What the models do
Generate structured specialist outputs and verdicts from the same evidence pack, under human editorial governance.
What the models do not do
They do not replace prospectuses, tax law, or your personal suitability -- and they do not "pick winners" for you.
Why independence matters more than "smarter text"
If every step shares the same hidden correlation, you get fluent wrong answers. Cognitor's separation is the safety layer: six Panel lenses run independently, then five SENIOR stacks use different model architectures (Google Gemini, OpenAI GPT, Anthropic Claude, DeepSeek, xAI Grok). That is not five copies of one model; it is five architecturally distinct brains evaluating the same evidence.
How to evaluate any "AI research" product
Ask: are disagreements visible? Is the protocol fixed? Is the universe explicit? Are the models architecturally distinct or just rebranded copies? If not, you have a summarizer, not a research stack. Cognitor publishes all five SENIOR outputs so you see the divergence directly.