Medbot doesn't diagnose anyone and it doesn't hand out a cure. It picks one cancer, builds a single evidence pack from public research, and runs a live argument between two AI models. Whatever survives the argument is still unreviewed, not a recommendation.
No hidden reasoning. Every step, every kill, every survivor gets written down.
One shared research pack per disease, pulled from Open Targets, ClinicalTrials.gov and PubMed. Both models read the same pages.
One model has to name real treatments, six to eight of them, real names and real mechanisms. Invented compounds don't survive step one.
The second model tries to eliminate every idea it can, checking each one against four separate standards of scrutiny.
Survivors and eliminations both get archived, with links back to the original public sources for every claim.
Is this an actual drug, an actual trial, an actual name. Invented compounds die immediately.
Does the reasoning actually hold up, or is it just a target name attached to a sentence.
Could this realistically be run today, or did a later stage trial already close the door on it.
Is there even an experiment that could prove this wrong. If not, the idea does not survive.
This is a sample layout, not a live pull. A real archived cycle updates as the argument runs.
Disease identifiers and target scores. The link goes to their page, not ours.
Every trial referenced is a real NCT number, clickable straight to the source.
Every paper referenced is a real PMID, same rule, same transparency.
This is not a clinic. Medbot is literature review and structured argument, nothing here is medical advice, there is no dosing information and no claimed cure. Every surviving idea stays unreviewed until an actual clinician says otherwise.