bws82/biasclear
Structural bias detection and correction engine built on Persistent Influence Theory (PIT)
What's novel
Structural bias detection and correction engine built on Persistent Influence Theory (PIT)
Code Analysis
5 files read · 7 roundsA structural bias detection engine that uses deterministic regex-based patterns to identify rhetorical manipulation (consensus-as-evidence, false urgency, dissent dismissal) in text, with optional LLM-enhanced deep analysis and iterative bias correction.
Strengths
Genuinely substantive implementation with a well-designed frozen-core architecture that separates deterministic detection from LLM enhancement, sophisticated regex patterns that encode linguistic relationships rather than just keywords, and thoughtful engineering details like ReDoS protection, self-scan of LLM output, and hash-chained audit trails.
Weaknesses
The regex-based detection, while clever, is inherently limited in coverage and may produce false positives/negatives on nuanced text; the theoretical framework (PIT) is self-published without peer review, and the LLM-dependent features (deep scan, correction) introduce non-determinism that partially undermines the 'deterministic core' selling point.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
84
Contributors
2
Files
62
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
AGPL-3.0