October baseball has a completely different feel from the regular season. There is less room for mistakes, managers make decisions faster, and one bad inning can change the entire direction of a series. That is what makes NLDS predictions so interesting. You are not just trying to figure out which team is better. You are trying to figure out which team is better in this particular matchup, under these particular conditions, with these particular pitchers and players available.
That sounds obvious, but it is surprisingly easy to forget.
A team can be excellent over 162 games and still have a rough matchup against another club in a five-game series. Maybe its best hitters struggle against the opposing starter's pitch mix. Maybe its bullpen is strong on paper but two of the most important relievers are coming off heavy workloads. Maybe the ballpark takes away some of the team's power. Maybe the opposing manager has a bench that creates problems late in games.
Those details matter.
When I look at an NLDS matchup, I do not want to start with the standings or simply ask which team has the better record. I want to build the game from the ground up. Who is pitching? What does that pitcher throw? Which hitters are likely to face him? How long is he expected to stay in the game? Which relievers are actually available? What does the park do to the ball? What is the weather like? What happens if the game gets into the seventh inning with a one-run difference?
That last question is a big one.
The playoffs are when bullpen depth really starts to matter. Managers don't have much time to mess around, so they might pull a starter after five innings even if he's been pitching pretty well. If the next part of the lineup is dangerous, they may just decide not to take the chance. You can also see relievers used in spots that would be unusual during the regular season. A guy who normally comes in for the eighth might be brought in during the sixth because the other team's best hitters are coming up.
This is why I prefer probability over certainty. Baseball is too unpredictable to say a team is going to win. What we can do is estimate how often each outcome should happen and then compare that estimate with the price being offered.
That is basically the philosophy behind the approach below.
Table Of Contents
Series context and schedule
Pitching matchups and run prevention
Lineup dynamics and defense
Data-driven modeling approach and scenarios
What to watch live and how to adjust
Practical betting notes for NLDS
How to run your own quick simulation
Frequently Asked Questions
Key Takeaways
The biggest thing I look at in an NLDS is run prevention. Starting pitchers obviously matter, but I also want to know how quickly managers are likely to go to the bullpen and which relievers will actually be available when the game gets tight.
The matchup matters more than the team name. A lineup that looks great against right-handed pitching might have a completely different outlook against a lefty with a nasty breaking ball. A team with plenty of home run power may also lose some of that advantage in a park that suppresses carry.
Recent results are useful, but they have to be handled carefully. A three-game hitting streak does not suddenly turn an average hitter into an MVP-level bat. At the same time, if a pitcher has changed his pitch mix, gained velocity, or developed a new weapon, ignoring that information would be just as bad.
I like building a baseline from a larger sample and then adjusting it with current information. That gives you something stable without pretending the players are exactly the same as they were six months ago.
The other big lesson is that an NLDS prediction should never be treated as a guarantee. A 56% favorite still loses plenty of the time. In a short series, even a legitimately better team can go home early.
ATSwins can be useful here because it gives you a way to organize projections, betting information, player props, betting splits, and results tracking. The point is not to blindly follow a number. It is to have a more organized process when there are a lot of numbers competing for your attention.
Series Context and Schedule
Before I even start looking at individual players, I want to know how the series is laid out.
The NLDS is a best-of-five, which means the schedule can have a pretty big influence on how teams use their pitching. The standard format gives the higher seed home-field advantage with Games 1 and 2 at home, Games 3 and 4 on the road, and Game 5 back at home if the series gets that far.
That might sound like a basic scheduling detail, but it affects the entire series.
A manager who knows there is a travel day coming may be more aggressive with his bullpen. If there is no day off between games, the same manager may have to think twice before using his best relievers for multiple innings.
That is why I always start with the actual schedule rather than assuming every game will have the same conditions.
I want to know where each game is being played, when it starts, which pitchers are expected to start, and how much rest the bullpen gets before and after the game. If a rain delay changes the schedule, that information needs to go into the analysis immediately.
The starting time can matter too. A game starting in the afternoon can play differently from a late-night game. Temperature, wind, humidity, sunlight, and the transition into nighttime can all have some effect on how the ball carries.
I would never build an entire betting decision around weather, but if two teams are already closely matched, a small environmental advantage can matter.
The same idea applies to the ballpark.
Some parks reward home run power. Others take some of the life out of fly balls. Some create more doubles and triples because of their dimensions. A lineup that depends heavily on extra-base hits can therefore look slightly different depending on where the game is being played.
I prefer using multi-year park information instead of relying on one season. There is too much randomness in a single year. Once the general park environment is established, current weather can be added as another layer.
Travel deserves a little attention too.
Players are not suddenly going to forget how to hit because they flew to another city, obviously. The more important issue is how travel affects the pitching staff and schedule. If the bullpen is already tired and the next game starts quickly after a trip, the manager may have fewer options than the box score suggests.
This is one of those areas where the information does not look exciting. Nobody is going to make a viral post because a team has an extra hour of rest. But playoff betting is often about collecting these smaller details and seeing how they fit together.
I like keeping a basic schedule record for every game. I note the ballpark, first pitch time, expected starters, travel day, bullpen workload, and weather. That gives me a much clearer picture of how the series might develop.
And the key word there is develop.
An NLDS should not be treated like five completely separate games. What happens in Game 1 can change Game 2. What happens in Game 2 can change Game 3. A reliever who throws 30 pitches tonight may technically be available tomorrow, but that does not mean I should project him exactly the same way.
That connection between games is one of the biggest things I want the model to understand.
Pitching Matchups and Run Prevention
Starting Pitchers and What They Actually Throw
Starting pitching is usually where the first major edge appears.
I want to know the obvious stuff, like ERA and strikeout rate, but those numbers are not enough for me. I want to know how the pitcher is actually getting his results.
What does his fastball look like? How hard is he throwing it? Is the slider getting swings and misses? Is his changeup generating weak contact? Has his velocity changed recently? Is he walking more hitters than usual?
Those questions give me a better idea of what could happen next.
Pitch mix is particularly important in playoff matchups because the same pitcher can have completely different results against two different lineups.
Imagine a pitcher who relies heavily on a slider. If the opposing lineup has several hitters who consistently handle sliders, I am not going to value that pitcher's season numbers quite as highly. He may still be a very good pitcher, but the matchup is less favorable.
The opposite is also true. A hitter who has struggled overall might suddenly have a very interesting matchup if the opposing pitcher relies heavily on pitches that hitter handles well.
This is where matchup analysis becomes more useful than simply comparing season ERAs.
Velocity is another thing I watch.
If a starter normally sits around 95 mph and suddenly spends an entire game around 93 to 94, I want to know why. Maybe it is intentional. Maybe the pitcher is pitching differently. Maybe he is tired. Maybe there is a mechanical problem.
One bad velocity reading does not mean much by itself. A repeated change is more interesting.
Pitch movement is similar. A slider with less sweep or a changeup without its usual separation can become much easier for hitters to recognize.
Command might be even more important.
A pitcher can give up a couple of hard-hit balls and still be fine. But if he suddenly cannot throw strikes, the entire projection changes. More walks mean longer innings, higher pitch counts, and earlier bullpen exposure.
That is especially important in October because managers tend to have shorter leashes.
Times Through the Order
One of the biggest questions I ask is how long the starter is actually expected to pitch.
A pitcher might be capable of throwing 100 pitches, but that does not mean the manager wants him facing the lineup a third time.
The times-through-the-order penalty is not identical for every pitcher. Some starters handle the third trip surprisingly well. Others lose effectiveness pretty quickly.
The opponent matters too.
A deep lineup can punish a tired starter more than a weaker lineup can. If the opposing manager has several dangerous hitters coming up in the sixth inning, the starting pitcher might get pulled earlier than expected.
That creates a direct connection between starting pitching and bullpen analysis.
If I expect the starter to give me six strong innings, I can give more weight to the starting matchup. If I think he is likely to go only four or five innings, the bullpen becomes a much larger part of the projection.
Short Rest
Short rest can create some interesting playoff situations.
Some pitchers handle it better than others. There is no reason to automatically apply the same penalty to every starter simply because he is working on less rest.
I look at how the pitcher has handled previous short-rest situations, his recent pitch count, velocity, command, and whether the team is likely to limit his workload.
A short-rest starter may have a lower expected pitch count even if the manager says he is available. That matters because every lost inning has to be covered by somebody else.
Bullpen Availability
The bullpen is where an NLDS can get really interesting.
A regular-season bullpen ranking tells you which team has generally pitched well. It does not necessarily tell you which relievers will be available tonight.
That distinction matters.
A team can have a great bullpen, but if its two best relievers both threw 25 pitches yesterday, I cannot project that bullpen exactly the same way.
I think about relievers in tiers rather than just saying a bullpen is good or bad.
There are the high-leverage arms. Then there are setup relievers who can handle important innings. Then there are multi-inning pitchers and lower-leverage options.
The question is who is available and who the manager trusts.
Some managers are comfortable using their closer in the seventh if the best hitters are coming up. Others stick more closely to traditional inning assignments. Some managers will use a reliever three days in a row. Others avoid it unless absolutely necessary.
Those tendencies can become part of the model.
Pitch count matters. So does warm-up activity.
A reliever can warm up twice without entering the game. That does not show up in the basic box score, but it is still information. If a high-leverage pitcher has been constantly preparing to enter, I am at least curious about whether that affects his next appearance.
Velocity can provide another clue.
A small drop is not automatically concerning. But if a pitcher is down close to a mile per hour and also missing his usual spots, I am going to be more cautious about projecting him for another high-leverage appearance.
The point is not to turn every tiny change into a dramatic story. The point is to notice when several signals are pointing in the same direction.
Turning Pitching Into Expected Runs
Once the starters and bullpens are evaluated, I want to turn all of that information into an expected scoring range.
The basic idea is pretty simple.
Estimate how many innings the starter is likely to pitch. Estimate how effective he will be against this particular lineup. Then estimate who handles the remaining innings.
After that, adjust for the ballpark, weather, and defense.
Defense matters because pitchers do not control every ball hit into play. A strong defensive team can turn difficult balls into outs, while a weaker team can allow extra baserunners even when the pitcher does not make a huge mistake.
I also care about the type of contact.
A lineup that hits a lot of ground balls may interact differently with a pitcher who generates grounders than a lineup built around high fly-ball rates. A home run-heavy offense can have a different ceiling in a small park compared with a large one.
Once those pieces are combined, the goal is to produce an expected run distribution rather than one exact score.
That distinction is important.
If my model says a team should score 4.2 runs, that does not mean I expect the final score to be exactly four runs. It means the average outcome across many simulations is around that number.
That is the way I want to think about baseball probabilities.
Lineup Dynamics and Defense
Projecting the Starting Nine
The lineup is not just a list of nine players.
The order matters. The handedness matters. The opposing pitcher's style matters. The bench matters. And the manager's willingness to make changes matters.
I usually start by projecting the lineup against the expected starter. Then I look at how the lineup changes if the opposing manager goes to a lefty or righty reliever later in the game.
That is where playoff benches become interesting.
A team may start a hitter who is not necessarily the best overall player because the matchup makes sense. Then, later in the game, a different hitter can enter against a particular reliever.
The more predictable those moves are, the easier they are to model.
Platoon splits are useful, but I am careful with small samples. If a hitter has a massive split over a limited number of plate appearances, I do not want the model treating that number as a permanent skill.
I would rather blend career or multi-year performance with current-season results and then adjust for the specific pitcher.
That keeps extreme numbers from taking over the projection.
Contact Quality
Batting average can tell you what happened. It does not always tell you why.
That is why I like looking at contact quality and plate discipline.
Is the hitter making hard contact? Is he chasing pitches outside the zone? Is he making contact when he does chase? Does he handle fastballs? What happens against breaking balls?
Those questions can reveal things that traditional statistics miss.
A hitter may have a .240 average but strong underlying contact quality. Another hitter might be hitting .320 while benefiting from a string of favorable outcomes.
Neither situation guarantees what happens in the next game.
I want to understand the process underneath the result.
Pitch fit is particularly useful in an NLDS because the sample is so small. If a hitter is especially good against the exact pitch a starter relies on, that can create a meaningful matchup advantage.
On the other hand, a lineup full of high-chase hitters can have a rough night against a pitcher who gets a lot of swings outside the zone.
Baserunning
Baserunning is another small piece that can matter more in a tight playoff game.
A stolen base can move a runner into scoring position without requiring another hit. A strong runner can pressure a catcher and pitcher in ways that do not appear directly in the box score.
I look at stolen-base attempts, success rate, pitcher delivery time, catcher throwing ability, and the game situation.
Obviously, teams are not going to run every time they see a slow pitcher. Managers understand the risk. But when the matchup is favorable, the opportunity is there.
That is another reason I like looking beyond season averages.
The question is not simply, "How many bases did this player steal?"
The better question is, "How likely is this player to attempt a steal against this pitcher and catcher combination?"
Defensive Value
Defense is easy to ignore because offensive statistics are more visible.
But one defensive play can change an entire playoff game.
A good outfielder can take away an extra-base hit. A shortstop can turn a difficult ground ball into an out. A catcher can block a pitch with runners on base.
Those things are not always obvious when you look at the final score.
For modeling purposes, I want to include defensive quality as part of run prevention. I do not want to make the adjustment huge, but I also do not want to pretend every team converts balls in play at the same rate.
Catcher defense deserves attention too. Receiving, blocking, throwing, and controlling the running game can all have an effect.
Again, these are usually secondary factors. But if two teams are very close, secondary factors can become the difference.
Bench and Pinch-Hitting Decisions
The bench becomes more important as the game gets later.
I like to know which hitters are available against left-handed and right-handed relievers. I also want to know whether a manager is comfortable using the backup catcher early or saving him for an emergency.
The order of substitutions can change the end of the game.
If a manager uses his best pinch hitter in the seventh, that player is no longer available in the ninth. If a defensive replacement enters early, another bench option disappears.
This creates a sort of chain reaction.
That is why a good playoff projection should not only ask who starts the game. It should ask who is likely to finish it.
Data-Driven Modeling Approach and Scenarios
Building the Baseline
The model starts with a baseline for both teams.
For hitters, that includes offensive production, plate discipline, contact quality, power, and performance against different handedness.
For pitchers, it includes expected run prevention, strikeout and walk ability, pitch mix, velocity, movement, and contact quality.
For defenses, I look at overall run prevention and individual strengths that may matter in the matchup.
Then I add the bullpen.
This is where I want to be more flexible because bullpen performance and availability change quickly.
A reliever's season numbers still matter, but today's availability may matter more than his average from three months ago.
Recent Form Without Overreacting
There is always a temptation to chase recent results during the playoffs.
A hitter gets hot and suddenly everybody wants to move his projection way up. A pitcher has two great starts and suddenly his old weaknesses disappear from the conversation.
I try not to do that.
Recent form matters when it represents something real. A pitcher who has changed his pitch selection or gained velocity is different from a pitcher who simply had two low-scoring outings.
Likewise, a hitter who has changed his swing or is making better contact may deserve an adjustment.
But a five-game streak is not enough by itself.
The model should have a memory.
That means keeping a stronger prior from larger samples and then allowing current information to move the number when there is enough evidence.
The exact weight depends on the player. A veteran with years of reliable performance should not have his entire projection rewritten because of two weeks of results. A young pitcher who has just added a new pitch may require a much more aggressive update.
Building Game-Level Projections
Once the inputs are ready, the next step is estimating expected runs.
I like splitting the game into two parts: the expected starter innings and the expected bullpen innings.
For example, if a starter is expected to go five and a half innings, the model needs to estimate what happens in the remaining innings. That means identifying the likely relievers and accounting for their availability.
Then the park and weather adjustments are added.
The final number becomes the team's expected run output.
But again, I do not want one exact number to represent the entire game.
A team expected to score 4.2 runs could score zero, three, four, eight, or more. The distribution around the average is what makes baseball difficult.
That distribution is also what makes simulation useful.
Monte Carlo Simulation
Once I have the game-level distributions, I can simulate the series thousands of times.
For every simulation, the model plays the games according to the schedule. If a bullpen gets heavily used in one simulated game, that affects its availability in the next simulated game.
That is important because the games are connected.
If Team A's closer throws 28 pitches in Game 1, the model should not treat him as perfectly fresh in Game 2. If Team B wins Game 2 and avoids using its best relievers, those pitchers may be more available in Game 3.
The simulation can then estimate the probability of winning the series, the most likely series length, individual game probabilities, and expected scoring ranges.
Imagine the simulation gives Team A a 56.8% chance to win and Team B a 43.2% chance.
That is not a prediction that Team A will definitely win.
It is saying that under the assumptions being used, Team A wins in roughly 57 out of 100 equivalent simulations.
That distinction sounds small, but it is incredibly important when betting.
Stress Testing
I always want to know how much the result changes when I change the assumptions.
What happens if the starter loses half a mile per hour?
What happens if the lineup's best hitter sits?
What happens if the top reliever is unavailable?
What happens if the wind changes?
What happens if the manager uses the bullpen aggressively?
These tests show whether the edge is strong or fragile.
If Team A's series probability stays around 55% to 57% through several reasonable scenarios, that is more reassuring.
If it jumps from 51% to 61% because of one tiny assumption, I would be much more cautious.
That is where confidence ranges become useful. Rather than pretending the model knows the exact probability, I can communicate the central estimate and how much it could reasonably move.
Integrating ATSwins Into the Process
This is where I like having a platform such as ATSwins as part of the workflow.
The goal is not to let the platform make every decision. The goal is to make it easier to compare your own analysis with market information.
Suppose I make a team a -130 favorite. If the market is offering -118, that gets my attention.
But I do not immediately bet it.
I first ask why the difference exists.
Maybe the market knows something I missed. Maybe the starting pitcher is not fully healthy. Maybe the bullpen has a hidden availability problem. Maybe the lineup is different from what I projected.
If I can explain the difference and the underlying assumptions still hold, then the price becomes interesting.
Player props work the same way.
A strikeout over might look attractive until I realize the pitcher is only expected to throw five innings. A total-bases over might look good until the hitter's expected plate appearances are lower than normal.
That is why I want the projection to include opportunity, not just talent.
ATSwins can also help with tracking betting splits and keeping a record of results. This is useful because memory is terrible at evaluating betting performance.
People remember the big wins.
They often forget the five smaller bets that lost before it.
A tracker gives you a more honest picture.
It also helps with sizing. If the edge is small, the stake should generally be smaller. If the model is uncertain, there is no reason to force a large position.
That is the difference between having a prediction and having a betting process.
What to Watch Live and How to Adjust
Bullpen Velocity and Usage
Once the game starts, the information changes.
The most obvious thing to watch is who is warming up.
If a high-leverage reliever starts throwing in the sixth, that tells you something about how the manager views the situation.
If that same reliever enters and throws 25 pitches, his availability tomorrow changes.
Velocity is worth monitoring too.
A small decline is not enough to make a major decision. But if the velocity drop is combined with poor command or reduced movement, it becomes more meaningful.
The same applies to pitch shape.
A slider that is not moving normally may not generate the same number of whiffs. That can turn a pitcher from a strikeout threat into more of a contact pitcher.
Starter Command
I care more about command than one bad result.
A starter can give up a home run and still be pitching well. Maybe the hitter made a great swing on a good pitch.
But if the starter is constantly missing arm-side, falling behind hitters, and throwing too many pitches, that tells me something different.
The expected innings projection should come down.
The bullpen exposure goes up.
That can change the next game's projection as well.
Catcher and Umpire Effects
The strike zone can quietly change the game.
A generous zone allows pitchers to attack. A tight zone forces them to work deeper into counts.
That affects walks, strikeouts, pitch counts, and eventually how long the starter stays in the game.
Catcher setup can also matter. If the pitcher and catcher are consistently targeting a certain area and the umpire is giving them that location, the pitcher can become more aggressive.
These are not huge standalone factors. I treat them as smaller adjustments that become more useful when they line up with the rest of the information.
Bench Usage
Keep an eye on which bench players have already entered.
If a team's best pinch hitter is gone by the seventh inning, that changes the later innings.
If the backup catcher has been used early, there is less flexibility later.
This is why live betting can become interesting in playoff baseball. The game state tells you which players are still available.
Weather Changes
Weather can move during a game.
Temperature drops after sunset. Wind can shift. A rain delay can suddenly change the pitching plan.
If the weather changes dramatically, the model should respond. But I do not want to overreact to every small forecast change.
A few degrees is not necessarily meaningful.
A major wind shift or a delay that forces a pitcher to cool down is much more important.
Between-Game Updates
After every game, I want to update the next game without rewriting everything.
Confirm the starting pitchers.
Review bullpen usage.
Check injuries.
Look at the projected lineup.
Update the weather.
Check which relievers are actually available.
Then compare the new projection with the market.
The important thing is to separate new information from emotional reaction.
A blowout does not necessarily mean the winning team is suddenly much better.
A one-run loss does not necessarily mean the losing team was much worse.
The underlying information is what matters.
Practical Betting Notes for NLDS
The first thing I would say about NLDS betting is simple: do not bet the team name.
A famous team can be overpriced.
A less exciting team can have the better matchup.
The price is what matters.
If your model gives a team a 55% chance to win and the market price implies 52%, there might be an opportunity. But the size of the wager should depend on how reliable that 55% estimate is.
If the projection is built on several stable factors, I have more confidence.
If it depends on one questionable lineup assumption and a weather forecast that could change, I am more cautious.
Totals require the same thinking.
Playoff unders can look attractive because everyone expects good pitching. But the market knows that too. You need to find the specific reason your projection differs.
Maybe the weather is warmer than expected.
Maybe the game starts earlier than usual.
Maybe one bullpen is more tired than the market realizes.
Maybe a starter has a shorter leash.
Player props can be even more sensitive.
A pitcher strikeout prop depends on both strikeout ability and opportunity. If he is unlikely to reach the seventh inning, his ceiling is lower.
A hitter's total-bases prop depends on plate appearances, lineup position, pitcher matchup, bullpen matchup, and park.
The more variables involved, the more important it becomes to avoid false precision.
If my model says a player has a 51.5% chance of going over and another calculation says 52.2%, I do not consider that a huge difference. The uncertainty in the projection is likely larger than the difference between those numbers.
Sometimes the best decision is to pass.
That is especially true in a short playoff series where variance is high.
How to Run Your Own Quick Simulation
You do not need a giant machine-learning system to get started.
A basic simulation can still be useful if the inputs are sensible.
Start with the offensive strength of each lineup. Adjust it for the starting pitcher's handedness and pitch mix. Estimate the starter's expected innings. Then estimate the bullpen innings and the quality of the relievers likely to pitch.
Add the park.
Add the weather.
Add a small defensive adjustment.
Now you have a basic expected run total for both teams.
From there, generate thousands of possible game scores using a suitable run distribution.
Count how many times each team wins.
That gives you a basic win probability.
For a series, repeat the process for each game and carry bullpen usage forward.
This is where the simulation becomes more interesting because the series has memory.
If a team burns its best relievers in Game 2, the model should know that when Game 3 starts.
I also like running different scenarios.
One scenario assumes both teams have full bullpen availability.
Another removes one key reliever.
Another changes the lineup.
Another changes the weather.
You can then see what actually moves the series probability.
That is more useful than simply generating one giant number and assuming it is correct.
And always sanity-check the result.
If your model says a game should have a total of 5.8 runs while the market is sitting at 8.0, ask why.
Maybe you have forgotten a hitter.
Maybe the weather input is wrong.
Maybe the bullpen projection is unrealistic.
Maybe the market is wrong.
The last possibility is exciting, but it should not be your first assumption.
Good models challenge themselves before they challenge the market.
Final Thoughts
The best way to approach NLDS predictions is to stop thinking about them as a simple question of which team is better.
The better question is which team has the better chance of winning this specific series under these specific circumstances.
That means starting with the pitchers, but not stopping there.
Look at the pitch mix.
Look at the lineup.
Look at platoon advantages.
Look at expected innings.
Look at the bullpen.
Look at who is actually available.
Look at the park.
Look at the weather.
Look at defense.
Then look at the price.
The numbers should help you understand the game, not make you feel like you have eliminated uncertainty.
That distinction is important because playoff baseball can get weird very quickly. A bloop can become a double. A bad hop can extend an inning. A reliever can lose his command for five pitches. One swing can turn a 2-1 game into a 5-1 game.
The model has to accept that randomness.
What it can do is make sure you are not making decisions based entirely on the latest score or whichever team looked better on television the night before.
That is where I think data-driven analysis has the most value.
You build the baseline.
You make the matchup adjustments.
You simulate the possible outcomes.
Then you keep updating when something meaningful changes.
ATSwins can be part of that workflow by giving you access to data-driven picks, player props, betting splits, and profit tracking across major sports. Used correctly, the goal is not to blindly follow a recommendation. It is to have another source of structured information that you can compare with your own analysis.
In the end, there is no perfect NLDS prediction.
There is only a better or worse estimate of what is likely to happen.
The goal is to find the situations where your estimate is meaningfully different from the price being offered, understand why that difference exists, and avoid risking too much when the uncertainty is high.
That is how I would approach October baseball.
Not with certainty.
With probabilities, context, and a willingness to change the number when the information actually changes.
Frequently Asked Questions
What matters most when making NLDS predictions?
I would start with starting pitching and bullpen availability.
The reason is pretty straightforward. In a short series, teams are usually much more aggressive with their pitching than they are during the regular season. Starters have shorter leashes, top relievers enter earlier, and managers are constantly thinking about the next game.
After pitching, I look closely at the lineup matchup. Platoon splits, pitch mix, contact quality, and expected plate appearances can all make a difference.
Then I add the park, weather, defense, and schedule.
The most important thing is not to isolate one statistic. A great starting pitcher does not automatically create a great betting opportunity if the opposing lineup matches up extremely well against his primary pitches.
How should the model react when a starting pitcher changes?
A starting pitcher change can be one of the biggest reasons to move an NLDS projection.
I would first adjust the starting pitcher component itself. Then I would revisit the lineup matchup because the handedness and pitch mix may have changed.
After that, I would check the bullpen.
A replacement starter might be expected to pitch only four or five innings, which means the bullpen has to cover more outs. That can create an additional adjustment.
The same process applies when a starter is scratched shortly before the game. Do not just replace one name with another. Rebuild the expected innings and matchup from there.
Should recent playoff results be heavily weighted?
Usually not.
Recent results are important when they reveal a genuine change in player skill or usage. For example, if a pitcher has added a new pitch, gained velocity, or changed his approach, that deserves attention.
But a handful of playoff games is still a very small sample.
A hitter going 8-for-15 does not suddenly become a .500 hitter. A pitcher allowing one run across two starts does not suddenly eliminate all of his previous weaknesses.
I prefer to keep a larger baseline and then make measured adjustments when the new information is meaningful.
How can ATSwins be used for NLDS predictions?
ATSwins can be used as part of the broader research process.
You can compare model-based projections with available market prices, review player prop opportunities, examine betting splits, and track your results over time.
The tracking part is especially useful.
It is easy to remember a big winning bet and forget the losing bets around it. Keeping a record gives you a more realistic idea of whether your approach is actually producing value.
I would use ATSwins as another piece of the process rather than treating any single projection as a guarantee.
Are NLDS predictions different from regular-season predictions?
Yes, mainly because the decision-making environment changes.
Managers are more aggressive. Starters can be pulled earlier. Elite relievers can appear in unusual innings. Bench players become more important. Bullpen availability can change from one day to the next.
The games are also connected.
If a team uses three of its best relievers in a long Game 2, that affects Game 3. If a starter throws a high pitch count, the rotation plan for the next game can change.
That is why I prefer a series-level approach instead of simply projecting five independent games.
The basic statistics still matter, but the postseason context changes how those statistics should be interpreted.