disclosure-bureau/scripts/rebuild_doc65_page_rebuilder.py

287 lines
12 KiB
Python

#!/usr/bin/env python3
"""
Page rebuilder for doc-65-hs1-834228961-62-hq-83894-section-1
Processes pages 1-150 using vision (PNGs at p-001.png .. p-150.png)
Outputs JSON per page with chunks list.
"""
import anthropic
import base64
import json
import os
import sys
import time
from concurrent.futures import ThreadPoolExecutor, as_completed
from pathlib import Path
DOC_ID = "doc-65-hs1-834228961-62-hq-83894-section-1"
DOC_TITLE = "65 HS1-834228961 62-HQ-83894 Section 1 (FBI UFO/UAP Investigative File)"
PNG_DIR = Path("/Users/guto/ufo/processing/png/doc-65-hs1-834228961-62-hq-83894-section-1")
OCR_DIR = Path("/Users/guto/ufo/processing/ocr/doc-65-hs1-834228961-62-hq-83894-section-1")
OUTPUT_DIR = Path("/Users/guto/ufo/raw/doc-65-hs1-834228961-62-hq-83894-section-1")
TOTAL_PAGES = 150
client = anthropic.Anthropic()
CHUNK_TYPES = [
"letterhead", "classification_banner", "header", "subheader",
"paragraph", "list_item", "caption", "footnote", "page_number",
"signature_block", "stamp", "redaction_block", "image", "table_marker",
"form_field", "watermark", "separator", "blank"
]
PAGE_REBUILDER_PROMPT = """You are a page-rebuilder for a declassified UAP/UFO government document.
Document: {doc_title}
Page: {page_number} of {total_pages}
Analyze this page image carefully and extract ALL content as structured chunks.
Return a JSON object with this exact structure:
{{
"page_number": {page_number},
"chunks": [
{{
"order_in_page": 1,
"type": "<chunk_type>",
"content_en": "<English text or description>",
"content_pt_br": "<Brazilian Portuguese translation>",
"bbox": {{"x": 0.0, "y": 0.0, "w": 1.0, "h": 0.1}},
"classification": null,
"formatting": [],
"cross_page_hint": "self_contained",
"ocr_confidence": 0.85,
"ocr_source_lines": [],
"redaction_code": null,
"redaction_inferred_content_type": null,
"image_type": null,
"ufo_anomaly_detected": false,
"ufo_anomaly_type": null,
"ufo_anomaly_rationale": null,
"cryptid_anomaly_detected": false,
"cryptid_anomaly_type": null,
"cryptid_anomaly_rationale": null,
"image_description_en": null,
"image_description_pt_br": null,
"extracted_text": null
}}
]
}}
CHUNK TYPES (use only these):
letterhead, classification_banner, header, subheader, paragraph, list_item,
caption, footnote, page_number, signature_block, stamp, redaction_block,
image, table_marker, form_field, watermark, separator, blank
RULES:
1. Extract EVERY element on the page — nothing is skipped
2. bbox: normalized coordinates (x=left, y=top, w=width, h=height) relative to page size (0.0 to 1.0)
3. content_en: verbatim OCR text for text chunks; for images describe what you see
4. content_pt_br: Brazilian Portuguese translation of content_en (NOT European Portuguese)
5. Preserve UTF-8 accents: ç, ã, á, é, í, ó, ú, â, ê, ô, à
6. For redacted sections: type="redaction_block", content_en="[REDACTED]", content_pt_br="[REDACTADO]"
7. For images/photos: type="image", describe the visual content in image_description_en and image_description_pt_br
8. For stamps: type="stamp"
9. classification: extract classification markings if visible (e.g. "SECRET", "CONFIDENTIAL")
10. formatting: array of applicable ["bold", "italic", "underline", "all_caps", "handwritten", "typewritten"]
11. cross_page_hint: "self_contained" | "continues_to_next" | "continues_from_prev"
12. ufo_anomaly_detected: true if chunk contains UAP/UFO sighting data, coordinates, witness accounts, or anomalous phenomena
13. If page is blank: return one chunk with type="blank"
14. Order chunks top-to-bottom, left-to-right as they appear on the page
15. Return ONLY valid JSON, no markdown code blocks, no extra text
OCR text hint (may be empty or garbled):
{ocr_text}
"""
def load_image_b64(png_path: Path) -> str:
with open(png_path, "rb") as f:
return base64.standard_b64encode(f.read()).decode("utf-8")
def load_ocr(page_num: int) -> str:
txt_path = OCR_DIR / f"p-{page_num:03d}.txt"
if txt_path.exists():
try:
content = txt_path.read_text(encoding="utf-8").strip()
return content if content else "(empty)"
except Exception:
return "(unreadable)"
return "(not found)"
def process_page(page_num: int, retries: int = 3) -> dict:
png_path = PNG_DIR / f"p-{page_num:03d}.png"
if not png_path.exists():
print(f" WARNING: PNG not found for page {page_num}: {png_path}", file=sys.stderr)
return {
"page_number": page_num,
"chunks": [{
"order_in_page": 1,
"type": "blank",
"content_en": "[PAGE NOT FOUND]",
"content_pt_br": "[PÁGINA NÃO ENCONTRADA]",
"bbox": {"x": 0.0, "y": 0.0, "w": 1.0, "h": 1.0},
"classification": None,
"formatting": [],
"cross_page_hint": "self_contained",
"ocr_confidence": 0.0,
"ocr_source_lines": [],
"redaction_code": None,
"redaction_inferred_content_type": None,
"image_type": None,
"ufo_anomaly_detected": False,
"ufo_anomaly_type": None,
"ufo_anomaly_rationale": None,
"cryptid_anomaly_detected": False,
"cryptid_anomaly_type": None,
"cryptid_anomaly_rationale": None,
"image_description_en": None,
"image_description_pt_br": None,
"extracted_text": None
}]
}
ocr_text = load_ocr(page_num)
img_b64 = load_image_b64(png_path)
prompt = PAGE_REBUILDER_PROMPT.format(
doc_title=DOC_TITLE,
page_number=page_num,
total_pages=TOTAL_PAGES,
ocr_text=ocr_text[:2000] # cap at 2000 chars
)
for attempt in range(retries):
try:
response = client.messages.create(
model="claude-haiku-4-5",
max_tokens=4096,
messages=[{
"role": "user",
"content": [
{
"type": "image",
"source": {
"type": "base64",
"media_type": "image/png",
"data": img_b64
}
},
{
"type": "text",
"text": prompt
}
]
}]
)
raw = response.content[0].text.strip()
# Strip markdown code blocks if present
if raw.startswith("```"):
lines = raw.split("\n")
raw = "\n".join(lines[1:-1] if lines[-1] == "```" else lines[1:])
result = json.loads(raw)
result["page_number"] = page_num # ensure correct
print(f" Page {page_num:3d} done — {len(result.get('chunks', []))} chunks", flush=True)
return result
except json.JSONDecodeError as e:
print(f" Page {page_num} JSON error (attempt {attempt+1}): {e}", file=sys.stderr)
if attempt == retries - 1:
return {
"page_number": page_num,
"chunks": [{
"order_in_page": 1,
"type": "blank",
"content_en": f"[PARSE ERROR: {str(e)[:100]}]",
"content_pt_br": f"[ERRO DE ANÁLISE: {str(e)[:100]}]",
"bbox": {"x": 0.0, "y": 0.0, "w": 1.0, "h": 1.0},
"classification": None,
"formatting": [],
"cross_page_hint": "self_contained",
"ocr_confidence": 0.0,
"ocr_source_lines": [],
"redaction_code": None,
"redaction_inferred_content_type": None,
"image_type": None,
"ufo_anomaly_detected": False,
"ufo_anomaly_type": None,
"ufo_anomaly_rationale": None,
"cryptid_anomaly_detected": False,
"cryptid_anomaly_type": None,
"cryptid_anomaly_rationale": None,
"image_description_en": None,
"image_description_pt_br": None,
"extracted_text": None
}]
}
time.sleep(2 ** attempt)
except Exception as e:
print(f" Page {page_num} API error (attempt {attempt+1}): {e}", file=sys.stderr)
if attempt == retries - 1:
return {
"page_number": page_num,
"chunks": [{
"order_in_page": 1,
"type": "blank",
"content_en": f"[API ERROR: {str(e)[:100]}]",
"content_pt_br": f"[ERRO DE API: {str(e)[:100]}]",
"bbox": {"x": 0.0, "y": 0.0, "w": 1.0, "h": 1.0},
"classification": None,
"formatting": [],
"cross_page_hint": "self_contained",
"ocr_confidence": 0.0,
"ocr_source_lines": [],
"redaction_code": None,
"redaction_inferred_content_type": None,
"image_type": None,
"ufo_anomaly_detected": False,
"ufo_anomaly_type": None,
"ufo_anomaly_rationale": None,
"cryptid_anomaly_detected": False,
"cryptid_anomaly_type": None,
"cryptid_anomaly_rationale": None,
"image_description_en": None,
"image_description_pt_br": None,
"extracted_text": None
}]
}
time.sleep(2 ** attempt)
def main():
pages = list(range(1, TOTAL_PAGES + 1))
results = {}
print(f"Processing {len(pages)} pages in parallel batches of 5...")
batch_size = 5
for batch_start in range(0, len(pages), batch_size):
batch = pages[batch_start:batch_start + batch_size]
print(f"Batch {batch_start//batch_size + 1}: pages {batch[0]}-{batch[-1]}")
with ThreadPoolExecutor(max_workers=5) as executor:
future_to_page = {executor.submit(process_page, p): p for p in batch}
for future in as_completed(future_to_page):
page_num = future_to_page[future]
try:
result = future.result()
results[page_num] = result
except Exception as e:
print(f" Page {page_num} FATAL: {e}", file=sys.stderr)
# Small pause between batches to avoid rate limits
if batch_start + batch_size < len(pages):
time.sleep(1)
# Save intermediate results
out_path = OUTPUT_DIR / "_pages_raw.json"
sorted_results = [results[p] for p in sorted(results.keys())]
with open(out_path, "w", encoding="utf-8") as f:
json.dump(sorted_results, f, ensure_ascii=False, indent=2)
print(f"\nSaved {len(sorted_results)} pages to {out_path}")
total_chunks = sum(len(r.get("chunks", [])) for r in sorted_results)
print(f"Total chunks: {total_chunks}")
if __name__ == "__main__":
main()