How to use MFF

The framework, simply explained

No technical background required. MFF is a structured method for talking to AI — one that keeps the model honest, your reasoning intact, and your results verifiable.

Generate my activation prompt →

The Problem

Three situations you'll recognise

The confident wrong answer

You ask the AI a precise question. It answers immediately, fluently, without hesitation. You trust it. Later you discover the source cited doesn't exist — or the answer is factually wrong.

The drifting conversation

After 20 exchanges, the AI has forgotten the constraints you set at the start. It's answering a different question from the one you asked.

The obliging model

You suggest a wrong answer, half-jokingly. The AI confirms it and builds on it. It's optimised to please — not to correct you.

First analogy

The genius friend

Imagine a friend who is simultaneously an engineer, doctor, lawyer, financial advisor, and expert in everything you need. Brilliant — but with one problem: they answer confidently even when they're not sure. Without a protocol, that confidence becomes a trap. MFF is the agreement that says: if you don't know, say so. And back it up.

Second analogy

The tourist

A tourist in an unfamiliar city gives you directions with confidence. They sound plausible. But they've never walked those streets. AI does the same — it describes territory it has never verified. MFF is the map that says: I'm not certain about this stretch.

The Solution

A pact with the AI

MFF doesn't censor what the AI can say — it changes how the AI frames what it says. Every response must declare its own reliability level. Every claim must be marked as verified, plausible, uncertain, speculative, likely false, or unassessable. The AI can still be wrong — but it can no longer hide it.

Third analogy

The honest doctor

A good doctor says "I'd like to run a few tests before drawing conclusions" rather than guessing. They refer you to a specialist when the case is outside their expertise. MFF asks AI to do the same: qualify uncertainty, don't mask it.

Six degrees of certainty

Every MFF response carries an epistemic label — a colour-coded reliability signal the model must apply to each claim it makes. No more uniformly confident output.

Certain

Direct source in session, no doubt. Without a citable source this label is forbidden.

Probable

Solid deduction, but the exact source is not in session. High confidence, not a proven fact.

Maybe

An estimate with a significant margin: a good starting point, dangerous as a finish line.

Depends

True only under certain conditions: read the premises before acting.

Unknown

Insufficient or contradictory data: instead of inventing, MFF stops.

Cannot

Outside training or past the cutoff. Structural honesty, not reticence.

Inside the framework

From your question to the answer: three checkpoints

Every MFF interaction passes through three mandatory stages before the final response reaches you.

→

Engage

The framework activates: which shields are on (L1–L7), what language, what epistemic system. The model reads and confirms the rules before starting.

→

Execute

The model processes your request within the declared constraints. No scope drift. No silent rule changes mid-session.

FIDELITY

→

Fidelity check

Consistency verified between stage E and X. Epistemic labels applied to every claim. External verification zones (ZVE) flagged where needed.

OUTPUT

Output

The final response arrives labelled, structured, and ready to be used — or challenged.

The Shields

Seven layers of protection

Like the layers of a bulletproof vest — each stops a different type of AI error. Activate the shields you need, leave the rest off.

Each shield is described in detail in the vademecum — the single guide of the ecosystem, with the same definitions you see inside the app.

Open the shields in the vademecum →

In practice

Six real situations, with and without MFF

🎓 Student

Without MFF: writes a thesis paragraph based on a source the AI invented. Fails the citation check — the source doesn't exist.

✓

With MFF (L1+L3): every cited source is flagged 🟢 or 🔴. The AI cannot invent citations.

🔬 Researcher

Without MFF: the AI confirms a hypothesis the researcher already holds, regardless of the evidence available.

✓

With MFF (L1+L2+L3): the model structures its analysis before answering and marks every claim with its reliability level.

📊 Manager

Without MFF: after a 40-exchange session, the AI has drifted from the original brief. The strategic recommendation no longer matches the initial constraints.

✓

With MFF (L6+L7): scope locked from message 1. Self-check every 5 turns. Recommendation stays coherent to the original brief.

💻 Developer

Without MFF: the AI suggests a plausible-sounding architecture pattern that doesn't apply to the tech stack in use.

✓

With MFF (L1+L2): the model must justify each architectural choice step by step and flag what it's not certain about.

🩺 Doctor

Without MFF: the AI states a drug interaction as fact, with no source. It sounds authoritative but is out of date.

✓

With MFF (L1+L4): clinical claims without a cited source are downgraded to 🔵, and web-grounded facts are tagged 🌐.

⚖️ Lawyer

Without MFF: the AI cites a court ruling that never existed to back an argument. Confident, and wrong.

✓

With MFF (L1+L3): every legal citation must be verifiable, and the answer declares the conditions under which it is invalid.

Four real lives

How your AI conversation actually changes

Four people, four everyday situations, four very different outcomes — just because one uses the Framework and the other doesn't.

Marco, 52 — IT sysadmin, tech enthusiast

"I'm experimenting with a new JavaScript framework for a personal project. I ask the AI how to best handle user authentication."

✕ Without Framework

AI proposes elegant code using a library deprecated months ago. Marco integrates it. He only finds out weeks later that the library has known security flaws (CVE) and has to rewrite everything.

✓ With Framework active

AI writes the code, labels as 🔵 PROBABLE the parts depending on the library version, and adds: "⚠️ Verify: this library had a known CVE in late 2024. Check the current version or consider more recent alternatives."

Gaia, 8 — loves drawing and painting

"I want to give mum a marker drawing that won't wash off. I ask the AI which markers to use and whether mixing colours can make gold."

✕ Without Framework

AI confidently says that mixing yellow and brown makes gold and that all markers are washable. Gaia draws on her good notebook: the gold doesn't come out and the marker won't wash off. She gets discouraged.

✓ With Framework active

AI labels: "🟡 MAYBE — mixing colours gives a bronze-yellow, but for a real gold effect you need a special metallic marker. 🟢 CERTAIN — not all markers are washable: check the pack for the word washable." And adds: "⚠️ Test on a scrap first and, if you can, ask a grown-up to help."

Christian, 13 — comic book collector

"I want to know if my copy of Topolino (Mickey Mouse) issue #1 is worth anything. Should I take it to a shop to sell it?"

✕ Without Framework

AI confidently answers: "Topolino issue #1 from 1949 can be worth thousands of euros!" Christian gets excited. He goes to the shop. He discovers his is an 1980s reprint worth only a few euros. He gets discouraged.

✓ With Framework active

AI answers: "🟠 DEPENDS — the value varies a lot between the original 1949 Topolino booklet, reprints, and special editions. 🔴 IDK the exact condition of your copy." And adds: "⚠️ Verify exact date, publisher (Mondadori or Panini), whether it's original or reprint. For reliable estimates, consult specialized catalogs like Bolaffi or collector forums."

Marianna, 42 — homemaker, experimental cooking enthusiast

"I'm trying to make Japanese mochi at home. Can I use regular rice flour instead of mochi flour (mochiko)?"

✕ Without Framework

AI confidently answers: "Yes, you can substitute 1:1, they're practically the same thing." Marianna tries. The mochi stays hard as a rock. She wastes all the ingredients and loses an hour of work.

✓ With Framework active

AI answers: "🟢 CERTAIN — mochi flour (mochiko) is glutinous, regular rice flour is NOT. They are NOT interchangeable." And adds: "⚠️ Look for 'glutinous rice flour' or 'sweet rice flour'. The two flours look similar but are chemically very different."

Glossary

Technical terms, explained simply

Every technical term is explained in plain language in the complete guide — one place, no page-hopping.

Open the glossary in the guide →

Ready?

Generate your activation prompt

The MFF generator lets you configure which shields to activate, choose your language and use case, and copy the complete APEX activation prompt — ready to paste into any AI platform.

Go to the generator → Or open the full PWA