File size: 11,614 Bytes
25872e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
#!/usr/bin/env python3
"""Validate the CropMath Hugging Face release package."""

from __future__ import annotations

import argparse
import json
from pathlib import Path
import re


EXPECTED = {
    "default": {"dev": 252, "test": 751, "gold": 95},
    "eval_prompts": {"dev": 1512, "test": 4506, "gold": 570},
}
EXPECTED_HIDDEN_PUBLIC = 160
CONDITIONS = {"C", "K_name", "K_formula", "K_domain", "K_distractor", "K_wrong"}
PUBLIC_CONTENT_SUFFIXES = {".json", ".jsonl", ".csv", ".yaml", ".yml"}
FORBIDDEN_PUBLIC_CONTENT = {
    "specific framework name": re.compile(
        r"RiceGrow|CH4MOD|AgroC|\bSIMPLE\b|\bDSSAT\b|\bAPSIM\b|"
        r"\bWOFOST\b|\bORYZA\b|\bDNDC\b|\bCENTURY\b|\bDayCent\b|"
        r"\bCERES\b|\bAquaCrop\b|\bSTICS\b|\bEPIC\b"
    ),
    "source-code wording": re.compile(
        r"source[ -]?code|codebase|源代码|源码|代码溯源|代码实现|代码位置",
        re.IGNORECASE,
    ),
    "implementation file reference": re.compile(
        r"(?:^|[\s'\"])(?:src|scripts)/[^\s'\"]+|\b[^\s/]+\.(?:py|R)\b",
        re.IGNORECASE,
    ),
    "generated citation": re.compile(
        r"\bet\s+al\.?\b|\bdoi\s*:|https?://|\bcited in\b|\bcitation\b",
        re.IGNORECASE,
    ),
}

# Code-area scan: catches model-name and provenance leaks in the public repo's
# Python/Markdown sources. Unlike FORBIDDEN_PUBLIC_CONTENT this does NOT reject
# legitimate ``.py`` references (code files legitimately live in the repo) and
# does NOT reject generic URLs (eval scripts reference model hubs); it focuses
# on the specific residue that would reveal the private formula provenance.
REPO_ROOT_SCAN_DIRS = ("src", "evaluation", "scripts", "tests")
REPO_ROOT_SCAN_FILES = ("README.md", "AGENTS.md")
REPO_ROOT_CODE_SUFFIXES = {".py", ".md", ".yaml", ".yml", ".sh"}
REPO_ROOT_FORBIDDEN = {
    "specific framework name": re.compile(
        r"RiceGrow|CH4MOD|AgroC|\bSIMPLE crop\b|\bSIMPLE model\b|"
        r"\bDSSAT\b|\bAPSIM\b|\bWOFOST\b|\bORYZA\b|\bDNDC\b|"
        r"\bCENTURY\b|\bDayCent\b|\bCERES\b|\bAquaCrop\b|\bSTICS\b|\bEPIC\b"
    ),
    "provenance citation in code": re.compile(
        r"\b(?:Tang|Huang|Herbst|Zhao)\s+(?:19|20)\d{2}\b|"
        r"\bfrom\s+(?:CH4MOD|RiceGrow|AgroC)\s+code\b|"
        r"\bCommon\.R\b|\bfunc_[A-Za-z][A-Za-z0-9_]+",
        re.IGNORECASE,
    ),
    "implementation file reference": re.compile(
        r"\bcropmath\.formulas\b|\bcropmath\.generator\b|"
        r"\b[^\s/]+\.R\b",
        re.IGNORECASE,
    ),
}
# Files that legitimately encode the forbidden patterns as a *defense* (the
# validator itself and the clean-release builder, both of which carry the
# blocklist). Scanned but whitelisted by relative path so the defense keeps
# working without false positives.
REPO_ROOT_SCAN_ALLOWLIST = {
    "scripts/validate_cropmath_release.py",
    "scripts/build_clean_public_release.py",
}


def load_jsonl(path: Path) -> list[dict]:
    with path.open(encoding="utf-8") as f:
        return [json.loads(line) for line in f if line.strip()]


def require(condition: bool, message: str, issues: list[str]) -> None:
    if not condition:
        issues.append(message)


def validate_public_content(release_dir: Path) -> list[str]:
    """Reject private implementation and citation residue in public data files."""
    issues: list[str] = []
    for path in sorted(release_dir.rglob("*")):
        if not path.is_file() or path.suffix.lower() not in PUBLIC_CONTENT_SUFFIXES:
            continue
        text = path.read_text(encoding="utf-8", errors="replace")
        for label, pattern in FORBIDDEN_PUBLIC_CONTENT.items():
            match = pattern.search(text)
            if match:
                relative = path.relative_to(release_dir)
                issues.append(f"{relative}: {label}: {match.group(0)!r}")
    return issues


def validate_repo_root_content(repo_root: Path) -> list[str]:
    """Reject private provenance residue in the public repo's own source files.

    Scans ``src/``, ``evaluation/``, ``scripts/``, ``tests/`` plus top-level
    markdown for crop-model names, paper-year citations, and implementation
    references that would reveal the private formula provenance. The validator
    itself is allowlisted so its defensive blocklist keeps working.
    """
    issues: list[str] = []
    candidates: list[Path] = []
    for name in REPO_ROOT_SCAN_DIRS:
        directory = repo_root / name
        if directory.is_dir():
            for path in directory.rglob("*"):
                if path.is_file() and path.suffix.lower() in REPO_ROOT_CODE_SUFFIXES:
                    candidates.append(path)
    for name in REPO_ROOT_SCAN_FILES:
        path = repo_root / name
        if path.is_file():
            candidates.append(path)
    for path in sorted(candidates):
        try:
            relative = path.relative_to(repo_root).as_posix()
        except ValueError:
            relative = str(path)
        if relative in REPO_ROOT_SCAN_ALLOWLIST:
            continue
        text = path.read_text(encoding="utf-8", errors="replace")
        for label, pattern in REPO_ROOT_FORBIDDEN.items():
            match = pattern.search(text)
            if match:
                issues.append(f"{relative}: {label}: {match.group(0)!r}")
    return issues


def validate_jsonl(release_dir: Path) -> list[str]:
    issues: list[str] = []
    rows: dict[tuple[str, str], list[dict]] = {}
    for config, split_counts in EXPECTED.items():
        for split, expected_n in split_counts.items():
            path = release_dir / "data" / config / f"{split}.jsonl"
            require(path.exists(), f"missing file: {path}", issues)
            if not path.exists():
                continue
            data = load_jsonl(path)
            rows[(config, split)] = data
            ids = [row["id"] for row in data]
            require(len(data) == expected_n, f"{config}/{split}: {len(data)} rows, expected {expected_n}", issues)
            require(len(ids) == len(set(ids)), f"{config}/{split}: duplicate IDs", issues)
            require(all(row.get("answer") not in (None, "") for row in data), f"{config}/{split}: missing answer", issues)
            if config == "eval_prompts":
                require(all(row.get("prompt") for row in data), f"{config}/{split}: missing prompt", issues)
                require({row.get("condition") for row in data} <= CONDITIONS, f"{config}/{split}: invalid condition", issues)

    dev_ids = {row["id"] for row in rows.get(("default", "dev"), [])}
    test_ids = {row["id"] for row in rows.get(("default", "test"), [])}
    gold_ids = {row["id"] for row in rows.get(("default", "gold"), [])}
    require(not (dev_ids & test_ids), "default/dev and default/test overlap", issues)
    require(gold_ids <= test_ids, "default/gold is not a subset of default/test", issues)

    for split in ["dev", "test", "gold"]:
        default_ids = {row["id"] for row in rows.get(("default", split), [])}
        eval_rows = rows.get(("eval_prompts", split), [])
        eval_sample_ids = [row["sample_id"] for row in eval_rows]
        require(set(eval_sample_ids) == default_ids, f"eval_prompts/{split}: sample IDs do not match default/{split}", issues)
        for sample_id in default_ids:
            sample_conditions = {row["condition"] for row in eval_rows if row["sample_id"] == sample_id}
            require(sample_conditions == CONDITIONS, f"eval_prompts/{split}: {sample_id} does not have all six conditions", issues)

    hidden_path = release_dir / "hidden_public" / "hidden_public.jsonl"
    require(hidden_path.exists(), "missing hidden public file", issues)
    if hidden_path.exists():
        hidden_rows = load_jsonl(hidden_path)
        hidden_ids = {row["id"] for row in hidden_rows}
        require(len(hidden_rows) == EXPECTED_HIDDEN_PUBLIC, f"hidden_public: {len(hidden_rows)} rows, expected {EXPECTED_HIDDEN_PUBLIC}", issues)
        require(len(hidden_rows) == len(hidden_ids), "hidden_public: duplicate IDs", issues)
        require(not (hidden_ids & (dev_ids | test_ids | gold_ids)), "hidden_public overlaps dev/test/gold IDs", issues)
        forbidden = {"answer", "solution", "effective_parameters", "answer_float"}
        leaked = [row["id"] for row in hidden_rows if forbidden & set(row)]
        require(not leaked, f"hidden_public leaks answer-bearing fields: {leaked[:5]}", issues)
        require(all(row.get("answer_hidden") is True for row in hidden_rows), "hidden_public rows must set answer_hidden=true", issues)
        bad_sensitivity = [row["id"] for row in hidden_rows if row.get("mode") == "sensitivity" and not row.get("change")]
        require(not bad_sensitivity, f"hidden_public sensitivity rows missing change: {bad_sensitivity[:5]}", issues)

    private_markers = list(release_dir.rglob("*private*")) + list(release_dir.rglob("*answer*.jsonl"))
    leaked_private = [str(path.relative_to(release_dir)) for path in private_markers if "hidden_public" not in str(path)]
    require(not leaked_private, f"private answer-like files found in public release: {leaked_private}", issues)

    return issues


def validate_hf_loader(release_dir: Path) -> list[str]:
    try:
        from datasets import load_dataset
    except ImportError:
        return ["datasets is not installed; HF loader validation requires this dependency"]

    issues: list[str] = []
    for config, split_counts in EXPECTED.items():
        dataset = load_dataset(str(release_dir), config)
        observed = {split: len(dataset[split]) for split in split_counts}
        if observed != split_counts:
            issues.append(f"HF loader {config}: {observed}, expected {split_counts}")
    hidden_dataset = load_dataset(str(release_dir), "hidden_public")
    hidden_observed = len(hidden_dataset["hidden_public"])
    if hidden_observed != EXPECTED_HIDDEN_PUBLIC:
        issues.append(f"HF loader hidden_public: {hidden_observed}, expected {EXPECTED_HIDDEN_PUBLIC}")
    return issues


def main() -> None:
    parser = argparse.ArgumentParser(description="Validate CropMath release/cropmath-v1")
    parser.add_argument("--release-dir", default="release/cropmath-v1")
    parser.add_argument("--skip-hf-loader", action="store_true")
    parser.add_argument(
        "--no-scan-repo-root",
        dest="scan_repo_root",
        action="store_false",
        default=True,
        help="Skip scanning the public repo source tree for provenance residue.",
    )
    args = parser.parse_args()

    release_dir = Path(args.release_dir).resolve()
    repo_root = release_dir.parents[1] if release_dir.name == "cropmath-v1" else Path.cwd().resolve()
    try:
        release_label = str(release_dir.relative_to(Path.cwd().resolve()))
    except ValueError:
        release_label = str(Path(args.release_dir))
    issues = validate_jsonl(release_dir)
    issues.extend(validate_public_content(release_dir))
    repo_root_issues: list[str] = []
    if args.scan_repo_root:
        repo_root_issues = validate_repo_root_content(repo_root)
    hf_issues = [] if args.skip_hf_loader else validate_hf_loader(release_dir)
    all_hard_issues = issues + repo_root_issues + hf_issues

    report = {
        "release_dir": release_label,
        "jsonl_pass": not issues,
        "hf_loader_pass": None if args.skip_hf_loader else not hf_issues,
        "repo_root_scan_pass": None if not args.scan_repo_root else not repo_root_issues,
        "issues": issues + repo_root_issues,
        "hf_loader_issues": hf_issues,
    }
    print(json.dumps(report, ensure_ascii=False, indent=2))
    if all_hard_issues:
        raise SystemExit(1)


if __name__ == "__main__":
    main()