We stopped printing 100%

Until today the MLB board said the Dodgers were 100% to make the playoffs. That came from a 10,000-season Monte Carlo in which they qualified 10,000 times — which is a real result, and still not 100%.

What the simulation actually established is that the answer is higher than 10,000 runs can measure. Roll the dice more times and you would eventually find the seasons where it goes wrong. Sampling is the small problem, though. The bigger one is that the model cannot see a torn elbow in August, a suspension, or a trade that guts a rotation. Every number here is conditional on the world behaving roughly as it has been, and 100% quietly drops that condition.

So the boards now top out at >99% and bottom out at <1%, in every sport. Not a flat 99, because that states a precision the simulation never produced — the greater-than sign is the honest part. It says: past here, we stop being able to tell.

A club that missed the playoffs in all 10,000 runs used to read 0%. That is the same overclaim pointing the other way: "it didn't happen in ten thousand tries" is not the sentence "it cannot happen." Those now read <1%.

This also brings the boards in line with something the per-game model has always done quietly. Every single-game win probability we lock is clamped to between 1% and 99% before it is stored and graded, for exactly these reasons. The season-long boards were the inconsistent ones.

With the deadline just past, the obvious question for a team already at >99% to play in October is whether a front-of-rotation arm even matters. We ran it: a genuine ace, worth about four wins over a full season, added to this Dodgers roster.

Their projected win total moves 100.5 to 101.8. Their playoff odds move by nothing at all — you cannot improve on already-there. If your instinct was that a big arm pads the win total, the model mostly disagrees: with roughly fifty games left, four full-season wins is worth about one and a third.

October is where it shows up. The pennant goes from 35% to 42%, and the World Series from 24% to 30%. Seven points of pennant equity from 1.3 wins.

Because championship odds are convex near the top. Down the standings, a marginal win buys you a marginal shot at a wild card. At the front, it buys a better seed, which buys a bye, which buys an easier draw, which compounds through three rounds. The same effect made us talk ourselves down on the Sixers in July — there it inflated a number we didn't trust, here it lifts one we do.

Worth saying plainly: we got this wrong in-house before we ran it. The back-of-envelope guess was two or three points, on the reasoning that a team this good is already close to its ceiling. The simulation said seven. The envelope was wrong because it was adding where the bracket multiplies.

Our postseason bracket is simulated on team strength, not on who takes the ball in Game 1. A real ace starts two of every five games in a short series, which is worth more in October than a season-long win total implies — so if anything, the 42% is conservative for a true number-one.

That is a known limit of the model and it is on the MLB model card, not buried here. The single-game engine weights the announced starter; the season simulation cannot, because nobody knows who starts a hypothetical Game 3 in October.