Wayne Kirkman here. I wrote "The AI Subscription Tax" a few weeks back about paying $200/month to beta test unreliable AI. The response was: okay, so what's the fix?
This is the fix. I built ASCERTAIN — a governance layer with seven validation gates, mandatory citation requirements, and bias detection. Every response passes through FORGEGATE before delivery. No unsourced assertions allowed.
The unicorn test case in the article shows it working: the AI tried to assert "unicorns are legendary creatures" without citation. The SOURCE gate caught it, penalized the quality score, and forced the model to provide Wikipedia and Science.org sources instead.
If one developer can build this as a side project, the $200/month vendors have no excuse.
Happy to answer questions about the architecture or the Five Pillars framework.
Wayne Kirkman here. I wrote "The AI Subscription Tax" a few weeks back about paying $200/month to beta test unreliable AI. The response was: okay, so what's the fix? This is the fix. I built ASCERTAIN — a governance layer with seven validation gates, mandatory citation requirements, and bias detection. Every response passes through FORGEGATE before delivery. No unsourced assertions allowed. The unicorn test case in the article shows it working: the AI tried to assert "unicorns are legendary creatures" without citation. The SOURCE gate caught it, penalized the quality score, and forced the model to provide Wikipedia and Science.org sources instead. If one developer can build this as a side project, the $200/month vendors have no excuse. Happy to answer questions about the architecture or the Five Pillars framework.