V1.3 Live · Windows Desktop

Historical probability
for every trade setup.

Know Your Odds Before You Trade

Define your setup. Prizm scans thousands of historical candles using a KNN engine, finds analogous market states, and returns a calibrated probability estimate — visualized as a live 8×8 heatmap in ~2.5s.

<6%
Calibration error
17,896
WF predictions
~2.5s
Heatmap render
PRIZM ENGINE — INTERACTIVE DEMO
BTC-USD · 1H · LONG
62%
TP hit probability
TP 2.0% / SL 1.0%
Selected cell
0%
100%

The Engine

How Prizm Works

A four-stage pipeline from raw market state to calibrated probability — no black boxes, no hand-waving.

01 / INGEST
Define Your Setup
Select asset, timeframe, direction, and TP/SL levels. Supports 10 assets across crypto and equities, 5 timeframes from 5M to 1D.
02 / ENCODE
18-Dimension Signal Vector
The current market state is encoded into 18 proprietary signal dimensions — momentum, volatility, structure, and regime indicators.
03 / SEARCH
KNN Historical Match
K-Nearest Neighbors scans thousands of historical candles to find analogous states. Matched setups are resolved against your exact TP/SL.
04 / OUTPUT
8×8 Probability Heatmap
64 TP/SL combinations simulate in ~2.5s. Each cell shows the historical win rate, giving you the full probability surface at a glance.
10
Assets supported
5
Timeframes (5M → 1D)
64
Heatmap cells / query
18
Signal dimensions

Methodology

Walk-Forward Validated

Prizm's probabilities are never backtested on data the model has already seen. The engine predicts, checks, and moves forward — every time.

<6%
Calibration error
17,896
Walk-forward predictions
3
Validated assets (BTC, SPY, QQQ)
What is walk-forward validation?
Most trading tools backtest — they run a strategy on data the model has already seen. Walk-forward validation trains on past data, predicts the next candle, checks if it was right, then slides the window forward. The model never sees future data before predicting. It's the difference between memorizing a test and actually passing one you've never seen.
What does calibration error mean?
If Prizm says a setup has a 60% probability of hitting TP, the actual win rate across all similar predictions should be close to 60%. Calibration error measures that gap. Under 6% means a 60% prediction corresponds to an actual outcome between roughly 54–66%. Measured across 17,896 predictions on BTC-USD, SPY, and QQQ.
Validation scope
Walk-forward calibration has been completed on 3 assets: BTC-USD, SPY, and QQQ — covering both crypto and traditional equities across multiple timeframes. The engine supports 10 assets total. Expanded validation across all 10 assets with tick-level resolution is currently in progress.

Pricing

Know Your Odds

Less than a single losing trade. All plans include the desktop app and live backend.

Scout
$0/yr
Free forever
  • 3 queries per day
  • 3 assets (BTC, SPY, QQQ)
  • 1H timeframe
  • Full heatmap locked
  • All timeframes locked
  • All 10 assets locked
Download Free
Pro
$149/yr
$12.42/mo · save 35%
  • Unlimited queries
  • All 10 assets
  • All 5 timeframes
  • Full 8×8 heatmap
  • Danger zone warnings
  • Optimal TP/SL overlay
Get Pro
Coming Soon
In development
Elite
$299/yr
$24.92/mo · save 36%
  • Everything in Pro
  • Trade journal
  • Email alerts
  • API access (Beta)
  • Founder support
  • Priority requests
Get Elite
All plans · Windows desktop app · Live backend at prizmtrading.com Billing: support@prizmtrading.com
Download

Start free. No card required.

Download the Prizm Trading desktop app, define your first setup, and get a calibrated probability estimate in under 3 seconds.

Download Prizm Trading

Windows installer · GitHub Releases · v1.3

FAQ

Common Questions

What exactly does Prizm tell me?+
Prizm tells you what percentage of historical setups similar to yours resulted in take profit being hit before stop loss. It's a calibrated historical win rate — not a prediction, not a signal, not financial advice. Quantified context for a decision you're already making.
How is this different from a backtest?+
A backtest runs a strategy on data the model has already seen — prone to overfitting. Prizm uses walk-forward validation: train, predict, check, slide forward. The engine never sees future data before predicting. Validated across 17,896 predictions on BTC-USD, SPY, and QQQ.
What is calibration error and why does it matter?+
If Prizm says 60% probability, the actual historical win rate should be close to 60%. Under 6% means a 60% prediction corresponds to an actual outcome between ~54–66%. Measured across 17,896 walk-forward predictions on validated assets.
What assets and timeframes are supported?+
10 assets across crypto and equities. Walk-forward calibration completed on BTC-USD, SPY, QQQ — expanded validation in progress. Five timeframes: 5M, 15M, 1H, 4H, 1D. Scout plan is limited to 3 assets and 1H only.
What platform does Prizm run on?+
Windows desktop app built on Electron, connecting live to a FastAPI/Python backend at prizmtrading.com. Mac and Linux not available in V1.3. Download the installer from GitHub Releases.
Is this financial advice?+
No. Prizm provides historical probability estimates based on pattern matching. Past performance does not guarantee future results. All trading involves risk. Users are responsible for their own decisions.