File size: 13,567 Bytes
5d05dd0
531d9de
 
 
 
 
5d05dd0
 
 
 
 
 
 
531d9de
 
 
 
 
 
 
 
 
5d05dd0
 
531d9de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5d05dd0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
531d9de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5d05dd0
531d9de
 
5d05dd0
531d9de
 
 
 
 
 
 
 
 
 
 
 
 
 
5d05dd0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
531d9de
 
5d05dd0
 
 
531d9de
 
 
 
 
5d05dd0
531d9de
 
 
5d05dd0
531d9de
 
 
 
5d05dd0
531d9de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5d05dd0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
531d9de
 
 
 
 
 
 
 
 
 
 
 
 
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
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
"""Topic and Term routing through the provider-neutral model interface."""

from __future__ import annotations

from pydantic import BaseModel, ConfigDict, Field

from gcmd_classifier.classification.candidates import (
    TermCandidate,
    TopicCandidate,
    build_term_candidates,
    build_topic_candidates,
    validate_term_candidate_relationship,
)
from gcmd_classifier.config import ModelSettings
from gcmd_classifier.errors import UnknownCandidateIDError
from gcmd_classifier.llm.base import (
    ModelClient,
    ModelRequest,
    ModelStage,
    RetryPolicy,
    generate_with_retries,
)
from gcmd_classifier.llm.prompts import ParentContext, build_term_prompt, build_topic_prompt
from gcmd_classifier.llm.schemas import CandidateDecision, TermResponse, TopicResponse
from gcmd_classifier.models import ArticleRecord, OutputError, OutputWarning, SupportType
from gcmd_classifier.vocabulary.index import VocabularyIndex


class ModelCallMetadata(BaseModel):
    """Safe model-call metadata retained in routing results."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    provider: str
    model_name: str
    prompt_version: str
    retry_count: int = Field(ge=0)
    duration_seconds: float | None = Field(default=None, ge=0.0)
    input_tokens: int | None = Field(default=None, ge=0)
    output_tokens: int | None = Field(default=None, ge=0)
    total_tokens: int | None = Field(default=None, ge=0)
    estimated_cost: float | None = Field(default=None, ge=0.0)


class TopicBranchSeed(BaseModel):
    """Seed for a later Term-routing branch under one selected Topic."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    branch_id: str = Field(min_length=1)
    topic_uuid: str = Field(min_length=1)
    topic_name: str = Field(min_length=1)
    topic_level: str = Field(pattern="^Topic$")
    topic_canonical_path: str = Field(min_length=1)
    evidence: str = Field(min_length=1)
    support_type: SupportType
    confidence: float | None = Field(default=None, ge=0.0, le=1.0)
    reason: str | None = None
    candidate_id: str = Field(min_length=1)
    prompt_version: str
    model_provider: str
    model_name: str
    retry_count: int = Field(ge=0)


class TopicRoutingResult(BaseModel):
    """Result of Topic routing for one article."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    branches: tuple[TopicBranchSeed, ...] = Field(default_factory=tuple)
    no_selection_reason: str | None = None
    model_metadata: ModelCallMetadata
    warnings: tuple[OutputWarning, ...] = Field(default_factory=tuple)
    errors: tuple[OutputError, ...] = Field(default_factory=tuple)

    @property
    def selected_count(self) -> int:
        """Number of selected Topic branches."""
        return len(self.branches)

    @property
    def is_no_topic(self) -> bool:
        """Whether routing completed successfully with no selected Topics."""
        return not self.branches


class TermBranchSeed(BaseModel):
    """Seed for a later Variable-routing branch under one selected Term."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    branch_id: str = Field(min_length=1)
    parent_topic_uuid: str = Field(min_length=1)
    parent_topic_name: str = Field(min_length=1)
    term_uuid: str = Field(min_length=1)
    term_name: str = Field(min_length=1)
    term_level: str = Field(pattern="^Term$")
    term_canonical_path: str = Field(min_length=1)
    evidence: str = Field(min_length=1)
    support_type: SupportType
    confidence: float | None = Field(default=None, ge=0.0, le=1.0)
    reason: str | None = None
    candidate_id: str = Field(min_length=1)
    parent_branch_id: str = Field(min_length=1)
    prompt_version: str
    model_provider: str
    model_name: str
    retry_count: int = Field(ge=0)


class TopicStopResult(BaseModel):
    """Successful Term-routing stop at the selected Topic parent."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    branch_id: str = Field(min_length=1)
    topic_uuid: str = Field(min_length=1)
    topic_name: str = Field(min_length=1)
    topic_level: str = Field(pattern="^Topic$")
    topic_canonical_path: str = Field(min_length=1)
    stop_reason: str = Field(min_length=1)
    parent_evidence: str = Field(min_length=1)
    parent_support_type: SupportType
    parent_confidence: float | None = Field(default=None, ge=0.0, le=1.0)
    topic_candidate_id: str = Field(min_length=1)
    prompt_version: str
    model_provider: str
    model_name: str
    retry_count: int = Field(ge=0)


class TermRoutingResult(BaseModel):
    """Result of Term routing under one selected Topic branch."""

    model_config = ConfigDict(extra="forbid", frozen=True)

    term_branches: tuple[TermBranchSeed, ...] = Field(default_factory=tuple)
    stop_at_topic: TopicStopResult | None = None
    model_metadata: ModelCallMetadata
    warnings: tuple[OutputWarning, ...] = Field(default_factory=tuple)
    errors: tuple[OutputError, ...] = Field(default_factory=tuple)

    @property
    def selected_count(self) -> int:
        """Number of selected Term branches."""
        return len(self.term_branches)

    @property
    def stopped_at_topic(self) -> bool:
        """Whether routing stopped successfully at the parent Topic."""
        return self.stop_at_topic is not None


def route_topics(
    *,
    article: ArticleRecord,
    vocabulary: VocabularyIndex,
    model_client: ModelClient,
    settings: ModelSettings,
    retry_policy: RetryPolicy | None = None,
) -> TopicRoutingResult:
    """Route an article to zero, one, or more valid GCMD Topics."""
    candidates = build_topic_candidates(vocabulary)
    candidates_by_id = {candidate.candidate_id: candidate for candidate in candidates}
    prompt = build_topic_prompt(
        article=article,
        candidates=tuple(candidate.prompt_candidate for candidate in candidates),
        prompt_version=settings.prompt_version_topic,
    )
    request = ModelRequest.from_settings(
        stage=ModelStage.TOPIC,
        prompt=prompt,
        response_schema=TopicResponse,
        settings=settings,
        metadata={
            "DOI": article.DOI,
            "candidate_ids": tuple(candidates_by_id),
        },
    )
    response = generate_with_retries(
        model_client,
        request,
        retry_policy or RetryPolicy.from_settings(settings),
    )
    _validate_selected_candidate_ids(response.parsed.selected, candidates_by_id, stage="Topic")
    metadata = _model_metadata(response)
    branches = tuple(
        _topic_branch_seed(
            decision=decision,
            candidate=candidates_by_id[decision.candidate_id],
            vocabulary=vocabulary,
            metadata=metadata,
        )
        for decision in response.parsed.selected
    )
    return TopicRoutingResult(
        branches=branches,
        no_selection_reason=response.parsed.no_selection_reason if not branches else None,
        model_metadata=metadata,
    )


def route_terms(
    *,
    article: ArticleRecord,
    topic_branch: TopicBranchSeed,
    vocabulary: VocabularyIndex,
    model_client: ModelClient,
    settings: ModelSettings,
    retry_policy: RetryPolicy | None = None,
) -> TermRoutingResult:
    """Route one selected Topic branch to zero, one, or more direct Term children."""
    candidates = build_term_candidates(vocabulary, topic_uuid=topic_branch.topic_uuid)
    candidates_by_id = {candidate.candidate_id: candidate for candidate in candidates}
    parent_context = ParentContext(
        candidate_id=topic_branch.candidate_id,
        name=topic_branch.topic_name,
        level="Topic",
        canonical_path=topic_branch.topic_canonical_path,
    )
    prompt = build_term_prompt(
        article=article,
        parent=parent_context,
        candidates=tuple(candidate.prompt_candidate for candidate in candidates),
        prompt_version=settings.prompt_version_term,
    )
    request = ModelRequest.from_settings(
        stage=ModelStage.TERM,
        prompt=prompt,
        response_schema=TermResponse,
        settings=settings,
        metadata={
            "DOI": article.DOI,
            "parent_topic_uuid": topic_branch.topic_uuid,
            "parent_branch_id": topic_branch.branch_id,
            "candidate_ids": tuple(candidates_by_id),
        },
    )
    response = generate_with_retries(
        model_client,
        request,
        retry_policy or RetryPolicy.from_settings(settings),
    )
    _validate_selected_candidate_ids(response.parsed.selected, candidates_by_id, stage="Term")
    metadata = _model_metadata(response)
    term_branches = tuple(
        _term_branch_seed(
            decision=decision,
            candidate=candidates_by_id[decision.candidate_id],
            topic_branch=topic_branch,
            vocabulary=vocabulary,
            metadata=metadata,
        )
        for decision in response.parsed.selected
    )
    stop_at_topic = None
    if response.parsed.stop_at_parent:
        stop_at_topic = _topic_stop_result(
            topic_branch=topic_branch,
            stop_reason=response.parsed.stop_reason,
            metadata=metadata,
        )
    return TermRoutingResult(
        term_branches=term_branches,
        stop_at_topic=stop_at_topic,
        model_metadata=metadata,
    )


def _validate_selected_candidate_ids(
    selected: list[CandidateDecision],
    candidates_by_id: dict[str, TopicCandidate] | dict[str, TermCandidate],
    *,
    stage: str,
) -> None:
    seen: set[str] = set()
    for decision in selected:
        if decision.candidate_id not in candidates_by_id:
            raise UnknownCandidateIDError(
                f"{stage} model selected unknown candidate_id {decision.candidate_id!r}."
            )
        if decision.candidate_id in seen:
            raise UnknownCandidateIDError(
                f"{stage} model selected duplicate candidate_id {decision.candidate_id!r}."
            )
        seen.add(decision.candidate_id)


def _topic_branch_seed(
    *,
    decision: CandidateDecision,
    candidate: TopicCandidate,
    vocabulary: VocabularyIndex,
    metadata: ModelCallMetadata,
) -> TopicBranchSeed:
    topic = vocabulary.get(candidate.topic_uuid)
    return TopicBranchSeed(
        branch_id=f"topic:{candidate.candidate_id}",
        topic_uuid=topic.UUID,
        topic_name=topic.name,
        topic_level=topic.level,
        topic_canonical_path=topic.canonical_path,
        evidence=decision.evidence,
        support_type=decision.support_type,
        confidence=decision.confidence,
        reason=decision.reason,
        candidate_id=decision.candidate_id,
        prompt_version=metadata.prompt_version,
        model_provider=metadata.provider,
        model_name=metadata.model_name,
        retry_count=metadata.retry_count,
    )


def _term_branch_seed(
    *,
    decision: CandidateDecision,
    candidate: TermCandidate,
    topic_branch: TopicBranchSeed,
    vocabulary: VocabularyIndex,
    metadata: ModelCallMetadata,
) -> TermBranchSeed:
    validate_term_candidate_relationship(
        candidate,
        selected_topic_uuid=topic_branch.topic_uuid,
        index=vocabulary,
    )
    term = vocabulary.get(candidate.term_uuid)
    return TermBranchSeed(
        branch_id=f"{topic_branch.branch_id}/term:{candidate.candidate_id}",
        parent_topic_uuid=topic_branch.topic_uuid,
        parent_topic_name=topic_branch.topic_name,
        term_uuid=term.UUID,
        term_name=term.name,
        term_level=term.level,
        term_canonical_path=term.canonical_path,
        evidence=decision.evidence,
        support_type=decision.support_type,
        confidence=decision.confidence,
        reason=decision.reason,
        candidate_id=decision.candidate_id,
        parent_branch_id=topic_branch.branch_id,
        prompt_version=metadata.prompt_version,
        model_provider=metadata.provider,
        model_name=metadata.model_name,
        retry_count=metadata.retry_count,
    )


def _topic_stop_result(
    *,
    topic_branch: TopicBranchSeed,
    stop_reason: str | None,
    metadata: ModelCallMetadata,
) -> TopicStopResult:
    if not stop_reason:
        raise ValueError("stop_at_parent responses require a stop_reason.")
    return TopicStopResult(
        branch_id=topic_branch.branch_id,
        topic_uuid=topic_branch.topic_uuid,
        topic_name=topic_branch.topic_name,
        topic_level=topic_branch.topic_level,
        topic_canonical_path=topic_branch.topic_canonical_path,
        stop_reason=stop_reason,
        parent_evidence=topic_branch.evidence,
        parent_support_type=topic_branch.support_type,
        parent_confidence=topic_branch.confidence,
        topic_candidate_id=topic_branch.candidate_id,
        prompt_version=metadata.prompt_version,
        model_provider=metadata.provider,
        model_name=metadata.model_name,
        retry_count=metadata.retry_count,
    )


def _model_metadata(response) -> ModelCallMetadata:
    token_usage = response.token_usage
    return ModelCallMetadata(
        provider=response.provider,
        model_name=response.model_name,
        prompt_version=response.prompt_version,
        retry_count=response.retry_count,
        duration_seconds=response.duration_seconds,
        input_tokens=None if token_usage is None else token_usage.input_tokens,
        output_tokens=None if token_usage is None else token_usage.output_tokens,
        total_tokens=None if token_usage is None else token_usage.total_tokens,
        estimated_cost=response.estimated_cost,
    )