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
391
392
393
394
395
396
397
398
399
400
401
402
403"""Pydantic response models for the Pipeline Viewer dev tool."""
from __future__ import annotations
from enum import Enum
from typing import Literal
from pydantic import BaseModel, Field
# ---------------------------------------------------------------------------
# Phase 1 — Structure Analysis models
# ---------------------------------------------------------------------------
class LayoutType(str, Enum):
"""Page layout classification based on text flow."""
SINGLE_COLUMN = "single_column"
DOUBLE_COLUMN = "double_column"
PRESENTATION = "presentation"
POSTER = "poster"
class PageAttributes(BaseModel):
"""Compositional page characteristics detected during structure analysis.
Each attribute maps to a procedure fragment that provides targeted
correction guidance for Phase 2 agents.
"""
layout: LayoutType
is_academic: bool = False
has_images: bool = False
has_tables: bool = False
has_lists: bool = False
has_equations: bool = False
is_scanned: bool = False
class HeadingRecommendation(BaseModel):
"""A heading identified on a page with its recommended level."""
text: str
"""Heading text as read from the page image (ground truth)."""
recommended_level: int
"""Correct heading level (1-6) based on document outline context."""
reasoning: str
"""Why this level was chosen, referencing the outline."""
class FootnoteInfo(BaseModel):
"""A footnote found on a page."""
number: str
"""Footnote marker (e.g. "1", "2", "*")."""
body_text: str
"""The footnote content text."""
source_page: int
"""Page where this footnote appears."""
class OutlineEntry(BaseModel):
"""A single entry in the accumulated document outline."""
level: int
"""Heading level (1-6)."""
text: str
"""Heading text."""
page: int
"""Page number where this heading appears."""
reasoning: str = ""
"""Why this level was chosen (from per-page analysis)."""
class CodeBlockInfo(BaseModel):
"""A code block found on a page."""
language: str
"""Detected programming language (e.g. "python", "java", "sql")."""
first_line: str
"""First line of the code block as visible in the page image."""
last_line: str
"""Last line of the code block as visible in the page image."""
page: int
"""Page where this code block appears."""
reasoning: str
"""How the language was identified (syntax, surrounding context, etc.)."""
class StructurePageOutput(BaseModel):
"""What the Phase 1 agent returns for a single page.
The agent analyzes the page image + markdown and reports structural
findings. It does NOT modify the markdown — that happens in Phase 2.
"""
page_attributes: PageAttributes
"""Compositional page characteristics detected from the page image."""
headings: list[HeadingRecommendation] = Field(default_factory=list)
"""Headings found on this page with recommended levels."""
footnotes: list[FootnoteInfo] = Field(default_factory=list)
"""Footnotes found on this page (marker + body text)."""
code_blocks: list[CodeBlockInfo] = Field(default_factory=list)
"""Code blocks found on this page with detected language."""
class StructureResult(BaseModel):
"""Complete Phase 1 output, accumulated across all pages.
Built by the orchestrator from individual StructurePageOutput results.
Consumed by Phase 2 (per-page agents) and Phase 3 (cross-page).
"""
page_attributes: dict[int, PageAttributes] = Field(default_factory=dict)
"""Page number -> compositional attributes detected for that page."""
outline: list[OutlineEntry] = Field(default_factory=list)
"""Full document outline in page order."""
footnotes: list[FootnoteInfo] = Field(default_factory=list)
"""All footnotes found across all pages."""
code_blocks: list[CodeBlockInfo] = Field(default_factory=list)
"""All code blocks found across all pages."""
# ---------------------------------------------------------------------------
# Phase 1c — Heading Reconciliation models
# ---------------------------------------------------------------------------
class HeadingReconciliationOutput(BaseModel):
"""Output from heading reconciliation — corrected outline."""
outline: list[OutlineEntry]
"""Globally reconciled outline with corrected heading levels."""
reasoning: str
"""Global rationale for hierarchy decisions."""
# ---------------------------------------------------------------------------
# Phase 1d — Section Map models
# ---------------------------------------------------------------------------
class SectionEntry(BaseModel):
"""A single section in the document."""
index: int
"""0-based section index (0 = preamble)."""
heading_text: str
"""Heading text ('(Preamble)' for index 0)."""
heading_level: int
"""1-6 (0 for preamble)."""
pages: list[int]
"""Page numbers this section spans."""
markdown: str
"""Section markdown (heading through next heading)."""
outline_position: int
"""Index in the full outline (-1 for preamble)."""
class SectionMap(BaseModel):
"""Maps the document outline to concrete page ranges and markdown slices."""
sections: list[SectionEntry] = Field(default_factory=list)
# ---------------------------------------------------------------------------
# Phase 2 — Content Correction models
# ---------------------------------------------------------------------------
class ImageDescriptionResult(BaseModel):
"""Result from the image description subagent."""
image_category: Literal["decorative", "informative", "complex_informative"]
"""Classification of the image's accessibility role."""
alt_text: str
"""WCAG-compliant alt text (empty string for decorative images)."""
is_decorative: bool = False
"""True if the image is purely decorative (borders, spacers, logos)."""
confidence: str = "high"
"""Confidence level: high, medium, or low."""
reasoning: str = ""
"""Why this alt text was chosen."""
class TableReconstructionResult(BaseModel):
"""Result from the table reconstruction subagent."""
table_format: Literal["markdown", "html"]
"""Whether the output is markdown or HTML."""
reconstructed_content: str
"""The corrected table content (markdown or HTML)."""
caption: str
"""WCAG table caption describing the table's purpose."""
has_row_headers: bool = False
"""True if the table has row headers (leftmost column)."""
has_merged_cells: bool = False
"""True if the table has merged cells (colspan/rowspan)."""
confidence: str = "high"
"""Confidence level: high, medium, or low."""
reasoning: str = ""
"""Why this reconstruction was chosen."""
class ListReconstructionResult(BaseModel):
"""Result from the list reconstruction subagent."""
list_type: Literal["unordered", "ordered", "definition"]
"""Classification of the list type."""
reconstructed_content: str
"""The corrected list content (markdown or HTML for definition lists)."""
confidence: str = "high"
"""Confidence level: high, medium, or low."""
reasoning: str = ""
"""Why this reconstruction was chosen."""
class SectionCorrectionResult(BaseModel):
"""What a section correction agent returns."""
section_index: int
heading_text: str
corrected_markdown: str
changes: list[DocumentChange] = Field(default_factory=list)
issues: list[str] = Field(default_factory=list)
pages: list[int] = Field(default_factory=list)
input_tokens: int = 0
output_tokens: int = 0
describer_input_tokens: int = 0
describer_output_tokens: int = 0
# ---------------------------------------------------------------------------
# Phase 2 (legacy) — Page Content Correction models
# ---------------------------------------------------------------------------
class PageCorrectionResult(BaseModel):
"""What a Phase 2 page agent returns."""
page: int
"""Page number that was corrected."""
corrected_markdown: str
"""The page markdown after all corrections applied."""
changes: list[DocumentChange] = Field(default_factory=list)
"""Log of every str_replace applied."""
issues: list[str] = Field(default_factory=list)
"""Uncertainties or failed edits flagged for human review."""
input_tokens: int = 0
"""LLM input tokens consumed for this page."""
output_tokens: int = 0
"""LLM output tokens generated for this page."""
describer_input_tokens: int = 0
"""Input tokens consumed by image describer subagent calls."""
describer_output_tokens: int = 0
"""Output tokens from image describer subagent calls."""
table_reconstructor_input_tokens: int = 0
"""Input tokens consumed by table reconstructor subagent calls."""
table_reconstructor_output_tokens: int = 0
"""Output tokens from table reconstructor subagent calls."""
list_reconstructor_input_tokens: int = 0
"""Input tokens consumed by list reconstructor subagent calls."""
list_reconstructor_output_tokens: int = 0
"""Output tokens from list reconstructor subagent calls."""
# ---------------------------------------------------------------------------
# Phase 3 — Cross-page models
# ---------------------------------------------------------------------------
class BoundaryContext(BaseModel):
"""Context for a boundary fix between two adjacent pages."""
page_before: int
page_after: int
tail_text: str
"""Last ~10 lines of the earlier page."""
head_text: str
"""First ~10 lines of the later page."""
# ---------------------------------------------------------------------------
# Viewer response models (used by API)
# ---------------------------------------------------------------------------
class DocumentChange(BaseModel):
"""A single change made by a pipeline step."""
page: int
old_text: str
new_text: str
reasoning: str
stage: str
class FigureData(BaseModel):
"""An extracted figure/picture from the document."""
ref_id: str
caption: str
page_number: int
image_base64: str
class TableData(BaseModel):
"""A markdown table extracted from a page."""
ref_id: str
"""Table reference ID (e.g. "table-1")."""
page_number: int
"""Page where this table appears."""
markdown_content: str
"""The extracted markdown table text."""
class StepResult(BaseModel):
"""Result of a single pipeline step."""
name: str
display_name: str
version_before: str | None = None
version_after: str
elapsed_ms: int
changes: list[DocumentChange] = Field(default_factory=list)
metadata: dict = Field(default_factory=dict)
skipped: bool = False
error: str | None = None
input_tokens: int = 0
output_tokens: int = 0
cost_cents: float = 0.0
class PipelineViewerResult(BaseModel):
"""Top-level response from the pipeline viewer."""
filename: str
total_pages: int
versions: dict[str, str] = Field(default_factory=dict)
page_images: dict[str, str] = Field(default_factory=dict)
page_markdowns: dict[str, dict[str, str]] = Field(default_factory=dict)
figures: list[FigureData] = Field(default_factory=list)
steps: list[StepResult] = Field(default_factory=list)
stats: dict = Field(default_factory=dict)
warnings: list[str] = Field(default_factory=list)
"""Human-readable warnings from PDF classification."""
# Resolve forward references
PageCorrectionResult.model_rebuild()
SectionCorrectionResult.model_rebuild()