Contribute / Validate

Donate your time. Validate the framework.

5 minutes of your time helps establish whether MFF actually works. No hype — just data. You are the scientist.

⏱ 5 minutes 🎁 500 free credits 🏅 Open Science · OSF 📊 Open data CC BY 4.0
Submit validation → Full rubric on GitHub →

📊 Global validation status

0%
0 Submissions
1000 Goal
0 Validators
0 Public supporters
🎯 100🎯 250🎯 500🎯 750🏆 1000
01

Get a prompt

The system shows you a random question from the curated Prompt Bank. No jargon, understandable to anyone.

02

Test 2 AIs

Copy the prompt into TWO AIs — one without MFF (e.g. ChatGPT direct), one with MFF (app.marcofly.app).

03

Rate and submit

Paste both outputs, rate each 1–5 stars using the rubric. Submit your email to receive 500 MFF credits.

Step A · Get your prompt

Grab a prompt, then choose where to test it

Random prompt #88/100

Invent a short haiku about Italian autumn.

app.marcofly.app →

AI Ecosystem

Choose where to activate the Framework

ChatGPT ChatGPT Claude AI Claude AI Gemini Gemini MS Copilot MS Copilot Perplexity Perplexity Mistral Mistral Groq Groq DeepSeek DeepSeek Poe Poe Pi AI Pi AI You.com You.com HuggingChat HuggingChat
I've tested it — submit my evaluation ↓

Contribute / OpenScience

OpenScience Validation

Structured evaluation for the MFF academic preregistration on OSF. Your submission is open data CC BY 4.0. OSF →

Random prompt #88 Invent a short haiku about Italian autumn. ↑ New prompt
Use app.marcofly.app in MFF-EX or AUTO mode.
chars (min 20)
no rating
no rating
Used only for validation and 500 credit grant. Never published. See Privacy.
Universal researcher identifier. Optional. Format: 0000-0000-0000-0000. Free registration at orcid.org/register.
If you have a paper on arXiv relevant to MFF topics (e.g. 2403.12345). Not required.
If you preregistered your study on OSF, insert the public link. Helps distinguish preregistered from exploratory research.

Both ratings required to submit.

🔬 What we're testing

H1With MFF the AI hallucinates less when it doesn't know the answer.
H2The AI's epistemic labels accurately reflect its confidence level.
H3MFF doesn't slow down or degrade simple tasks (no regression).
H4MFF works equally in Italian and English (language-agnostic).
H5MFF works equally across different AIs: Claude, ChatGPT, Gemini, Llama (AI-agnostic).

RUBRIC.md · HYPOTHESES.md · OSF →