Everton and Manchester United served up a topsy-turvy affair at Goodison Park that swung decisively in the final moments. United struck first after the break through Mbeumo, assisted by Mainoo, but Everton clawed back when George levelled it up in the 83rd minute with help from Hackney. United looked poised to nick it when Šeško restored their lead courtesy of Luke Shaw's assist, only for Maitland-Niles to haul Everton level again deep into stoppage time. The match finished 2-2—a draw neither side quite managed to hold onto.

Our model went into the game fancying a 1-2 United win, assigning a 29% chance to the draw that actually materialised. That was a lean among plausible outcomes rather than a confident call; the model had flagged both sides as motivated enough for a proper contest despite mid-table stakes and mixed recent form. The draw was always live in the probability picture, but we weighted United's superior pedigree and head-to-head record heavily. The extra goals and the sheer drama—both teams trading blows right to the whistle—reflected the kind of competitive balance the model had identified, even if the specific scoreline proved beyond our read.

What emerged was a proper advert for why football doesn't follow scripts. Both sides had quality on the ball, neither could quite seal it, and Everton's late leveller meant neither went home happy. That's how 2-2s happen—not through accident, but through two teams who had a go and ran out of time to find a winner.