Kwaker: forum

I love to see developers developing great apps for Torn! Good job. Keep it up!

You're right that absolute capital itself isn't relevant when comparing percentage returns. What I meant is that someone's final % can still differ based on position sizing, allocation, which signals they act on, and execution.

And I understand what you're asking. You're curious what % actual TornX users make over a longer period. Once people have actually used TornX for that long, those results can be compared.

For now, the 7-day trial lets you judge the part TornX actually controls: the predictions and signals made before the market outcome is known.

That's a fair way to look at it, but you're mixing your personal trading performance with predictor performance.

If you already make 2.5% per month consistently on your own, then TornX only has value to you if it improves your decisions enough to justify using it. I agree with that.

What I'm not going to do is invent a monthly % and pretend every TornX user will make the same return. Results depend on capital, position sizing, which signals someone follows, timing and execution.

TornX is meant to be judged on whether its predictions and signals add useful information before the market moves. That's why the trial lets you compare its prediction → window → outcome cycles directly against what you would have done yourself.

If after properly testing it you perform better without TornX, then there's obviously no reason for you personally to use it.

Just to clarify, TornX has never claimed or targeted 0.5% per month. I'm not sure where that number is coming from.

TornX predicts the next market state and provides signals based on the expected movement. It does not promise a fixed percentage return per trade or per month.

I'm also interested in hearing users' results after they've completed multiple prediction windows, which is exactly what the 7-day trial is for.

Just to correct a few things here, because the question was directed at me and some of what you're saying is not how TornX works.

TornX does not predict “a market state every 24 hours.” It predicts the next distinct market state and estimates the window in which that state is expected to occur. Sometimes that window can be around 24 hours, but it is not a fixed 24-hour prediction interval. (Read our help guide)

The 0.1% Torn selling fee is also already accounted for. TornX separates predicted direction from trade eligibility, which is exactly why stocks are marked Fee-cleared or Below fee edge.

The window you're referring to was also a low-confidence case with the Selective tier on ABSTAIN. That means TornX itself was explicitly not classifying that situation as a high-confidence Selective prediction. Judging the entire system from one such window and saying “the predictions failed” is therefore not a fair representation of how TornX is meant to be evaluated.

For entries and exits, the signal is actionable when issued. The prediction window tells you roughly when that expected state should be evaluated. After that, you check the next prediction again because the target and signal can update. SELL / AVOID is the exit warning.

The 7-day trial exists specifically so people can follow multiple prediction → window → outcome cycles instead of drawing conclusions from a single low-confidence window.

If anyone has questions about how TornX actually works, feel free to ask me directly. I'd rather explain the system myself than have incorrect information posted on my behalf.

Thanks for reporting this. You were right, the 500ms portfolio refresh was not intended.

I found a rare edge case in the refresh scheduler that only occurred under certain conditions, which could cause an older portfolio timestamp to trigger immediate repeated refreshes instead of waiting for the normal interval.

That bug has now been fixed. Portfolio data will no longer continuously refresh like that.

Appreciate you catching it, since it helped uncover an issue that hadn't shown up for the other users I tested with. If you decide to try TornX again later, this specific problem should no longer occur.

It's closer to a custom statistical / machine-learning prediction engine than a fuzzy or language-model system.

TornX looks at historical 15-minute snapshots of all 35 stocks together and compresses repeated market conditions into distinct market states. It then models how those states tend to transition and uses the current market state to estimate the next one, including the expected prices and approximate timing window.

The live signal generation itself is rule-driven around that model output. Confidence, expected movement and the 0.1% selling fee are then applied to decide whether something is actionable, fee-cleared, or should be ignored/abstained from.

So there is learned/modelled behaviour behind the prediction, but the live output is not an LLM making subjective guesses. The exact weighting, thresholds and calculations are the proprietary part I keep private.

Just to correct one thing here: TornX does not miss the 0.1% Torn selling fee. It is already calculated into the trade eligibility of every stock. The interface explicitly marks predictions as Fee-cleared or Below fee edge.

A BUY/SELL signal describes the predicted direction of the next distinct market state. It does not automatically mean every predicted move is worth trading after fees. That's exactly what the fee-cleared indicator is there for.

You're also still evaluating predictions before several of their predicted windows have resolved, which is why I gave you the full 7-day trial in the first place. I'm completely fine with you posting your results after the seven days, but I'd prefer the review to be based on completed prediction → window → outcome cycles and on how TornX actually works.

As for recommending another stock service inside the TornX thread, you're obviously free to have your preference, but I'd appreciate keeping this thread focused on TornX and its actual results.

Sounds good haha. I've activated your 7-day trial, so you can compare TornX directly against your own model. Once the trial is over, I'd genuinely appreciate your honest feedback; good or bad. I'm especially interested in how the predictions and signals compare with what your model produces.

You're welcome!

We're talking about two different things. I can't tell you what % I'll make in August, September or October before those months happen, and posting my personal monthly returns afterward still wouldn't be a clean measure of TornX. My results would depend on my capital, position sizing, which signals I act on, and my own execution.

TornX provides forecasts, action signals and an estimated next-state window. It predicts upcoming market states; it isn't a managed portfolio or a promise that users will reproduce my personal returns.

That's exactly why the 7-day trial exists: you can see the predictions and signals before the outcomes are known and judge TornX directly against the live market. If you'd rather wait and watch, that's completely fine, but my personal monthly portfolio performance is not what the service is designed to prove.

Fair question, but I think two different things are being mixed together here.

My Torn account being 11 days old only means I have 11 days of live history on this account. It does not mean the predictor was created 11 days ago or tested on only 11 days of data.

TornX works by taking consecutive 15-minute full-market snapshots, grouping unchanged periods into distinct market states, estimating when the next state change is likely, and then forecasting all 35 stock prices for that next distinct state. Those forecasts are converted into BUY / SELL / AVOID / HOLD signals only when the expected move clears the selling fee and the confidence conditions are met.

I am deliberately not publishing the exact internal rules/parameters behind those steps, because that is the part I am selling.

The important part is how it was tested. I tested the model chronologically on older Torn market data: it predicts later data using only information that would have been available before that point, rather than fitting a model and then judging it on the same data it learned from. That is much more useful for evaluating the algorithm than showing an 11-day personal portfolio.

I also can't honestly give you my "monthly percentage for the next year" because nobody knows what their future monthly return will be. Any number I gave you would just be made up.

What I can provide is the actual product and let you judge its forecasts going forward. That's exactly why I'm offering the 7-day trial. You can watch the predictions, timestamps and signals in real time and decide for yourself whether it provides value before paying anything.

I'm completely fine with skepticism, I'd rather someone test it than take my word for it.

Send me a message to claim your free 7-day trial.

https://www.torn.com/forums.php#/p=threads&t=16590536

I have posted it, thanks!

-- Can a moderator grave this thread due to moving to the proper channel? <3

Hello! I love this new mafia game!