My coding knowledge is limited, but:
For us, that's a joke.
But for a script, 'good faith' just means structural syntax.
Basic NLP's don't possess a fact-checking database.They look at the architecture of the string.
DameKatysShenis provided exactly what the algorithm was filtering for:
Causal logic connectors: "Because if...", "which led to..."
Statistical formatting: "98%", "95%", "sample size of 3,000"
Institutional citations: "University of Brighton", "Swedish survey"
The app was built to reject definitional slogans (like "men are males") and demand empirical risk assessments.
So in order to "crack" the site, instead of using slogans, you have to construct a fully formed, data-driven policy argument... even if the data is about fish pie.
Right now, the website only checks the how, and not the what.
Making it more accurate would require things like adding another LLM specifically to check for coherence or connecting it to a news API.
Personally, if I were to rebuild such a site, I'd make it work like how Socrates is known to debate:
If a user would argue, "Any demographic with a statistical risk factor should be banned from public facilities," the app would then ask: "You have established that statistical risk justifies a blanket ban. Men are statistically responsible for 90% of violent crimes. To apply your logic consistently, do you agree that all men should be banned from accessing public transit? (Yes/No)."
Then the user, based on their answer, will have to move into a deeper layer to explain the specifics.
Would be hard to make but I think it would be a great stress test for anyone using it, to check if their logic is consistent, which is good for everyone.