Both figures are right for what they are; they are produced at different levels of detail. The league scan is a fast comparative pass across every fixture in the competition, using consistent baseline assumptions, built for shortlisting and ranking. The single-fixture analysis rebuilds the calculation for that one match with more granular adjustments and produces a more precise, match-specific EV. When the two disagree, the single-fixture figure is the decision metric.
The numbers move for two reasons. Granularity: the scan is optimised for breadth, the single-fixture run for depth, and the deeper calculation can land somewhere different, sometimes by a wide margin. Timing: the two views are not computed at the same moment, and Gecko Edge works off live data, so odds drift and model updates between the two readings shift the EV in either direction.
The working routine: scan the league to shortlist, open the strongest few fixtures individually, and read the single-fixture EV, Sanity, and flags before deciding anything. If the single-fixture EV stays positive, consider it. If it flips negative, discard it. A match that looked strongly +EV in the scan and comes out negative in isolation is the deeper calculation doing its job; the first result was not wrong, the second is more refined.