Rank the next wireless charging point for one EV decision event.

A deployment-valid Graph Transformer that compares candidate WCPs without candidate IDs, proxy-label criteria, or cross-event batch aggregation.

Event-specific graphListwise rankingBatch invariant

01 · Decision event

EV context

02 · Candidate generation

Journey context

Outside model scoring

Source and destination describe the journey. Candidate WCPs are selected by minimum additional detour from the remaining trip. Coordinates and detours never enter the strict model.

Only the current location code is a trained model input.

Journey origin

Current EV position

Destination

03 · Graph ranking

Recommendations

Awaiting event

No ranking yet

Enter one EV decision event to construct its candidate graph.

REVIEWER-FACING CONTROLS

What the strict model does—and does not use

✓

Independent event graph

The result for an EV is unchanged when unrelated events enter the batch.

✓

No identifier embeddings

Event and WCP identifiers map records and topology only; they are never learned inputs.

✓

No proxy criteria

Distance, speed, popularity, TOPSIS intermediates, EM outputs, and derived ranks are excluded.