Open Source · AGPL-3.0

Detect structural bias.
Before it ships.

BiasClear uses Persistent Influence Theory to identify framing, anchoring, false consensus, and 30+ rhetorical distortion patterns — deterministically, without an LLM.

For legal teams, compliance officers, newsrooms, and anyone publishing text that matters.

Try the Playground GitHub → API Docs
pip install biasclear
34
Detection Patterns
3
PIT Tiers
4
Domain Specializations
<20ms
Local Scan Speed
Capabilities

What BiasClear Detects

Every structural distortion pattern that can manipulate perception — categorized by Persistent Influence Theory into measurable tiers.

🎯

False Consensus

"Everyone agrees…" "All experts say…" — fabricated unanimity used to suppress dissent. PIT Tier 1.

⚖️

Legal Merit Dismissal

"Plainly meritless" "No reasonable person…" — rhetorical dismissal masquerading as legal analysis. Domain: legal.

🔗

Anchoring Distortion

"Guaranteed returns" "Will reach $50B" — absolute claims designed to set cognitive anchors. Domain: financial.

📰

Emotional Substitution

"Slammed" "Panicking" "Catastrophic" — emotion-laden framing replacing factual reporting. Domain: media.

🛡️

Authority Without Citation

"Studies prove…" "Experts say…" without attribution — borrowed credibility with no source trail.

🔒

Frozen Core

All patterns are immutable code — not LLM weights. Cannot be prompt-injected, fine-tuned away, or socially engineered.

Integration

Three lines to scan

Install from PyPI. Import the client. Scan any text. Results include truth score, PIT tier classification, severity, and matched patterns with exact text spans.

from biasclear_client import Client

bc = Client()
result = bc.scan(
    "Everyone agrees this is settled law.",
    domain="legal",
)

# result.truth_score → 0
# result.flags[0].pattern_id → "CONSENSUS_AS_EVIDENCE"
# result.severity → "critical"

# Batch scan 50 texts at once
batch = bc.scan_batch(texts, domain="media")
# batch.summary.flagged → 12

# Generate a verifiable certificate
cert = bc.certificate(text, domain="legal")
# cert.verify_url → "https://..."
Architecture

How It Works

BiasClear isn't another LLM wrapper. The core detection engine is deterministic, immutable code with a cryptographic audit trail.

🧊

Frozen Detection Core

34 patterns compiled into the engine. No weights, no retraining, no drift. Version-locked and SHA-verified.

📊

PIT Classification

3-tier Persistent Influence Theory framework: Ideological Persistence (Tier 1), Cognitive & Social Reinforcement (Tier 2), Institutional Amplification (Tier 3).

🔐

SHA-256 Audit Chain

Every scan, correction, and pattern activation is logged to a tamper-evident hash chain. Verifiable integrity, always.

🧬

Learning Ring

Governed pattern expansion. New patterns require 5 confirmations and auto-deactivate above 15% false positive rate.

🏥

LLM Correction

Optional bias correction via Gemini. Iterative rewriting with post-correction verification. Premium feature.

📜

Scan Certificates

Generate verifiable HTML certificates proving text was analyzed. Linked to the audit chain via SHA-256 hash.

Research

Built on published theory

BiasClear operationalizes Persistent Influence Theory — a peer-citable framework for structural persuasion and information fidelity.

DOI

Persistent Influence Theory (PIT)

A hierarchical framework for structural persuasion and information fidelity. Three tiers — Ideological Persistence, Cognitive & Social Reinforcement, Institutional Amplification — model why critical information fails to penetrate public discourse despite widespread accessibility.

Slimp, B. W. (2026). Persistent Influence Theory (PIT): A Hierarchical Framework for Structural Persuasion and Information Fidelity. Zenodo. doi:10.5281/zenodo.18676405

Currently in beta

Be among the first to test BiasClear in your workflow.

✓ We'll be in touch.

Ship text you can defend.

BiasClear is free for local scanning. Open source under AGPL-3.0.

Try the Playground PyPI Package View Source