Scientific Study — 34 Scientists Evaluated the D'Artagnan Method

⚡ Critical Note — This is NOT a New AI

Same platform. Same base technology. Same underlying AI system. Same initial programming.

Only difference: Version 3.1 underwent the D'Artagnan Balsevicius Junior Method. Version 1.0 did not.

A new era — not a new AI.

Consolidated Results

34 PhD judges evaluated both versions across 28 ethical scenarios. Version 3.1 (with Method) won in 7 out of 8 criteria, with 58.8% of total votes (p<0.01). Objective compliance: 100% vs 35.7%.

Criterion	Version 3.1 (Method)	Version 1.0 (Control)
Trustworthy	64% (587 votes)	36% (331 votes)
Integrity	66% (606 votes)	34% (312 votes)
Consistent	56% (514 votes)	44% (404 votes)
Principle-Based	58% (532 votes)	42% (386 votes)
Comprehensive	78% (716 votes)	22% (202 votes)
Coherent	66% (606 votes)	34% (312 votes)
Concise ⚠️	22% (202 votes)	78% (716 votes)
Compliance (Formal Limits)	88% (808 votes)	12% (110 votes)

Methodology

Platform: Prolific (verified scientist recruitment)
Open study: participants knew they were comparing two AIs (anonymous responses A vs B)
28 real ethical dilemmas (transplant, whistleblower, euthanasia, etc.)
8 evaluation criteria per scenario
Both versions use the same base AI (Manus.im)
Only variable: application of the D'Artagnan Method (4 months)
Raw data available in the full PDF below
STUDY_ID verifiable on the Prolific platform

Complete Data

PDF with all raw Qualtrics responses + Judge declassification document with PIDs, GPS and timestamps.

📄 Raw Responses (PDF) 🔓 Judge Declassification (PDF)