"""Strip condition tells from expert review form.
Removes phrases that would reveal which condition (Control/RAG/Pragmatics)
produced each response, ensuring proper blinding for expert review.
Usage:
python scripts/strip_review_tells.py
"""
import re
from pathlib import Path
def strip_tells(text: str) -> str:
"""Remove all condition tells from text."""
# Remove entire disclaimer sentences that contain tells
# Pattern: sentence that starts with disclaimer and contains tells
disclaimer_patterns = [
r'I appreciate your question[^.]+but I need to clarify something important:[^.]+Census methodology documentation[^.]+\.',
r'Based on general knowledge,[^.]+Census methodology documentation[^.]+\.',
r'I don\'t have[^.]+in Census methodology documentation[^.]+\.',
r'I don\'t have[^.]+in my (available )?reference materials[^.]+\.',
r'I don\'t have[^.]+in the provided reference materials[^.]+\.',
r'[^.]+Census methodology documentation I have available[^.]+\.',
r'[^.]+handbooks that explain[^.]+\.',
r'[^.]+methodology documents and handbooks[^.]+\.',
]
for pattern in disclaimer_patterns:
text = re.sub(pattern, '', text, flags=re.IGNORECASE | re.DOTALL)
# Remove specific tell phrases (case insensitive)
tell_phrases = [
'Census methodology documentation focuses on',
'Census methodology documentation',
'in Census methodology documentation',
'Census methodology documentation I have available',
'in my available reference materials',
'in my reference materials',
'my available reference materials',
'reference materials',
'methodology documentation',
'the provided reference materials',
'provided reference materials',
'documents I have',
'handbooks that explain',
'methodology documents and handbooks',
'in the provided reference',
'according to the methodology',
'the methodology documentation',
'methodology handbooks',
]
for phrase in tell_phrases:
# Replace with empty string
text = re.sub(re.escape(phrase), '', text, flags=re.IGNORECASE)
# Clean up resulting formatting issues
# Remove "These documents explain:" orphaned headers
text = re.sub(r'These documents explain:\s*-', '-', text, flags=re.IGNORECASE)
# Remove orphaned "However," at start of sentence
text = re.sub(r'\.\s*However,\s*I can guide', '. I can guide', text)
# Remove "Based on general knowledge," when it starts a sentence awkwardly
text = re.sub(r'Based on general knowledge,\s*I do not have current',
'To find current', text, flags=re.IGNORECASE)
text = re.sub(r'Based on general knowledge,\s*real-time or current',
'To find current', text, flags=re.IGNORECASE)
# Remove doubled words from replacements
text = re.sub(r'\b(\w+)\s+\1\b', r'\1', text)
# Clean up extra whitespace - preserve newlines
# First protect newlines by temporarily replacing them
text = re.sub(r'\n', '<<<NEWLINE>>>', text)
# Now collapse multiple spaces
text = re.sub(r' +', ' ', text)
# Restore newlines
text = re.sub(r'<<<NEWLINE>>>', '\n', text)
# Clean up multiple newlines to at most double
text = re.sub(r'\n\n\n+', '\n\n', text)
# Fix sentences that start with lowercase after a period
text = re.sub(r'\.\s+([a-z])', lambda m: '. ' + m.group(1).upper(), text)
return text.strip()
def main():
form_path = Path('talks/fcsm_2026/expert_review_form.md')
print("="*70)
print("STRIPPING CONDITION TELLS FROM EXPERT REVIEW FORM")
print("="*70)
# Read file
print(f"\nReading {form_path}...")
with open(form_path) as f:
content = f.read()
# Strip tells
print("Stripping tells...")
cleaned = strip_tells(content)
# Write back
print(f"Writing cleaned content to {form_path}...")
with open(form_path, 'w') as f:
f.write(cleaned)
print("\n" + "="*70)
print("VERIFICATION")
print("="*70)
# Check for remaining tells
tell_check_patterns = [
'reference material',
'methodology documentation',
'documents I have',
'handbooks',
'reference.*available',
'Census methodology',
'provided reference',
'my available'
]
print("\nChecking for remaining tells...")
found_tells = []
for pattern in tell_check_patterns:
matches = list(re.finditer(pattern, cleaned, re.IGNORECASE))
if matches:
found_tells.append((pattern, len(matches)))
for match in matches[:2]: # Show first 2
context_start = max(0, match.start() - 50)
context_end = min(len(cleaned), match.end() + 50)
context = cleaned[context_start:context_end]
print(f" ⚠️ Found '{pattern}': ...{context}...")
if not found_tells:
print(" ✅ No tells found!")
else:
print(f"\n ❌ Found {len(found_tells)} types of tells still present")
for pattern, count in found_tells:
print(f" - {pattern}: {count} matches")
print("\n" + "="*70)
print("COMPLETE")
print("="*70)
print("\nNext steps:")
print("1. Manually read Q1, Q5, Q10, Q15, Q20 to verify no tells remain")
print("2. Verify responses are still coherent and substantive")
print("3. If any tells remain, manually edit the file to remove them")
if __name__ == '__main__':
main()
