…a stranger offered a kidney?
An has no patient of their own, so they can start a chain.
1 of 5
–
–
Something went wrong.
Synthetic patients only. Not medical advice.
Kidney swaps
you can check.
Some patients have a willing donor who doesn't match them. When a few of these pairs swap donors, everyone gets a kidney that fits. Swap finds the best swaps, then shows exactly why, for every patient, matched or not.
What makes Swap different
-
It explains the "no".
Pick any patient who missed out. Swap shows which donors could have helped, and what got in the way.
-
It never guesses.
Every answer carries its evidence. If the graph has nothing, it says
the graph doesn't know
, never "safe". The AI helper can only choose from fixed, human-written queries. -
It proposes. People decide.
Swap drafts the plan and records it in the graph. Only a person can approve it, and the AI helper has no way to.
-
It checks itself.
Every number is recounted by separate code that never touches the database. If they ever disagree, you see it.
Why a graph? A swap is a loop of "donor can give to patient" links. Finding loops, and asking "who could give to this patient?", are two things a does naturally, from the same stored links.
The idea in five pictures
Results
2-way swaps only
–
3-way swaps allowed
–
More patients helped
–
Looking for swaps…
Loading the current pool…
Change the pool
A is a batch of made-up pairs. Build a new one and everything on this page updates.
–
Every swap, as a list
Each line is one closed loop: every pair's donor gives to the next pair's patient. Tap one to find it on the map.
Who is left out, and why?
Not everyone can be matched. Pick a patient and the graph shows who could have helped, and what stood in the way.
Patients without a swap –
Loading…
What if…?
Each idea runs on a throw-away copy of the pool, so nothing you see here is lost.
…one pair dropped out?
Pairs sometimes leave. See who else is affected.
…we favoured the hardest cases?
By default Swap maximises how many people are matched. With priority on, a patient counts three times.
How was this measured?
From result to action
Swap can turn the best set into a plan for a coordinator, and record it in the graph. It only ever drafts. A person has to approve it, and the AI helper cannot.
Uses the swap length and priority chosen on the other steps.
Plans recorded in the graph
When a lab test fails
Most planned exchange transplants never happen, and a failed , the lab test just before surgery, is a common reason. Record one here and watch the graph react: every swap that needed that link dies, the approved swap breaks, its patients are stranded, and the graph looks for what can still be saved.
Approve a plan first. A failed test matters when it breaks a swap someone already approved.
1. Who is recording the result, and which link did the lab test?
Lab results recorded in the graph
Can you trust it?
Don't take our word for it. Every answer comes with its working, and you can open any line below.
Made-up pairs→ graph finds every ring→ best set is chosen→ answer + evidence + check
Is it right? an independent recount
A second count is made in plain Python from the same pool. If the graph database ever disagreed, you'd see a cross here instead of a tick.
Ask in plain English the agent
The agent may only choose from fixed, human-written queries. The answer shows whether a language model or the keyword router replied.
Try to trick it a pair that doesn't exist
Swap never guesses. Ask about a pair that isn't in the pool and it says so, which is different from "not matched".
What was asked of the database? the trace
Every request: which tool, what it was given, how long it took, how many rows came back.
The four kinds of answer
The queries themselves Cypher
Written by a person, with blanks filled in safely. The AI never writes these ().