Live, 11th inning, MIL 3-2 SD: our model likes Under 6.5 at +125 — and we still aren't betting it

This is a live, in-game price, not a pregame line. MIL @ SD is in progress — top of the 11th, Milwaukee leading 3-2. The total we're discussing is Under 6.5 at +125, offered right now with the game already past regulation. A pregame Under 6.5 and an 11th-inning Under 6.5 with five runs already on the board are completely different instruments. Read every number below as in-game.

What we found

Our simulation, re-run from the current game state, puts Under 6.5 at 57.6%. The market is offering +125, which is a materially longer price than a coin flip. Against our number, that's +29.6% expected value — the largest disagreement on our board tonight, and a 4-star rating on our internal scale.

And our engine would not place the bet.

That is not a contradiction we're papering over — it's the entire point of publishing this. A large EV figure is an *output of a model*, not a fact about the world. When a game is in extra innings with a one-run margin, the remaining-outcome distribution is thin: a single swing resolves it. Our position-sizing layer treats a high-EV signal derived from a small remaining sample differently than the same figure computed over nine innings of simulated play. The EV can be real and the bet can still fail our risk gate.

The same pattern shows up across tonight's live board. MIL ML at -153 with a model probability of 69.8% carries +15.35% EV. NYM ML at +308 in the bottom of the 2nd, 0-1, prices at a model 29.9% — +21.85% EV. CLE ML at +190 in the top of the 4th, trailing 2-3, models at 37.5% for +8.81% EV, 3 stars. Four live disagreements, four flagged edges, zero the engine chose to fire.

What it means

If you only read the EV column, tonight looks like a feast. If you read the whole system, tonight is a night our model talks and our risk layer says no. We think the second reading is the honest one, and we'd rather show you a board of declined signals than pretend every flagged edge is a recommendation.

The useful, falsifiable claim here is narrow: at top-11 in a 3-2 game, we believe the true Under 6.5 probability is nearer 57.6% than the ~44% the +125 price implies. That claim gets graded whether or not we backed it with money. Over a large number of such spots, either our live re-simulation beats the closing in-game price or it doesn't — and we'll publish the record either way.

We are not telling you what Milwaukee or San Diego will do. We publish probabilities and the track record of those probabilities. 21+ where legal.

What would change our mind: if declined-but-flagged live signals like these grade out profitably over a few hundred logged instances, the risk gate is too tight and we'll loosen it — and say so here.