NeutralEye

How NeutralEye
Works

Article text goes in. A structured bias analysis — direction, evidence, confidence, and next reads — comes out.

01Submit

Paste article text or provide a URL. NeutralEye accepts both — URL submissions extract the article body automatically.

02Validate

The system confirms the material is article-like. Short passages, navigation text, and blocked pages are flagged before analysis starts.

03Analyze

Tone, framing, attribution, source balance, and omission are reviewed together in one pass — not as five independent checks.

04Score

A confidence score is calibrated against how consistently signals appeared. Articles where all five checks point the same way score higher than those with mixed or thin evidence.

05Return

The result comes back with a direction label, confidence score, quoted evidence, and sources to read alongside — everything needed to check the reasoning yourself.

One pipeline

Article intake, content extraction, signal review, and evidence packaging run as a single sequence — not independent steps stitched together.

Structured output

Every result uses the same schema: direction, confidence, summary, examples, sources, and recommendations — making results comparable across runs.

No black box

The system returns what it found and where. Quoted language and sourcing patterns are included so the output can be checked against the original text.

Processing

One pass, five signals

Every article goes through tone, framing, attribution, source balance, and omission in a single sequence — not as independent checks stitched together. The output stays tied to what was actually in the text.

System flow
Readable article material confirmed
Signal families checked together

Framing, source balance, attribution, and omission reviewed in one pass.

Structured result assembled

Summary, examples, confidence, and next-reading context stay attached.

Output

Evidence exits with the result

Every result includes the direction label, a confidence score, quoted examples of the signals that shaped it, and sources to read alongside — so the analysis is a starting point, not a final word.

Structured result

Readable explanation with evidence attached.

Inspectable
+Summary grounded in article text1
+Quoted evidence linked2
+Next-read suggestions included3
Evidence packet
1 Quoted phrase with explanation
2 Source balance note
Confidence

A score, not just a label

Every result includes a confidence score reflecting how consistently the detected signals appeared across the article. When evidence is sparse or ambiguous, the score drops — so you know when to read the analysis with more caution.

Confidence report
DirectionModerate bias — toward government sources
Analysis confidence0.74
Signal strengths
Framing
High
Language
Moderate
Attribution
Low

Put the pipeline to work

Open Analyzer