Learn FALSEWORK: from zero to expert, then sell it

No AI background needed. Five levels, each with something to actually do, at your own pace. Tick items as you go (your ticks stay in this browser).

Evidence, proof and selling tools for FALSEWORK

Level 1 Beginner: understand it

AI coding tools are very good at producing code that looks finished. Sometimes it is not: a function returns a typed-in "98% accuracy", a security check can never fail, an error is quietly swallowed, a "score" is a random number. Human reviewers miss these because the code looks tidy.

FALSEWORK reads Python code (it never runs it) and grades each module REAL, PARTIAL, STUB or UNKNOWN, pointing at the exact lines. It also checks that each web route really calls the engine it claims to call. No AI makes the decision; it is rule-based, so the same code always gets the same grade.

Think of it like this: A home inspector who knocks on the walls. The house looks finished, but some walls are painted cardboard. FALSEWORK knocks on every wall and lists the hollow ones.

Read the full plain-English explanation: FALSEWORK, explained.

I read the explanation above.
I can say what it does in one sentence, without notes.

Level 2 User: see it work

Run it live: open the live demo. Before each result, guess what will happen. These are checks it must pass:

Sample: A signature check that never checks the signature: A deterministic detector, not the model, finds the defect the AI is then asked to fix.
Sample: A metric that reports a random number as accuracy: A deterministic detector, not the model, finds the defect the AI is then asked to fix.
Sample: A webhook signature compared with == (timing leak): A deterministic detector, not the model, finds the defect the AI is then asked to fix.
Sample: An alert function that says 'sent' and sends nothing: A deterministic detector, not the model, finds the defect the AI is then asked to fix.
Your code: honest constant-time compare: Real code must grade REAL or nobody trusts the alarms.
Your code: verify() that always says yes: A security check that cannot fail is fabrication.

I ran the live demo.
I can explain what one of these checks shows.

Level 3 Verifier: prove it yourself

This is the skill that sets you apart: you don't have to believe us, and neither do your customers.

Install Python 3 if you don't have it (python.org/downloads).
Download the kit and unzip it.
Open a terminal in that folder. On Windows, click the folder's address bar, type cmd and press Enter. On a Mac, right-click the folder and choose New Terminal at Folder.
Run python3 verify.py (on Windows, python verify.py). It checks the seal without the internet.
Run python3 verify.py --live. It repeats every check from your own computer.
Try to cheat it: open transcript.json in a text editor, change one character, save, and run python3 verify.py again. It must now say FAIL. Unzip a fresh copy afterward.

I ran the kit's check on my own computer.
I changed a file and saw it fail.

Level 4 Expert: know the edges

An expert knows what a product does not do, and says it first. That is what makes a buyer trust you.

Read its limits page in full: what FALSEWORK does not do.

You can say

It finds code that looks finished but fakes its results, and shows the exact lines.
It is deterministic: no AI judge.
On our benchmark it raised 0 false alarms at the blocking grade.
A person reviews every finding before the report goes out.

Never say

"It finds all fake code": on a fresh blind set it caught 3 of 12.
"REAL means the code is correct": it means none of the known patterns fired.
"Works for any language": Python only.

Read the questions real buyers send, with honest answers: answers to buyer questions.

Test yourself (click a question to see the answer)

Does FALSEWORK run the code?

No, it reads it.

What does a REAL grade mean?

None of the known fabrication patterns fired; not a guarantee.

Standard audit price and turnaround?

$1,500, 48 hours.

Blind-set result to be honest about?

3 of 12 on a fresh blind set: precise but narrow.

In the live check "Sample: A signature check that never checks the signature", what answer must come back, and why?

"PARTIAL", "STUB". A deterministic detector, not the model, finds the defect the AI is then asked to fix.

In the live check "Sample: A metric that reports a random number as accuracy", what answer must come back, and why?

"PARTIAL", "STUB". A deterministic detector, not the model, finds the defect the AI is then asked to fix.

In the live check "Sample: A webhook signature compared with == (timing leak)", what answer must come back, and why?

"PARTIAL", "STUB". A deterministic detector, not the model, finds the defect the AI is then asked to fix.

Does a passing kit prove the product is safe for every possible input?

No. It proves what the live product answered to the inputs listed in the kit, and lets anyone check that again.

Has an independent party audited or certified this product?

No. No third-party audit is complete and no certification is held. Say so if asked.

I got every quiz question right without looking.
I can name one thing it does not do.
I explained it to a friend, and they understood it.

Want a formal study path? The CAIN Academy has courses and exams. Finishing this page is your own practice; it is not a qualification or certification.

Level 5 Seller: put it in front of a customer

Copy, paste, send: the shortest route, no technical skill needed.
Who buys it and where to find them.
The sales conversation, then the offer and the close.
The rules: read these before your first message.
Selling tools with this product's evidence built in: pitch, emails, call script, objection answers, one-pager.

I read the rules.
I sent my first personal message.

About money: your share is agreed in writing with FIRME AI CORPORATION before your first message (email vincent@firmeai.com). Nothing here promises income. Most messages get no reply; that is normal.