Profits with Speculation: PART DEUX – AI, Sports Simulation, and Analytics

[NemoTheOne]

Hi BHW friends,

So about almost (exactly) a year ago, I created a HORSE RACING SIMULATOR – or, in plain terms, a “CHATGPT prompt”. I used this simulator (again, ‘just a prompt’) to forecast horse races and cash in on the winning race order combinations. I had no prior knowledge, or experience using, FanDuel, AI/Programming etc etc. I’m just a writer, which is my “main source of income”

It’s here if you’d like to open up another tab and take a look.

Profits from Speculation

Now, it’s a year later. I’ve discovered that the sports that I can make money on are:

Baseball (my favourite sport)
Basketball (The Euroleague is kinda cool.)
Horse Racing (Great, but the races go by too fast. Easy to fall into trap of “chasing”)

I’ve always wanted to “make money online” through some type of online casino type of deal, so this is filling that potential.

_ _ __ _ _ _ _ _ _

So, after a year of testing/backtesting, we are going to be, once again, targeting PROFITS THROUGH SPECULATION via my

BASEBALL SIMULATOR FEATURE ENGINEERING MODEL (a ‘prompt’). Let’s just dive right in, and explore the method.
_ _ _ _ _ _ _ _ _____________ _ _ _

First off, a Feature Engineering Model is a series of processes (in the form of a prompt) that identifies probabilities of certain things happening. For our purposes, this is the function. Basically, the model (GPT) runs 150,000 Monte Carlo simulations on any given game. Now, I’ve created CUSTOM METRICS, that seem to accurate indicate GAME NARRATIVES and KEY CONTRIBUTORS.

STEP 1: Run THE ENTIRE GAMES SLATE OF MLB GAMES FOR THE DAY

Let’s take this one step at a time. I’m changing some of the names of the Metrics, or Simulator names etc, since there’s no copyright etc, or anything – it’s a prompt (NOTE: These games are from 13th July. There was a typo in the prompt below, where "june" is written, instead)







So, that’s 3 pages, of the 5 page prompt.

We go to a site that has ALL THE MATCHUPS AND LINEUPS LISTED. I usually pick BASEBALL MONSTER. The below image shows the listings. (since I’m writing this the day after the games, the scores are shown)


I just copy and paste the entire page for the day, into the prompt (where it says ‘GAME INFO ENTERED HERE’)

My goal is always to DO AS MANY SIMULATIONS AS POSSIBLE, quickly, without compromising accuracy.

I do not want to be the guy that says “After HOURS OF RESEARCH I did these parlays! F8DE ME BRO”

_ _ _ _ _ _

So, now we have our initial GAME ANALYSIS OF THE ENTIRE DAY OF GAMES. Here’s what the output looks like:



STEP 2:

Now, that we have our initial report, let’s get some HUMAN INFLUENCE here. So, I then go to YOUTUBE, and copy the transcripts of about 7 or 8ish of these analyses. I then run a SECONDARY PROMPT, basically now having the SIMULATOR (same structure), ANALYZE MY SIMULATOR’S FORECASTS VS/INTEGRATED WITH HUMAN ANALYSES (transcript provided to the GPT from YOUTUBE). It then spits out a new report, like this:





___
By the way, that YELLOW HIGHLIGHTED, where it indicated Elly De La Cruz would produce, did indeed lead to Elly De La Cruz 2+TOTAL BASES, being on my PARLAY SLIP (and hitting). Same with the Freddy Peralta suggestion. And actually, Christopher Sanchez: UNDER 6.0Ks suggestion above, led me to (thankfully) REMOVE an OVER 7K leg that I had, earlier.

STEP 3:

Read the combined report. Make bets. Get results:






_ + _ _ _ --___

My challenge has always been organization. That’s part of the purpose of doing this type of journey again.

The other part is to show, that ANYONE can do this. For me, it gives me something that I like to do (baseball/sports analytics and simulation), in between my other work. Plus, money's cool, too.

Lastly, when wanting to turn this into a MAKING MONEY METHOD, basically, take a look at that winning parlay ticket…..It’s a mix of different games, events, and based on many variables including ballpark factors, player metrics – just like, the prompt described – so, basically, during one day’s game slate, I’ll make about 5-10 of those “Parlay tickets”.

The good thing is, if they/some miss – I’m only making $1-$4 bets most of the time. That’s the amount that I would invest in AMAZON MARKETING SERVICES, when I did Kindle. There are some times when I would make larger bets, or “safe plays”, but that’s what most are already doing, anyway.

STEP 4: TESTING

To be continued....

 

[NemoTheOne]

STEP 4: TESTING

This is the key. Here’s a snapshot:




...and so on….

Lastly, as the accuracy began to improve, I made a glossary of terms for this LLM structure


____+++


So, with that said, much like my last journey this will document MAKING MONEY ONLINE THROUGH SPECULATION.

I’m not much of a fan of “gambling” – but I like analytics and sports. This came as a result of my experiencing playing SUPER BASEBALL SIMULATOR 1000 many, many, many years ago.



Thanks for being part of this journey, again.

LET’S GET IT!

 

[jas reed]

Never really thought about combining AI with sports sims like that. If you can get even halfway decent prediction accuracy, there’s serious upside. I’d just start small, test a bunch, and see where the data actually lines up with real outcomes. Super curious how far you can take this, following for sure!

 

[NemoTheOne]

Never really thought about combining AI with sports sims like that. If you can get even halfway decent prediction accuracy, there’s serious upside. I’d just start small, test a bunch, and see where the data actually lines up with real outcomes. Super curious how far you can take this, following for sure!

Thanks for following! Welcome to BHW, by the way! I see you joined not too long ago.

And precisely - that's exactly where I am, here, with this. I began last year - started testing small, saw that there was accuracy. But that was during the FINAL month of Baseball season. So, during the off season, I worked on other sports, saw accuracy, and NOW applied it to baseball.

I found that whilst the EXACT EVENTS (i.e. HOME RUN, SINGLE) might differ - by doing the SIMULATIONS, the prompt/GPT/SIMULATION MODEL (whatever you want to call it) is able to identify basic narrative and KEY CONTRIBUTORS.

So, that's what we're able to leverage here, with this tool.

 

[NemoTheOne]

Hello all - as an update, I have succeeded with the MLB Baseball Simulator model using GPT to run Monte Carlo simulations and other processes that I'm learning more about, but couldn't even begin to understand. If you have GPT, and mybe can string together a word or two, in the form of instructions, you can do this.

The idea is for each sport, a different approach is needed.

For baseball, I've developed metrics that identify key cluster zones - innings of high votality - and players' probability of being associated with those clusters of activity. That's all in a nutshell. Key is backtesting.

I've been able to focus in, with success on speciality markets like TOTAL RUNS, FIRST 5 INNINGS, WINNING MARGIN, HITS, TBS, and a few others. Sometimes, especially with K'S, some strong value can be found. My goal as/is to do as much in as little time. If I had to ask AI to explain it, iw ould be THREE STEPS:
1. SIMULATE. 2. INVESTIGATE. 3.SPECULATE.

That phrase was born from a very intentional design philosophy: BHW MLB BaseBallPrompt (just a really long chatgpt prompt) may be a deep and technical simulation framework, but it had to be operationally simple in its user flow.

Let’s break it down:

---

SIMULATE – The Engine Phase​

This is where the BHW MLB BaseBallPrompt Simulator does what it was built for:

• It ingests full pre-game datasets (lineups, weather, park factors, pitcher form, etc.)
• Runs multi-phase scenario cycles (F3, F5, Full, 7–9 Window)
• Generates metric outputs (GSSM, BEI, DEA, PUI, CDT, BDI)

This phase answers: "What are the likely shapes and volatility zones of this game?"

The beauty is—you hit "run," and the engine executes 5,000+ scenario seeds per phase in seconds. No manual guesswork.

---

INVESTIGATE – The Analyst Phase​

Once the sim outputs, you investigate.

• You cross-check flags (Tier-1 Locks, Suppression Zones, Late-Inning Volatility)
• You interpret CI ranges and sequence projections
• You overlay human insights (Transcripts from youtube videos and other "experts") to contextualize results

This is the interpretive intelligence step—where the data is translated into actionable ideas.
It’s not about accepting the output blindly—it’s about interrogating it.

---

SPECULATE – The Action Phase​

Now you act—whether that means:

• Targeting short-leg, high-confidence plays
• Structuring prop combinations (HR + RBI + Team Win, Race to X Runs, F5 Overs)
• Strategizing series-level plays (bullpen fatigue exploitation, GSSM carryover)
• Timing live market entries based on volatility triggers

This phase is where strategy turns into speculative positioning, powered by simulation-backed probabilities.

---

"1, 2, 3 GO!" – Why We Say It​

Because the simulator is designed for velocity of decision-making.
You don’t get lost in endless toggles or reams of raw data—the flow is:
Run → Review → Act.
It’s engineered to go from data ingestion to confident strategy in minutes, not hours.

---

In practical terms:

Simulate – trust the engine’s computational muscle.
Investigate – apply your human context and tactical mind.
Speculate – place calculated, high-clarity positions in the market.

That’s it.
Three steps. One loop. Repeat as needed.

And if you want to sound cool while doing it—say it with me:
"Simple as 3: Simulate. Investigate. Speculate. 1, 2, 3—GO!"

BIGGEST ONE DAY LOSS TO DATE: $-30
BIGGEST ONE DAY ROI TO DATE: $2,122.56

 

[spajki323]

Hello all - as an update, I have succeeded with the MLB Baseball Simulator model using GPT to run Monte Carlo simulations and other processes that I'm learning more about, but couldn't even begin to understand. If you have GPT, and mybe can string together a word or two, in the form of instructions, you can do this.

The idea is for each sport, a different approach is needed.

For baseball, I've developed metrics that identify key cluster zones - innings of high votality - and players' probability of being associated with those clusters of activity. That's all in a nutshell. Key is backtesting.

I've been able to focus in, with success on speciality markets like TOTAL RUNS, FIRST 5 INNINGS, WINNING MARGIN, HITS, TBS, and a few others. Sometimes, especially with K'S, some strong value can be found. My goal as/is to do as much in as little time. If I had to ask AI to explain it, iw ould be THREE STEPS:
1. SIMULATE. 2. INVESTIGATE. 3.SPECULATE.

That phrase was born from a very intentional design philosophy: BHW MLB BaseBallPrompt (just a really long chatgpt prompt) may be a deep and technical simulation framework, but it had to be operationally simple in its user flow.

Let’s break it down:

---

SIMULATE – The Engine Phase​

This is where the BHW MLB BaseBallPrompt Simulator does what it was built for:

• It ingests full pre-game datasets (lineups, weather, park factors, pitcher form, etc.)
• Runs multi-phase scenario cycles (F3, F5, Full, 7–9 Window)
• Generates metric outputs (GSSM, BEI, DEA, PUI, CDT, BDI)

This phase answers: "What are the likely shapes and volatility zones of this game?"

The beauty is—you hit "run," and the engine executes 5,000+ scenario seeds per phase in seconds. No manual guesswork.

---

INVESTIGATE – The Analyst Phase​

Once the sim outputs, you investigate.

• You cross-check flags (Tier-1 Locks, Suppression Zones, Late-Inning Volatility)
• You interpret CI ranges and sequence projections
• You overlay human insights (Transcripts from youtube videos and other "experts") to contextualize results

This is the interpretive intelligence step—where the data is translated into actionable ideas.
It’s not about accepting the output blindly—it’s about interrogating it.

---

SPECULATE – The Action Phase​

Now you act—whether that means:

• Targeting short-leg, high-confidence plays
• Structuring prop combinations (HR + RBI + Team Win, Race to X Runs, F5 Overs)
• Strategizing series-level plays (bullpen fatigue exploitation, GSSM carryover)
• Timing live market entries based on volatility triggers

This phase is where strategy turns into speculative positioning, powered by simulation-backed probabilities.

---

"1, 2, 3 GO!" – Why We Say It​

Because the simulator is designed for velocity of decision-making.
You don’t get lost in endless toggles or reams of raw data—the flow is:
Run → Review → Act.
It’s engineered to go from data ingestion to confident strategy in minutes, not hours.

---

In practical terms:


That’s it.
Three steps. One loop. Repeat as needed.

And if you want to sound cool while doing it—say it with me:
"Simple as 3: Simulate. Investigate. Speculate. 1, 2, 3—GO!"

BIGGEST ONE DAY LOSS TO DATE: $-30
BIGGEST ONE DAY ROI TO DATE: $2,122.56

How much are you in profit?

 

[NemoTheOne]

So far it is %250, or just under triple roi, with is amazing considering I'm counting in some 'testing' funds from earlier. So, these are the metrics (abbreviated), somewhat explained. The point is, not the names of the various metrics - but how the model LEARNED and derived these correlations.

Let's take today's game slate, MLB 5th August, 2025.​

Alright—class is in session, and today we’re going to talk about the August 5th game slate like it’s a baseball treasure map. Every mark, arrow, and number on this map is a clue, and if you understand the clues, you can tell where the “X” is before the game even starts. We’re going to take the big, techy simulator talk—things like GSSM, BEI, DEA, PUI—and make them easy enough that you can see how they work together, especially for the pitchers.

---

1. GSSM – The Fuse and Firecracker Gauge
Think of GSSM (Game Script Sequencing Metric) like the “when will the fireworks go off?” meter.
If GSSM is high—say, over 1.3—it means there’s a strong chance that runs will explode in a certain part of the game. It’s not saying if the runs will happen, it’s saying when.
Example from today’s slate:

• TOR @ COL has a GSSM of 1.36 in the 3rd–5th inning. That’s like knowing the fuse is lit and will pop before the halfway mark.
• MIN @ DET has its own firecracker at innings 4–5, meaning a pitcher who’s cruising early might suddenly give up a crooked number.

For pitching, this means if your starter is on the mound during the “fuse window,” their ERA can balloon fast. That’s why pitchers with low GSSM danger zones—like Logan Webb (SF)—are safer bets to keep the scoreboard quiet.

---

2. BEI – The Slugging Power Thermometer
BEI (Barrel Expectancy Index) is like checking how hot the bats are. High BEI means hitters are more likely to smash the ball hard—doubles, triples, homers.
Example:

• Bo Bichette and Vladimir Guerrero Jr. have BEI values over 1.6 today in Coors Field. That’s the baseball equivalent of putting fireworks in a microwave—high heat, high danger.
• Bobby Witt Jr. in KC @ BOS? His BEI is down near 0.48 because he’s facing Garrett Crochet, who’s a strikeout machine with stuff that’s hard to barrel.

For pitchers, BEI tells us how risky a lineup is. A low BEI means the pitcher’s margin for error is bigger. A high BEI lineup means even one mistake pitch can wreck the night.

---

3. DEA – The Ballpark Weather Report
DEA (Dynamic Environment Adjuster) is like the “how much does the air help the ball?” score. High DEA means the air, temperature, and park are helping the hitters. Low DEA means it’s like hitting with a wet towel.
Example:

• TOR @ COL is through the roof—DEA 1.35 because of Coors Field, warm air, and wind that helps carry the ball.
• KC @ BOS has DEA 0.82 thanks to wind blowing in and a park that’s shrinking the ball flight.

Pitchers in high DEA environments have to be extra careful. Even good pitches can turn into extra-base hits. Pitchers in low DEA games can pitch more aggressively without worrying as much about cheap home runs.

---

4. PUI – The Jenga Block Index
PUI (Pitcher Unsteadiness Index) is like a stack of Jenga blocks—if the number is high, the tower can wobble and fall any inning.
Example:

• Anthony Molina (COL) has a PUI of 31%. That means he’s already leaning over, and the first three innings are the wobbliest.
• Mike Burrows (PIT) is at 27%, which screams “collapse watch” if his pitch count gets too high.
• Logan Webb (SF)? PUI of 14%. That’s a tower built like concrete—hard to knock over.

High PUI means a pitcher is one bad inning away from losing control. Low PUI means the pitcher can absorb pressure without crumbling.

---

5. CAI – The “Tag Team” Alert
CAI (Convergent Alignment Index) is like knowing which two batters are going to tag-team the same inning to create trouble.
Example:

• SF @ PIT has a CAI of 4.3 in the 5th inning, meaning Chapman and Devers are lined up to hit back-to-back in a danger zone for Burrows.
• STL @ LAD shows Ohtani and Freeman lined up for an 8th-inning burst—exactly when Mikolas is at his most vulnerable.

For pitchers, this tells us when to expect the “storm cloud” in the lineup—two or three tough outs in a row, in the wrong part of the game.

---

Putting It All Together

• If GSSM is high and PUI is high? You’ve got a pitcher standing on a short fuse—runs are coming, and they may come fast. (Example: Burrows in SF @ PIT.)
• If BEI is high and DEA is high? Expect the ball to fly—this is when even average hitters look like sluggers. (Example: Bichette/Guerrero in Coors.)
• If PUI is low and BEI is low? The pitcher has a good shot at cruising—think Logan Webb against a cooled-off Pittsburgh lineup.
• If CAI shows a late-inning storm cloud? Don’t count the game over—those rallies can flip a winner into a loser in one inning. (Example: STL @ LAD 8th inning Ohtani/Freeman.)

Now, if I had to rank the most dangerous situations for pitchers tonight, using all these metrics:

1. Anthony Molina (COL) – High PUI, High BEI against, High DEA. This is the nightmare trifecta.
2. Mike Burrows (PIT) – Collapse watch, mid-game fuse, CAI pairing against him.
3. Miles Mikolas (STL) – Collapse watch plus late-game CAI against LAD’s top hitters.
4. Zebby Matthews (MIN) – Inverted anchor PUI means the 4th–5th inning could get ugly.
5. Any SP in TOR @ COL – Even low PUI can get roasted here thanks to the BEI + DEA combo.

The safest arms by the numbers?

• Logan Webb (SF) – Low PUI, low DEA, high ground-ball rate.
• Garrett Crochet (BOS) – Suppression flag on opposing lineup, low BEI against, weather on his side.

 

[Venusaur]

this journey is written in greek for me, it seems advanced and hard to understand, but if it's working for you, keep going, you have my respect

 

[NemoTheOne]

Yup. Basically l, I'm working with custom measurements used to forecast game outcomes.

Like a video game simulation

 

[ritafairclough]

This is by far my favourite read in the mjt section.

If you ever do some tutorials on how to set up something like this up with AI - I'd be all over it!

 

[NemoTheOne]

This is by far my favourite read in the mjt section.

If you ever do some tutorials on how to set up something like this up with AI - I'd be all over it!

Thanks so much! You meanna tutorial on how to make a game simulator? What would be the title? I'll do it!

 

[ritafairclough]

Thanks so much! You meanna tutorial on how to make a game simulator? What would be the title? I'll do it!

Well, yes. Everything really. For example I'd be interested in creating a system that focuses on one small betting in play element, like the number of yellow cards or corners in Premier League soccer game.

From gathering all the required data sources to building the model with all variables covered. How cool would that be!

 

[BanX]

This is super interesting. Good luck with your journey. "Cricket" is a hugely popular sport that's somewhat similar to baseball. Bettors love it, give it a try when you’re done with this.

 

[Gritz]

Had to get back to BHW just because of this post. To OP, are you saying that your interactions with chatgpt and feeding it proper info over a year brought you this far, or it is all about refining your initial prompt, which then does the 'best speculation' ?

 

[NemoTheOne]

Had to get back to BHW just because of this post. To OP, are you saying that your interactions with chatgpt and feeding it proper info over a year brought you this far, or it is all about refining your initial prompt, which then does the 'best speculation' ?

Thanks so much! Well, I'd say it's been the info provided and dissected over the year that brought things this far. The specific formulas/calculations thats done are specific to that. But this level can be attained quicker with many interactions with chatgpt at any given time.

Or, someone could take what I have, and say "GPT, explain this to me and make something better /different /for this sport etc"

 

[Gritz]

Are you also using some sort of AI utility to fetch Youtube transcripts (summarize them) or you do that manually ?

 

[NemoTheOne]

Are you also using some sort of AI utility to fetch Youtube transcripts (summarize them) or you do that manually ?

I do that manually. Which is actually better because I DON'T want "every" opinion out there filtered into my simulations

 

[NemoTheOne]

Thanks for the interest guys! Next update will be running the simulator + AI AGENTS. Like someone said, what I'm doing here is targeting specific markets (ie FIRST 5 INNINGS), because it seems the simulator/prompt is very good at picking up 'clusters of activity'.





Even Biff can do it.



Let's make a funny HA HA out of it:

 

[Gritz]

What do you say about the latest version 5 release of chatgpt ? Any positive or negative impact, or is it too early to tell ?

 

[NemoTheOne]

What do you say about the latest version 5 release of chatgpt ? Any positive or negative impact, or is it too early to tell ?

Hey, thanks again for your interest, it's so very much appreciated that I'm not the only one that enjoys this unique cross-section of black hat, marketing, analytics, sports, and making money.

So, I think it's GREAT. I'm not sure if it's the vocal minority, or votality of the internet, about how much they miss the old models, but I think it depends on your purpose of usage.

If you want to ask gpt questions like 'roast me' and stuff like that, then the loss of personality is huge. I'm not saying there's anything wrong with that, as I, too used it for some aspects of personal accountability.

But for those using for business purposes - creative really (marketing, advertising, influencer etc), it's really amazing. For content, stats, ideas etc.

I am using AI Agents, and have been for a few years now. The personality is still there, but the over the top, yet strategic 'characters' are lacking. So, again, for fiction writers, it may be difficult.

I have a creative writing prompt (used to evade AI content detectors) that I've yet to try.

For analytical purposes, like what I'm doing here, in this journey, it's quite awesome. It doesn't randomly misread numbers, and it doesn't forget things (as much, me thinks).

####

So, for those of you new to the thread, I started a diary about 1.75 years ago, about speculation, based on my extensive, and quite embarrassingly long hours spent playing casino, simulation, and sports games from the 8-bit, and 16-bit console era.

So, long story short - I've been training for online speculation (fanduel etc) for DECADES. When AI came along, as documented in my first diary (link in first post), I used it for seo, writing, upwork, cpa marketing, blackhat stuff - ya know, the usual. Then they legalised online speculation in my region!

Knowing nothing about 1) online speculation 2) ai (like the rest of the world), I start building 'prompts', or models based on ideas and research found in papers, forums, journals, documented online from various years, 2015 etc, to build, the "AI " versions of those.

That, combined with my own experience with simulation modules from those early sports and business simulation video games. Yeah, business simulation. That's an important element, too. It was like the "BHW" of early strategy games. You would be able to find strategy and simulation games, in your local video game store, in the discount bin.

####
So, now back to your question.

GPT 5 allows my idea of an "AI Speculation Intelligence Agency" tangible, because it's not making the tedious, yet, important miscalculations that would allow for any type of long term positive backtesting results.

Also, I speak of "backtesting" like I'm some type of programmer. GPT-5 also allows for the very meticulous details that writers like myself put in, actually mean something, without knowing ANY programming. So everything that I say, I say, with respect to programmers, who actually know a think or two about Python. To me, it's a reptile.