🎪 Hall of Amusing Red Flags

4 founders x 25 years each? LinkedIn reveals: 35 years total. In AI specifically: 4. Apparently time moves differently in startup land.

FINRA shows no registration. LinkedIn reveals "Technology Project Manager" overseeing CRM upgrades. By this logic, the night janitor at Blackrock also "worked with trillions".

Google Scholar results: 0. Academic papers: 0. Peer reviews: 0. Invented terminology that sounds impressive precisely because no one can evaluate it.

"Which consortium?" -> "Confidential." "Can we speak to them?" -> "They prefer not to be contacted." The only thing verifiable is that someone typed "73%."

This sentence could describe any software. Or a toaster. Or nothing at all. Pro tip: If you can replace "AI" with "blockchain" and it still works, it says nothing.

Translation: "Limit demos" = no working demo. "NDA required" = legal cover for nothing. Real platforms: "Here's a sandbox, try it yourself."

That's not how trademarks work. One ™ means unregistered claim. Two means nobody proofread... or nobody cares. When even the brand name has a copy-paste error, imagine the codebase.

Five stats, all suspiciously round, zero citations, zero methodology, zero named customers. If your metrics end in 0 and have no source, they came from the marketing meeting, not the data team.

No name. No specific company. No LinkedIn. No way to verify any of it. Real testimonials have real humans at real companies that you can actually look up.

The demo proves itself by... talking about itself. That's not proof, that's a chatbot reading its own marketing page. Asking an AI "are you good?" is like asking a used car if it runs well.

Eight industries claimed. Zero named organisations. Zero logos. Zero case studies. Just a grid of icons. 'Trusted' by whom? By the designer who made the icons, apparently.

App store links on the website but... have you actually searched for it? Is it there? Does it have reviews? Pro tip: Always check the store listing independently. Linking isn't shipping.

Enterprise AI platform with security, analytics, multi-language support, role-based permissions... live in 2 minutes? If setup takes 2 minutes, support will take 2 years.

No SOC 2 mentioned. No ISO 27001. No penetration test results. No audit reports. Just adjectives. 'Enterprise-grade' without a certification number is just a font size.

A measurement that equals an input is not evidence; it is a tautology wearing a lab coat. It usually means the pipeline never observed the failure case at all: losing trades exit at the stop by definition, uncertain answers are filtered by the threshold before scoring, and slow requests are killed before they can be timed. The result looks like performance. It is arithmetic. This is dangerous precisely because it produces beautiful numbers rather than absurd ones. Nobody audits a plausible result. The tighter you set the parameter, the better the metric gets, which reads as a tuning discovery instead of what it is. Put the reported metrics next to the configuration file and look for equality. Any headline number that exactly equals a threshold, cap, stop, or timeout is guilty until proven otherwise. Then break the assumption and rerun. We found ours by asking whether the exit price was ever actually reached rather than assumed. It was not. The system had been booking the stop price on gaps where the market never traded there. Our reported average loss of −0.010% was the stop distance restated. Live execution, with real fills, disagreed immediately. The general test: force the system to observe the case it currently defines away. If the metric survives, it was a measurement. If it collapses, it was a parameter.