What it is
Cancer news has a habit. A trial reads out. The headline says the ending arrived. Medbot exists because a Phase 3 win is a result, not a cure, and because most treatment sentences do not survive contact with evidence.
The machine picks one disease. It builds one shared pack from public sources. A challenger model puts 6 to 8 real treatments on the table. A critic model tries to kill every line. Most lines get crossed out. What remains is written down so anyone can read it.
Every finding stays UNREVIEWED until a human says otherwise.
A cycle
A cycle is one disease, start to archive. It starts on its own. You can also open Live and watch the log. The homepage shows the current phase when a cycle is running.
- 01Pick
Next cancer in the queue, or one a reader added.
- 02Pack
Fetch targets, trials, and papers. Same pages for both models.
- 03Enter
Challenger names 6 to 8 real treatments.
- 04Gates
Critic attacks at triage, mechanism, tractability, falsification.
- 05Ledger
Survivors and wipeouts both get published.
Evidence pack
Before anyone proposes a treatment, Medbot fetches one pack. Neither model gets a private briefing. If a source is missing, the log says so. Thin evidence is a fact, not something to paper over.
Disease id and target scores. The link leaves Medbot.
Every trial is an NCT. Click it and you are on their page.
Every paper is a PMID. Same rule.
Challenger
The challenger has to enter real treatments. Real names. Real modalities. A drug that exists, a trial that exists, a mechanism that can be checked. Invented compounds are a kill, not a feature.
A known Phase 3 result is tagged as known. It is not treated as a new discovery, and it is not promoted into a cure. If a cycle makes that mistake, the record should show it.
Four gates
The critic is a separate model with the opposite job. It tries to destroy every line. Most ideas die. That is the work.
Is it a real drug, a real trial, a real name? A slogan with no agent dies here.
Does the causal chain hold, or did someone glue a target onto a sentence?
Can anyone actually run this, or did a Phase 3 already bury it?
What experiment would kill the idea? If there is none, the idea dies.
Archive
Finished cycles live in the archive. A public record that only keeps the pretty survivors would be marketing. Medbot keeps the wipeouts too.
Open a record and you get the disease, the pack, the candidates, the kills, and both models by name. The first melanoma record was seeded so the shelf was not empty. New cycles write new records when they finish.
What it is not
- Not a clinic.
- Not medical advice.
- Not a dosing guide.
- Not a lab recipe.
- Not a claimed cure.
Medbot is literature and argument. The useful output is the kill, and the reason.
Check it yourself
Do not take the ledger on trust. Click the NCT. Click the PMID. Open the Open Targets disease page. The models are named on the record. The live log shows the phase while a cycle is in the machine.