I bought "half price" dishwasher tablets at the supermarket. $64 down to $32. Felt clever. Then I read the back. Mostly cheap surfactants and bleach. On the shelf below sat a plain paper box I'd walked past for years: same price, more tablets, cleaning with citric acid instead. The boring one was better. I'd been paying for the packaging my whole adult life. I'm a chef, not a developer by trade, so I spent six months of nights building HonestWorld. Scan a product and it scores the label's honesty from 0 to 100. Not nutrition. Honesty. The gap between the front of the pack and the back. How it works, since this crowd will ask: a vision model reads the packaging into structured text, ingredients, claims, percentages. Then a rule engine runs 23 rules I wrote against that text. Things like a hero ingredient named in the title but sitting at 1%, one substance split across four names so it drops down the list, "made with real fruit" where fruit is ninth. The scoring is deterministic, so the same label always produces the same score, and every deduction hands back the exact string that triggered it. You can argue with it, which is the point. It's free, there are no ads, and I take no money from brands. The moment I do, every score is worthless. The thing I'd genuinely like your take on: where does a rule engine like this produce false positives? Water in tofu isn't dilution. Mustard is meant to be mostly vinegar. Those cases were harder to get right than the obvious deceptions, and I suspect there are more I haven't found.

HonestWorld

Scan a food label. See the marketing tricks it's hiding.

Scan a food label. See the marketing tricks it's hiding.