The model read: Ballpark Pal on a drop

Two clocks

A drop is news in seconds, and the arithmetic on it is a slot change: this batter now takes more trips to the plate, so his volume markets argue over. That is the first clock, and it is the one Lineup Dropper has always run on.

Ballpark Pal runs a full simulation of every MLB game, and it re-runs a game within about three minutes of that game's official lineups. That re-sim is the second clock, and this page is everything it adds. Nothing here delays a drop or an alert: the news arrives on its own clock, and the model catches up to it.

Because the simulation re-runs once both teams' cards are in, a game's re-sim usually lands with the second of its two drops. A lineup that posts first will often be showing you a pre-drop model read for a while, and it says so.

Every place Ballpark Pal speaks on screen carries its mascot as a small round badge. That is the tell: a badge means the simulation said it, and plain text means the lineup change did.

The Ballpark Pal fair price

Open any market and, underneath the drop-window chart, you may see a Ballpark Pal fair price: the simulation's own no-vig price for the shown side of exactly the line you are looking at.

It sits beside the book prices and never among them. The Bets list is a ranked list of prices you can actually take, and there is no book behind the fair price and nothing to take it at, so ranking it there would put the one unbettable number in the list. Use it as a reference point for whether the market has caught up to the news, not as an offer.

Two labels tell you which lineup it was priced off:

• Pre-drop sim means the simulation has not seen the news yet. It is pricing the same lineup the books are still pricing, which makes it a useful "before" number and nothing more.
• Simmed on this lineup means the re-sim has run against the lineups that actually posted, and this is the model's current read.

The fair price is shown for the line the simulation priced and for no other. If the simulation has no row at your line, or only reached it in the extreme tail where a probability stated to three decimals gives a price with a rounding band half as wide as the price itself, there is no fair price on that market at all rather than one read off a neighbouring line. Prices also always come from the model's newest run, so a line the newest run is silent about, such as a batter who left the order, drops its fair price rather than keeping a stale one.

Flags on batter rows

Batter rows read exactly as they always have unless the two projections disagreed about that batter. When they did, the row gains a small warning triangle beside his name. Opening it states what each one had, flatly and in both directions, for example that RotoWire had him batting third while Ballpark Pal did not have him in the lineup at all.

A flag is a disagreement, never a direction. It calls neither projection right and it never names a side to take, because direction in this tool comes from the lineup event and never from two projections arguing. The case worth catching most is the loudest one: one projection had him and the other missed him entirely. A batter neither projection listed is a batter neither was wrong about, so he carries no flag, and a game Ballpark Pal holds nothing for carries no flags at all.

The model projection is also a spare tyre. When RotoWire's projection is missing for a game, the drop is compared against Ballpark Pal's projected lineup instead of collapsing into a "Projection unavailable" card, so an upstream failure costs you the better baseline rather than the whole drop.

Pitchers: the opposing starter

Under the Batters list sits a Pitchers section: one row, the opposing starter of the lineup you are reading. That is the pitcher the batters who just moved are going to face.

The row states what the lineup change did to his strikeout outlook, a figure written ΔK. Expanding it lists his markets, and below them the arithmetic in full, one line per changed batter:

• his strikeout rate per plate appearance against this particular starter,
• how many times his slot comes up against that starter before and after the change, written for example as 4x → 3x, with a dash on the side of the change he was not in the lineup for,
• and the two multiplied together, which is his share of the swing.

The total sits underneath. Contributions are listed rather than quietly netted, because two substitutions can cancel out to almost nothing, and a total of nothing looks identical to a lineup where nothing happened unless you can see the parts.

These rates match Ballpark Pal's own site

Ballpark Pal works out every batter-versus-pitcher matchup twice: once with the ballpark and the weather, and once without. Its site shows the first, and so do we, so you can put the two screens side by side and read the same percentage for the same batter against the same starter.

There is one case where we cannot. A batter written onto the lineup card late is one Ballpark Pal has no prepared matchup for, and the endpoint that answers for him one at a time is given a batter and a pitcher but no game, so it has no ballpark to allow for. We carry that answer onto the same scale as the rest of the table by measuring what the ballpark did to a batter we do have both readings for. Where even that is not possible, the table says underneath that its rates are the unadjusted ones.

The adjustment is a single multiplier for the whole game, so it scales every batter in the table by the same amount. That means it can nudge the size of the total but never the order of the rows, and never which way the total points.

The number is in the units of that starter's outing, which is the same currency as his strikeout line. It reads on the row exactly the way a batter's expected plate appearances do, next to the name it belongs to: "-0.42 projected strikeouts", and then who projected it.

The swing says which way the strikeout outlook moved and never how far past any particular number the market is posting, so picking the line is left to you. Past a swing of a quarter of a projected strikeout in either direction, the starter's markets under the row say the side in their own words - "Bet under Player Strikeouts 3.5" - and each one opens the same bet detail panel a batter's does, with the same book list, the same fair price, and the same take. Inside that threshold the markets are watches, and the row is only the swing.

Two things stop any side from being named, and both are deliberate:

• An incomplete sum. If the simulation holds no strikeout rate for one of the changed batters, that batter's row reads "unknown", the swing is marked incomplete, and nothing signs a side. A sum with a hole in it that signed a side would be the one number on the screen you could not check.
• No named starter. Starter identity is a live fact rather than a morning one, and a projected opener can be replaced by the time the lineup posts. When Ballpark Pal cannot name who is starting, the whole Pitchers section is absent rather than showing a section about a guess.

Model verdicts on watch markets

RBIs and runs are marked watch at the drop, because a slot change gains plate appearances and loses runners on base at the same time and the lineup change alone cannot settle which wins.

The re-sim can settle it. Minutes after the drop, the simulation's projection for that batter in that market is compared against what it projected before the news, and if it moved far enough the market upgrades in place: the watch becomes a signed direction, and the chip that says it carries the Ballpark Pal badge. No animation, no transition, no new card. The row was a watch and now it is signed.

"Far enough" is currently a tenth of an expected run or RBI, and it is deliberately strict. Below that the market stays exactly the watch it was, because a simulation that moved a hundredth has not said anything worth acting on. Every movement is recorded either way, whether it cleared the bar or not, so the bar itself can be tuned against real evidence later.

Two consequences worth knowing:

• A market you looked at seconds after a drop can be signed the next time you open the panel. That is the verdict arriving, not the tool changing its mind.
• A market the drop already signed is never re-signed by the model. Volume markets belong to the lineup change.

When the model argues with the drop

When the re-sim moves against a direction the drop's arithmetic already gave a market, the row keeps that direction and gains the same warning triangle a batter row carries. Opening it states both readings side by side: what the drop said, and what the re-sim said, with its movement.

The simulation is allowed to sign a market the lineup change could not sign, and it is never allowed to overturn one it could. The arithmetic is a reading of the news itself, the simulation is a model of what the news is worth, and a model that could delete the news would make a wire into a slower simulation feed. So conflicting evidence is shown to you rather than resolved for you by whichever number happened to arrive last.

If you take a bet from a panel, what was on screen goes into your tracker with it: the fair price you were shown, the model's reading as it stood at that moment, and which of the two clocks signed the direction you took. That is what lets you compare, later and in your own results, how the two kinds of read have worked out for you.

When Ballpark Pal is unavailable

All of it disappears cleanly. No placeholders, no empty sections, no error cards: no fair price under the chart, no flags on batter rows, no Pitchers section, and watch markets that stay watch markets. What is left is the tool exactly as it worked before any of this existed, which is the point.

For the wire itself, movers, and the bet detail panel, see What Lineup Dropper is.