File size: 11,301 Bytes
d82db76
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
from __future__ import annotations

import importlib.util
import json
import sys
from argparse import Namespace
from pathlib import Path


def _load_tool():
    path = Path(__file__).resolve().parents[1] / "tools" / "audit_recognizer_eval_pack.py"
    spec = importlib.util.spec_from_file_location("audit_recognizer_eval_pack", path)
    assert spec is not None
    assert spec.loader is not None
    module = importlib.util.module_from_spec(spec)
    sys.modules[spec.name] = module
    spec.loader.exec_module(module)
    return module


def _args(tmp_path: Path) -> Namespace:
    return Namespace(
        label_file=tmp_path / "pack" / "eval.txt",
        data_dir=tmp_path / "pack",
        manifest_file=tmp_path / "pack" / "manifest.jsonl",
        summary_file=tmp_path / "pack" / "summary.json",
        out=tmp_path / "audit",
        language="newa",
        script="Newa",
        domain="real_handwritten",
        gt_status="human_verified",
        split="eval",
        eval_pack_version="fixture-v1",
        frozen_eval_pack=True,
        unseen_for_selection=True,
        claim_eligible=True,
        require_claim_ready=False,
    )


def _write_ready_pack(tmp_path: Path, *, rows: int = 30) -> None:
    pack = tmp_path / "pack"
    images = pack / "images"
    images.mkdir(parents=True)
    label = "𑐣𑐾𑐥𑐵𑐮"
    tool = _load_tool()
    label_lines: list[str] = []
    manifest_lines: list[str] = []
    content_rows: list[dict[str, object]] = []
    for index in range(1, rows + 1):
        image = images / f"line-{index}.png"
        image.write_bytes(f"line image {index}".encode("utf-8"))
        rel_path = f"images/line-{index}.png"
        image_sha256 = tool._hash_file(image)
        text_sha256 = tool._hash_text(label)
        label_lines.append(f"{rel_path}\t{label}")
        manifest_lines.append(
            json.dumps(
                {
                    "sample_id": f"sample-{index}",
                    "image": rel_path,
                    "text": label,
                    "sha256": image_sha256,
                    "text_sha256": text_sha256,
                    "sample_sha256": tool._sample_hash(image_sha256, text_sha256),
                    "source_image": f"source-page-{index}.jpg",
                    "page": index,
                    "bbox": [10, 20, 100, 40],
                    "metadata": {
                        "language": "newa",
                        "script": "Newa",
                        "domain": "real_handwritten",
                        "gt_status": "human_verified",
                        "split": "eval",
                        "claim_evidence_eligible": True,
                        "reviewed_by": "fixture-reviewer",
                    },
                },
                ensure_ascii=False,
            )
        )
        content_rows.append(
            {
                "sample_id": f"sample-{index}",
                "image_sha256": image_sha256,
                "gt_text_sha256_nfc": text_sha256,
                "language": "newa",
                "script": "Newa",
                "domain": "real_handwritten",
                "gt_status": "human_verified",
                "split": "eval",
                "claim_eligible": True,
            }
        )
    (pack / "eval.txt").write_text("\n".join(label_lines) + "\n", encoding="utf-8")
    (pack / "manifest.jsonl").write_text("\n".join(manifest_lines) + "\n", encoding="utf-8")
    (pack / "summary.json").write_text(
        json.dumps(
            {
                "dataset": "fixture",
                "eval_pack_version": "fixture-v1",
                "pack_content_sha256": tool._pack_content_digest(content_rows),
                "eval_pack": {"line_crops": rows},
            }
        )
        + "\n",
        encoding="utf-8",
    )


def test_audit_recognizer_eval_pack_accepts_claim_ready_pack(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path)

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["claim_ready"] is True
    assert audit["warnings"] == []
    assert audit["counts"]["manifest_matched"] == 30
    assert audit["counts"]["claim_eligible"] == 30


def test_audit_recognizer_eval_pack_warns_on_incomplete_provenance(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path, rows=1)
    manifest = tmp_path / "pack" / "manifest.jsonl"
    payload = json.loads(manifest.read_text(encoding="utf-8"))
    payload.pop("page")
    payload.pop("bbox")
    payload["metadata"]["claim_evidence_eligible"] = False
    manifest.write_text(json.dumps(payload, ensure_ascii=False) + "\n", encoding="utf-8")
    summary = tmp_path / "pack" / "summary.json"
    summary_payload = json.loads(summary.read_text(encoding="utf-8"))
    summary_payload["pack_content_sha256"] = tool._pack_content_digest(
        [
            {
                "sample_id": payload["sample_id"],
                "image_sha256": payload["sha256"],
                "gt_text_sha256_nfc": payload["text_sha256"],
                "language": "newa",
                "script": "Newa",
                "domain": "real_handwritten",
                "gt_status": "human_verified",
                "split": "eval",
                "claim_eligible": False,
            }
        ]
    )
    summary.write_text(json.dumps(summary_payload) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["claim_ready"] is False
    assert "missing_page: 1/1" in audit["warnings"]
    assert "missing_bbox: 1/1" in audit["warnings"]
    assert "not all rows are claim eligible: 0/1" in audit["warnings"]


def test_audit_recognizer_eval_pack_accepts_pdf_native_provenance_keys(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path, rows=30)
    pack = tmp_path / "pack"
    manifest_rows = []
    for line in (pack / "manifest.jsonl").read_text(encoding="utf-8").splitlines():
        payload = json.loads(line)
        payload.pop("source_image")
        payload.pop("page")
        payload.pop("bbox")
        payload["pdf_url"] = "https://example.test/source.pdf"
        payload["pdf_page_number"] = 6
        payload["bbox_rendered_pixels"] = [10, 20, 100, 40]
        manifest_rows.append(json.dumps(payload, ensure_ascii=False))
    (pack / "manifest.jsonl").write_text("\n".join(manifest_rows) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["counts"]["missing_source_document"] == 0
    assert audit["counts"]["missing_page"] == 0
    assert audit["counts"]["missing_bbox"] == 0
    assert not any(warning.startswith("missing_source_document") for warning in audit["warnings"])
    assert not any(warning.startswith("missing_page") for warning in audit["warnings"])
    assert not any(warning.startswith("missing_bbox") for warning in audit["warnings"])


def test_audit_recognizer_eval_pack_warns_on_human_verified_without_review_evidence(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path, rows=1)
    manifest = tmp_path / "pack" / "manifest.jsonl"
    payload = json.loads(manifest.read_text(encoding="utf-8"))
    payload["metadata"].pop("reviewed_by")
    manifest.write_text(json.dumps(payload, ensure_ascii=False) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["claim_ready"] is False
    assert "missing_human_review_evidence: 1/1" in audit["warnings"]


def test_audit_recognizer_eval_pack_surfaces_normalizer_defaults(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path, rows=1)
    summary = tmp_path / "pack" / "summary.json"
    payload = json.loads(summary.read_text(encoding="utf-8"))
    payload["counts"] = {
        "defaulted_domain": 1,
        "defaulted_gt_status": 1,
        "defaulted_split": 1,
    }
    summary.write_text(json.dumps(payload) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["claim_ready"] is False
    assert audit["normalizer_counts"]["defaulted_domain"] == 1
    assert "normalizer_defaulted_domain: 1/1" in audit["warnings"]
    assert "normalizer_defaulted_gt_status: 1/1" in audit["warnings"]
    assert "normalizer_defaulted_split: 1/1" in audit["warnings"]


def test_audit_recognizer_eval_pack_warns_on_tiny_claim_pack(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path, rows=1)

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is True
    assert audit["claim_ready"] is False
    assert "claim-facing sample size too small: 1 scoring units, minimum 30" in audit["warnings"]
    assert (
        "slice sample size too small: newa/Newa/real_handwritten/human_verified/eval has 1 scoring units, minimum 30"
        in audit["warnings"]
    )


def test_audit_recognizer_eval_pack_fails_on_missing_image(tmp_path: Path) -> None:
    tool = _load_tool()
    pack = tmp_path / "pack"
    pack.mkdir()
    (pack / "eval.txt").write_text("images/missing.png\ttext\n", encoding="utf-8")
    (pack / "manifest.jsonl").write_text('{"image":"images/missing.png","text":"text"}\n', encoding="utf-8")
    (pack / "summary.json").write_text('{"dataset":"fixture"}\n', encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is False
    assert audit["claim_ready"] is False
    assert any("missing image" in issue for issue in audit["issues"])


def test_audit_recognizer_eval_pack_fails_on_duplicate_manifest_image_basename(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path)
    manifest = tmp_path / "pack" / "manifest.jsonl"
    rows = [json.loads(line) for line in manifest.read_text(encoding="utf-8").splitlines()]
    duplicate = dict(rows[0])
    duplicate["sample_id"] = "sample-2"
    duplicate["image"] = "other/line-1.png"
    rows.append(duplicate)
    manifest.write_text("\n".join(json.dumps(row, ensure_ascii=False) for row in rows) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is False
    assert audit["claim_ready"] is False
    assert "duplicate manifest image basename: line-1.png" in audit["issues"]


def test_audit_recognizer_eval_pack_fails_on_stale_text_and_sample_hashes(tmp_path: Path) -> None:
    tool = _load_tool()
    _write_ready_pack(tmp_path)
    manifest = tmp_path / "pack" / "manifest.jsonl"
    rows = [json.loads(line) for line in manifest.read_text(encoding="utf-8").splitlines()]
    rows[0]["text_sha256"] = "0" * 64
    rows[0]["sample_sha256"] = "1" * 64
    manifest.write_text("\n".join(json.dumps(row, ensure_ascii=False) for row in rows) + "\n", encoding="utf-8")

    audit = tool.audit_pack(_args(tmp_path))

    assert audit["passed"] is False
    assert audit["claim_ready"] is False
    assert audit["counts"]["text_sha256_mismatch"] == 1
    assert audit["counts"]["sample_sha256_mismatch"] == 1
    assert any("text_sha256 mismatch for images/line-1.png" in issue for issue in audit["issues"])
    assert any("sample_sha256 mismatch for images/line-1.png" in issue for issue in audit["issues"])