else None,
666
+ "median": round(statistics.median(values), 6) if values else None,
667
+ "sample_std": round(statistics.stdev(values), 6) if len(values) >= 2 else None,
668
+ },
669
+ "threshold": threshold,
670
+ "years_at_or_above_threshold": len(threshold_rows),
671
+ "first_threshold_year": threshold_rows[0]["year"] if threshold_rows else None,
672
+ "last_threshold_year": threshold_rows[-1]["year"] if threshold_rows else None,
673
+ }
674
+ if year_aggregation:
675
+ summary["year_aggregation"] = year_aggregation
676
+ summary["source_row_count"] = len(raw_rows)
677
+ _write_json(output_dir / "summary.json", summary)
678
+
679
+
20260904-005149-science-environment-cee684/tests/reference/internal/workflow_runtime.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ from pathlib import Path
2
+ from workflow_adapters_runtime import csv_annual_series_outputs
3
+
4
+ def run_reference(input_dir: Path, output_dir: Path) -> None:
5
+ csv_annual_series_outputs(input_dir, output_dir)
20260904-005149-science-environment-cee684/tests/reference/reference_manifest.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "workflow_kind": "csv_annual_series_audit",
3
+ "task_id": "usgs_fdsn_eq_andes_1970_1970_m4p5_depth_mean_quota_repair_0778",
4
+ "domain": "science_environment",
5
+ "task": "usgs_andes_annual_focal_depth_audit",
6
+ "variant": "base",
7
+ "created_from": "standard reference workflow",
8
+ "required_outputs": [
9
+ {
10
+ "path": "summary.json",
11
+ "type": "json"
12
+ },
13
+ {
14
+ "path": "decadal_statistics.csv",
15
+ "type": "csv"
16
+ },
17
+ {
18
+ "path": "threshold_years.csv",
19
+ "type": "csv"
20
+ },
21
+ {
22
+ "path": "clean_series.csv",
23
+ "type": "csv"
24
+ }
25
+ ],
26
+ "numeric_tolerance": 1e-09,
27
+ "source_prior_sha256": "17ec83991f0cc3a43a3e2535b9b8f82aabe3465c42dcd550539eabd56a469ee7",
28
+ "gold_output_hashes": {
29
+ "summary.json": "5f1973ada56af746a9ed6567ecb798669f733efdaea5b52cc10c7d3986b7ffca",
30
+ "decadal_statistics.csv": "127c891478bed0a7b468d46f1d505fdbbec854890d4be20fa42c782cad17a451",
31
+ "threshold_years.csv": "b1c8cc3331d945668228fe1975cfda0d8dffab79c15b43681deb0698e27af28f",
32
+ "clean_series.csv": "9b4556992bf970940de195a19812b3b4d415783f8040d5e1a11dcc19a1709645"
33
+ }
34
+ }
20260904-005149-science-environment-cee684/tests/reference/reference_solution.py ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ import argparse
4
+ import sys
5
+ from pathlib import Path
6
+
7
+
8
+ def main() -> int:
9
+ parser = argparse.ArgumentParser()
10
+ parser.add_argument("--input", required=True)
11
+ parser.add_argument("--output", required=True)
12
+ args = parser.parse_args()
13
+ sys.path.insert(0, str(Path(__file__).resolve().parent / "internal"))
14
+ from workflow_runtime import run_reference
15
+ run_reference(Path(args.input), Path(args.output))
16
+ return 0
17
+
18
+
19
+ if __name__ == "__main__":
20
+ raise SystemExit(main())
20260904-005149-science-environment-cee684/tests/test.sh ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ set -u
3
+
4
+ TESTS_DIR="${HARBOR_TESTS_DIR:-/tests}"
5
+ WORKSPACE_DIR="${HARBOR_WORKSPACE_DIR:-/workspace}"
6
+ LOGS_DIR="${HARBOR_LOGS_DIR:-/logs/verifier}"
7
+
8
+ mkdir -p "${LOGS_DIR}" "${WORKSPACE_DIR}/output"
9
+
10
+ python "${TESTS_DIR}/evaluator_runner.py" \
11
+ --output-dir "${WORKSPACE_DIR}/output" \
12
+ --reference-dir "${TESTS_DIR}/reference" \
13
+ --reward-json "${LOGS_DIR}/reward.json" \
14
+ --details-json "${LOGS_DIR}/ale_evaluator_details.json"
15
+
16
+ if [ ! -f "${LOGS_DIR}/reward.json" ]; then
17
+ echo '{"score":0.0,"format":0.0,"output_present":0.0,"evaluator_available":0.0,"item_count":0.0}' > "${LOGS_DIR}/reward.json"
18
+ fi
19
+
20
+ python - <<'PY' "${LOGS_DIR}/reward.json" "${LOGS_DIR}/reward.txt"
21
+ import json
22
+ import sys
23
+ from pathlib import Path
24
+ reward = json.loads(Path(sys.argv[1]).read_text(encoding="utf-8"))
25
+ score = float(reward.get("score", 0.0))
26
+ Path(sys.argv[2]).write_text(f"{score}\n", encoding="utf-8")
27
+ PY
28
+
29
+ exit 0
20260904-005151-science-environment-7e595f/.DS_Store ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f460901604ffbe0a2dd82f6ba21089032e5ee22d96774e3556e4b56cd5ea0693
3
+ size 6148
20260904-005151-science-environment-7e595f/environment/Dockerfile ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ FROM python:3.12-slim
2
+
3
+ WORKDIR /workspace
4
+
5
+ COPY public_data/input /workspace/input
6
+ COPY public_data/software /workspace/software
7
+ COPY public_data/task_card.json /workspace/task_card.json
8
+ COPY public_data/main.py /workspace/main.py
9
+ COPY public_data/README.md /workspace/README.md
10
+
11
+ RUN mkdir -p /workspace/output /workspace/report /logs/agent /logs/artifacts \
12
+ && chmod -R 777 /workspace/output /workspace/report /logs
20260904-005151-science-environment-7e595f/environment/public_data/README.md ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Public Task Inputs
2
+
3
+ Files in this directory are copied into the workspace:
4
+
5
+ - `input/` -> `/workspace/input/`
6
+ - `software/` -> `/workspace/software/`
7
+ - `task_card.json` -> `/workspace/task_card.json`
8
+ - `main.py` -> `/workspace/main.py`
9
+
10
+ Hidden reference data and evaluator files are not present in the agent workspace.
11
+ Final deliverables must be written under `/workspace/output/`.
20260904-005151-science-environment-7e595f/environment/public_data/input/README.md ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ This directory contains the public staged input for this task.
2
+ input/analysis_spec.json is the authoritative public contract for input columns, grouping/filtering rules, required output files, and output schemas.
3
+ Primary data file: input/usgs_fdsn_eq_caribbean_1987_1989_m5p5_depth_mean_quota_repair_0779.csv.
4
+ Workflow kind: csv_annual_series_audit.
5
+ Required output files: output/summary.json, output/decadal_statistics.csv, output/threshold_years.csv, output/clean_series.csv.
6
+ Solve the task from the staged local files only; do not fetch fresh remote data during solve time.
20260904-005151-science-environment-7e595f/environment/public_data/input/analysis_spec.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input_file": "usgs_fdsn_eq_caribbean_1987_1989_m5p5_depth_mean_quota_repair_0779.csv",
3
+ "workflow_kind": "csv_annual_series_audit",
4
+ "year_column": "time",
5
+ "value_column": "depth",
6
+ "series_name": "USGS FDSN Caribbean plate boundary 1978-1997 earthquake annual mean focal depth (km) (min mag 4.5)",
7
+ "threshold": 70.0,
8
+ "year_from_timestamp": true,
9
+ "comment_prefix": "#",
10
+ "required_outputs": [
11
+ {
12
+ "path": "summary.json",
13
+ "type": "json"
14
+ },
15
+ {
16
+ "path": "decadal_statistics.csv",
17
+ "type": "csv"
18
+ },
19
+ {
20
+ "path": "threshold_years.csv",
21
+ "type": "csv"
22
+ },
23
+ {
24
+ "path": "clean_series.csv",
25
+ "type": "csv"
26
+ }
27
+ ],
28
+ "output_schemas": {
29
+ "summary.json": {
30
+ "series_name": "string",
31
+ "row_count": "integer valid annual rows",
32
+ "year_range": {
33
+ "start": "integer|null",
34
+ "end": "integer|null"
35
+ },
36
+ "value": {
37
+ "min": "number|null",
38
+ "min_year": "integer|null",
39
+ "max": "number|null",
40
+ "max_year": "integer|null",
41
+ "mean": "number|null",
42
+ "median": "number|null",
43
+ "sample_std": "number|null"
44
+ },
45
+ "threshold": "number",
46
+ "years_at_or_above_threshold": "integer",
47
+ "first_threshold_year": "integer|null",
48
+ "last_threshold_year": "integer|null",
49
+ "year_aggregation": "string annual aggregation applied to raw rows",
50
+ "source_row_count": "integer raw rows before annual aggregation"
51
+ },
52
+ "decadal_statistics.csv": [
53
+ "decade_start",
54
+ "start_year",
55
+ "end_year",
56
+ "count",
57
+ "mean_departure",
58
+ "median_departure",
59
+ "min_departure",
60
+ "max_departure",
61
+ "sample_std_departure"
62
+ ],
63
+ "threshold_years.csv": [
64
+ "rank",
65
+ "year",
66
+ "departure",
67
+ "threshold"
68
+ ],
69
+ "clean_series.csv": [
70
+ "Year",
71
+ "Departure from Average"
72
+ ]
73
+ },
74
+ "notes": "First group raw CSV rows by year and aggregate the value column with mean; then compute reproducible annual-series audit outputs. Do not fetch new data.",
75
+ "year_aggregation": "mean"
76
+ }
20260904-005151-science-environment-7e595f/environment/public_data/input/source_context_windows.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
20260904-005151-science-environment-7e595f/environment/public_data/input/usgs_fdsn_eq_caribbean_1987_1989_m5p5_depth_mean_quota_repair_0779.csv ADDED
The diff for this file is too large to render. See raw diff