[Script] BSP - Battle Stats Predictor

This script predicts total battle stats for each Torn player.
Predictions are fast and displayed in a bunch of useful places, almost everywhere actually, using a patented handy color system.
Predictions are based on a machine learning model trained with thousands of real spies.
Predictions can also be overridden by spies (link with TornStats/YATA), so you can see the most reliable information.

Load image

Load image

Chain window : Easy retals with the color system! (click on the colored area for quick attack too!)
[image: i.ibb.co]

The first week of use is free, then it's 1 xan/15 days.
Send to me. Process is automated. You can pay in bulk.
Price might increase in the future as I keep improving it.

Discord is the place to discuss installation / question / improvement / wrong predictions / roadmap or just anything really Load image: party
It has the most up to date information.

Disclaimer : Keep in mind predictions can't be 100% accurate, this script really shines to give you a super quick & relatively precise estimation.

Link to the Script - It requires ViolentMonkey/GreaseMonkey to work, see installation steps below for more info

Video showing how useful it can be in Elimination

PART A - First you need a tool such as Tampermonkey
 It's what injects scripts in the appropriate pages.

• On computer, if you use Chrome, install it directly from Chrome extensions. (don't forget to enable developer mode, in 
• chrome://extensions)
• On computer, if you use Firefox, install Greasemonkey instead
• On Android, use Firefox + Tampermonkey
• On Iphone, you can use TornPDA. 

Step 1: Install the script 
From within the browser having Tampermonkey installed, click on this:
https://github.com/tdup-torn/userscripts/raw/master/battle_stats_predictor.user.js
It should open the script within Tampermonkey, click install.

Step 3: Create a new API key. You can use the button for that .

Step4 : make sure you have this confirmation. Your API key is verified locally, and the BSP backend returned your subscription (1 free week for new user)

Step 5 : Configure the Color tab
It's important as it is what's going to display differently the players in Torn, depending on their stats (predictions or spies)
Generate a key with access to your battlestats, and fill it properly then press Import stats
You could also manually fill your stats if you wish. Note that those stats, or this API Key is not sent to the BSP backend. It's only stored locally on your machine.

Load image

Notice the star, along with the TBS, the star means it comes from a spy, and not a prediction (so it's "reliable", except if you have forged spies in your TornStats spies data!)
Load image

Step 9 : Enjoy, give feedback, and remember that:

- Those predictions are.. predictions. They can't be 100% accurate, so use this tool for what it is, not a cheat but a helper
- If there are issues with my script, let me know politely, it has been a serious hobby for me those past month, but it also takes time to support/fix, please understand I have a life outside of Torn

Step 10 : Advanced knowledge for TornStats spy integration
When a spy is older than the setting you have here (older than 30 days by default) .. a prediction is made. If the predicted tbs is lower than your old spy tbs, the old spy tbs is used instead. along with a specific symbol : a floppy disk [image: discord.com] ( [image: discord.com] btw, feel free to suggest another icon :p)

API Terms of Service (ToS) Guidelines (Updated 2025 May 15th)

- Data Storage -
Data stored (Persistent - forever, you can ask me to manually delete)
BSP User Player ID, BSP User Public API key, predictions computed for other Torn players.
Public HOF stats from the public thread

- Data Sharing - 
Predictions are shared between all BSP users (Service owners & their customers)
Public HOF stats from the public thread

- Purpose of Use -
QoL to display battle stats gathered through spies, through public forum, through predictions

- Key Storage & Sharing -

Stored / Only TDup has access to the database / API keys are not shared. API key is used only when userscript triggers a query. (ie : there's no automated feature using a pooling of API keys)

- Key Access Level -
BSP allows user to set multiple keys, so that only the minimum required is sent to BSP server.

[Normal usage: Public] This API key is sent to BSP server and stored in BSP server database.

[Special usage - Custom API Key: battlestats] For BSP users choosing to auto import their Battle stats in the tool (for stats comparison coloring). This key is NOT sent to BSP server.

[Special usage - Custom API Key: attacks,battlestats] For BSP users choosing to opt-in to feature reverse computing FF from recent stats 
(this API key is NOT stored on the server database, battle stats of the user are NOT saved and NOT shared)

[Special] For TornStats/Yata spies integration: API keys that fit those systems. Those API keys are NOT sent to BSP server.

All those API keys are independant and saved separately on the client cache.
BSP encourages through the UX to generate a different key for each usage, with proper permission on each.

BSP flowchart (as of 2025 May 15th)
[image: i.ibb.co]

Q&A

- How do I pay?
Send 1 or more xanax to TDup [2660552]
Each xanax gives 15 days worth of subscription. The process is automated so you don't have to wait for me to be online. Payment is processed within 1 minute.

- Is it legal?
Yes. It uses only API data, and an external service I created & trained with legit spies.
Also, I recently added protection so it doesn't do any API queries on pages you don't have the focus on.

- What do you mean "price might increase in the future"?
It's free for one week starting whenever you want (trying to hook you up so you can see why it's useful), then it's 1 xan / 15 days.
I received tons of feedback saying "it's way too cheap", but I want a broad adoption and young players to be able to use it.
I plan to keep working on this script for a while, and it's very likely to improve significantly (in UX, optimizations and more importantly : accuracy), and thus why price might increase in the future.
Note that if you subscribe for like 3 months right now, you obviously won't have to pay for a potential raise for this period.

- Can you predict all kind of TBS?
On March 2024, I pushed a new model that can better predict any kind of targets.
From weak players to strong ones.
For the megazord (>250SE), I'm not relying on a model, but from a simple math function, using SE as input. It's cheap but doesn't the job ok-ish.

- But, Ched changed the way gym gains are computed!
About that.. Yeah, It's a bit annoying. My current prediction model will indeed be offset with time, and predictions will be off especially for high TBS players (prediction lower than reality)
It's going to take time though, and getting new spies frequently / updating my model will mitigate this issue. I'm not too worried about that.

- But, TBS are irrelevant because balanced build and whore build are not similar
True, especially below 50m stat cap. I actually based my prediction model on the Battle Stats Score information
Problem is: normal players don't get the meaning of Battle Stats Score, so it makes little sense to display it to them.

- What is the API keys rate limit?
For now, I've set a hard lock at 50 calls/minutes per user. (max allowed by Torn is 100 calls/minute)
That being said, I've implemented different caches to reduce the required number of API calls (predictions are saved server side, and shared among users of this script with a 1 day timeout for example)

- I've heard about API keys being stolen and used to mug me!
I'm taking this topic really seriously and don't want any issue ever. I actually designed my previous project to avoid any possibility of API key breach, and there has been 0 complain ever.
I've built BSP so it can use different API Keys. The only API key sent to my backend is the global one, which requires no specific right (= I can't see your activity, your battle stats, your money).
The API key you can use in the "Color" section, is here to automate the process of getting your stats locally, to compare with other players. This key, or the stats are not sent to my backend.
Yata/TornStats integration : same thing: it's done on the client and not sent to my backend.
You can change your API key whenever you want.
You can also inspect the client (javascript) code (or ask a fellow developer player to do so), I didn't minify it for this very reason.

- What about the "upload data" thing? It sends my battle stats right?
The opt-in Upload data option (which is disabled by default), is a way to help BSP predicts players you faced. It relies on the well known way to reverse compute the Battle Score, from the result of a fight (and the FairFight value).
When you enable this option, you get credited 3 months of BSP subscription right away. (a way to recompense those turning that on).
You can turn that off whenever you want.
When you press the button, or once a day if you have the checkbox ticked, it will send your API key to BSP server, that will then
- Fetch your Battle stats.
- Fetch your last attacks (BSP only consider the most recent ones, ie : less than 48h old)
- It will then compute the Battle score of your opponents (both the ones you attacked, or attacked you)
- It will then store those battle score in the database. Those battle score will be shared with any other BSP user (even those who don't turn this opt-in feature) - The gray star display means the prediction comes this.
- It won't save your own battle score, so it won't share your battle score with other BSP users.
- It won't save that API key

- You mentioned Machine Learning, I'd love to know more!
Here's the current state:

The new model is based on 57k unique data points. Coming from spies, or from BScore calculated through faction attacks.
I'm still relying on a 3 layers approach.

• If a player has 250SE+, instead of the really bad model I previously had, I now run a simple math function to predict TBS based on SE alone (capped at 1q). It's not ideal but it's a nice improvement.
• Otherwise player gets bscore predicted through the main model, with those features: Attacks, Useractivity, Awards, Xantaken , Daysbeendonator , Refills , Statenhancersused, Revives, Lsdtaken ,Boostersused, Energydrinkused,Agenoninactive

If the player has <4000 bscore, then it goes through another model ("TheWeakModel") which has been trained with weak players only (22k data point) - I have a really better accuracy like thisBscore is then transformed into TBS, and increased a little bit, because most players aren't perfectly balanced.

If you want more info, join the Discord!

- Be honest, you have the Thief Title, you're just trying to mug me!
Nice! You either read until here (and it was long, congratz), or you noticed the Thief Title.
I spend a lot of time scripting.
I've developed a tool to know about Leslie Wheels, but outside the 2 DPs Sugarvalves sent me to appear in the newspaper, I made $0 out of a lot of hours programming. "Do it for the Science" they said Load image: party
So I've a bazaar checker script sometimes running on the side, and I mug lazy players who don't self hosp when they sell expensive stuff :D
(Fun fact, this prediction tool was first created as a feature for this bazaar checker, to filter out players I couldn't beat!)
Needless to say this mugging script runs with my own key and has 0 link with this. You can always check your key usage.

My plan is actually to make $$$ with this Predictor tool, because it scales (as opposed to some occasional mugs) and it's really useful for everyone!