Datasets:
File size: 10,117 Bytes
0bf2389 | 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 | from __future__ import annotations
from copy import deepcopy
import json
from pathlib import Path
import unittest
from barbet_bgd.validation import (
ACTION_TOOLS,
PRIMITIVES,
RecordValidationError,
assert_valid_record,
find_disallowed_terms,
validate_record,
)
def minimal_valid_record() -> dict:
return {
"record_id": "bgd_verify_000001",
"schema_version": "1.0.0",
"primitive": "VERIFY",
"mixture_category": "VERIFICATION_QE_ERROR_LOCALIZATION",
"task_family": "ASR_QUALITY",
"domain": "meeting",
"language": ["zh-TW", "en"],
"locale": "zh-TW",
"split": "train",
"source_group": "meeting_session_001",
"task_spec": {
"instruction": "比較參考文字與辨識結果，標記有證據支持的錯誤。",
"allowed_labels": ["PASS", "FAIL", "UNCERTAIN"],
"allowed_error_types": ["VALUE_CORRUPTION"],
"allowed_actions": ["EDIT", "ABSTAIN"],
"required_evidence": True,
"abstention_allowed": True,
},
"inputs": [
{
"id": "src",
"type": "reference_text",
"role": "reference",
"units": [
{"id": "src:w01", "text": "模型將於九月十五日上線。"},
],
},
{
"id": "hyp",
"type": "asr_hypothesis",
"units": [
{"id": "hyp:w01", "text": "模型將於九月五日上線。"},
],
},
],
"target": {
"decision": "FAIL",
"errors": [
{
"type": "VALUE_CORRUPTION",
"severity": "major",
"source_evidence": ["src:w01"],
"target_evidence": ["hyp:w01"],
}
],
"post_edit": "模型將於九月十五日上線。",
"action": {
"tool": "EDIT",
"arguments": {
"target": "hyp:w01",
"replacement": "模型將於九月十五日上線。",
},
},
"confidence_target": 0.98,
},
"component_targets": {
"source_token_span": [0, 1],
"target_token_span": [0, 1],
"audio_interval": None,
"image_region": None,
},
"presentation": {
"serializer": "json",
"serializer_version": "1",
"context_bucket": "0.5k-4k",
},
"provenance": {
"source_dataset": "unit_test",
"license": "CC0-1.0",
"source_hash": "sha256:test",
"transformation": "date_substitution_v1",
"teacher_model": None,
"validator_version": "1.0.0",
"human_review": True,
},
"quality": {
"intentional_negative_simplified_hits": [],
"deterministic_verified": True,
},
}
class ValidationTests(unittest.TestCase):
def test_minimal_record_is_valid(self) -> None:
record = minimal_valid_record()
self.assertEqual(validate_record(record), [])
self.assertIsNone(assert_valid_record(record))
def test_all_ten_primitives_are_accepted(self) -> None:
self.assertEqual(
PRIMITIVES,
{
"COPY",
"EDIT",
"CLASSIFY",
"EXTRACT",
"LOCATE",
"ALIGN",
"MATCH",
"RANK",
"VERIFY",
"ROUTE",
},
)
for primitive in PRIMITIVES:
with self.subTest(primitive=primitive):
record = minimal_valid_record()
record["primitive"] = primitive
self.assertEqual(validate_record(record), [])
def test_required_root_field_and_invalid_primitive_are_reported(self) -> None:
record = minimal_valid_record()
del record["source_group"]
record["primitive"] = "SUMMARIZE"
errors = validate_record(record)
self.assertIn("missing required field: source_group", errors)
self.assertTrue(any("primitive must be one of" in error for error in errors))
def test_language_and_locale_must_be_taiwan_first(self) -> None:
record = minimal_valid_record()
record["language"] = ["en", "zh-TW"]
record["locale"] = "zh-Hant"
errors = validate_record(record)
self.assertIn("language[0] must be zh-TW", errors)
self.assertIn("locale must be zh-TW", errors)
def test_duplicate_unit_id_is_rejected(self) -> None:
record = minimal_valid_record()
record["inputs"][1]["units"][0]["id"] = "src:w01"
errors = validate_record(record)
self.assertIn("duplicate input unit ID: src:w01", errors)
def test_all_symbolic_reference_families_must_resolve(self) -> None:
cases = {
"evidence": {"evidence": ["missing:evidence"]},
"locations": {"locations": ["missing:location"]},
"copy": {"copy_ids": ["missing:copy"]},
"alignment": {
"alignments": [{"source": "src:w01", "target": "missing:alignment"}]
},
}
for name, extra in cases.items():
with self.subTest(name=name):
record = minimal_valid_record()
record["target"].update(extra)
errors = validate_record(record)
self.assertTrue(
any("references unknown input ID" in error for error in errors),
errors,
)
def test_required_evidence_and_reasonable_abstention(self) -> None:
record = minimal_valid_record()
record["target"]["errors"] = []
self.assertIn(
"task_spec.required_evidence is true but target has no evidence",
validate_record(record),
)
abstain = minimal_valid_record()
abstain["task_spec"]["allowed_error_types"] = ["INSUFFICIENT_EVIDENCE"]
abstain["target"] = {
"decision": "ABSTAIN",
"errors": [{"type": "INSUFFICIENT_EVIDENCE"}],
"action": {"tool": "ABSTAIN", "arguments": {}},
"confidence_target": 0.25,
}
self.assertEqual(validate_record(abstain), [])
abstain["task_spec"]["abstention_allowed"] = False
self.assertIn(
"target abstains but task_spec.abstention_allowed is false",
validate_record(abstain),
)
def test_positive_supervision_rejects_simplified_and_mainland_terms(self) -> None:
record = minimal_valid_record()
record["task_spec"]["instruction"] = "请检查这个软件的输出。"
errors = validate_record(record)
self.assertTrue(any("task_spec.instruction" in error for error in errors), errors)
record = minimal_valid_record()
record["target"]["post_edit"] = "請檢查這個軟件的輸出。"
errors = validate_record(record)
self.assertTrue(any("target.post_edit" in error for error in errors), errors)
def test_intentional_negative_may_contain_simplified_text_when_audited(self) -> None:
record = minimal_valid_record()
bad_text = "这个软件会在九月上线。"
record["inputs"][1]["units"][0]["text"] = bad_text
record["task_spec"]["allowed_error_types"] = ["CONSTRAINT_VIOLATION"]
record["target"]["errors"] = [
{
"type": "CONSTRAINT_VIOLATION",
"severity": "major",
"source_evidence": ["src:w01"],
"target_evidence": ["hyp:w01"],
}
]
record["quality"]["intentional_negative_simplified_hits"] = [
{
"unit_id": "hyp:w01",
"hits": find_disallowed_terms(bad_text),
}
]
self.assertEqual(validate_record(record), [])
def test_intentional_negative_requires_error_evidence_and_quality_log(self) -> None:
record = minimal_valid_record()
record["inputs"][1]["units"][0]["text"] = "这个软件会在九月上线。"
record["target"]["errors"] = []
errors = validate_record(record)
self.assertTrue(any("not captured by target.errors evidence" in error for error in errors))
self.assertTrue(
any("intentional_negative_simplified_hits is missing ID" in error for error in errors)
)
def test_tool_and_confidence_are_closed_and_bounded(self) -> None:
self.assertNotIn("RUN_ARBITRARY_PYTHON", ACTION_TOOLS)
record = minimal_valid_record()
record["target"]["action"]["tool"] = "RUN_ARBITRARY_PYTHON"
record["target"]["confidence_target"] = 1.01
errors = validate_record(record)
self.assertIn("target.action.tool must be a closed-world action tool", errors)
self.assertIn(
"target.confidence_target must be a finite number in [0, 1]", errors
)
def test_assert_valid_record_exposes_all_errors(self) -> None:
record = minimal_valid_record()
record["locale"] = "zh-CN"
record["target"]["confidence_target"] = float("nan")
with self.assertRaises(RecordValidationError) as raised:
assert_valid_record(record)
self.assertGreaterEqual(len(raised.exception.errors), 2)
def test_schema_is_json_and_matches_closed_world_constants(self) -> None:
schema_path = Path(__file__).parents[1] / "schema" / "canonical_record.schema.json"
schema = json.loads(schema_path.read_text(encoding="utf-8"))
self.assertEqual(set(schema["required"]), set(minimal_valid_record()))
self.assertEqual(set(schema["$defs"]["primitive"]["enum"]), PRIMITIVES)
self.assertEqual(set(schema["$defs"]["action_tool"]["enum"]), ACTION_TOOLS)
if __name__ == "__main__":
unittest.main()
|