PRFX +227.01% CDT +98.94% REPL +84.72% VCIG +84.63% STG +76.47% IOTR +72.85% ASTC +72.48% OLOX +69.92% OKTG +59.29% DLLL +57.97% CMND +53.06% YMAT +43.14% SPCE +39.07% NEXR +38.36% ADVB +37.60% PRFX +227.01% CDT +98.94% REPL +84.72% VCIG +84.63% STG +76.47% IOTR +72.85% ASTC +72.48% OLOX +69.92% OKTG +59.29% DLLL +57.97% CMND +53.06% YMAT +43.14% SPCE +39.07% NEXR +38.36% ADVB +37.60%
Live — 80.3% accuracy

Your edge
before the bell.

AI scans 12,000+ stocks twice each morning — at 6 AM and 8 AM ET, before the bell. 21 signals. 80.3% accuracy. Strong Buy to Watch, every trading day.

Get Started View Dashboard
53.3%
Win Rate
$-106
P&L (28 days)
46.2%
ML Filter Win
0
Back-to-Back Loss
3 TRADING DAYS / $1K PER POSITION / TOP 5 PICKS PER DAY / LIVE FROM DB
The Engine Room

What powers Prismo

Six systems working together, every trading day.

Chapter 01 · Machine Learning

Gradient-boosted trees, tuned weekly.

21 engineered features feed a GBT classifier whose only job is to score the probability of a continuation. The model retrains every weekend on the freshly labeled ground truth from the past five trading days. No manual recalibration. No drift compounding.

80.3%
Out-of-sample accuracy
21
Features
Weekly
Retrain cadence
Chapter 02 · Prediction Engine

Six multipliers, institutionally calibrated.

After the score lands, the prediction engine layers six independent multipliers — gap momentum, dollar volume, float rotation, ATR ratio, history of follow-through, and catalyst posture — to derive expected price, stop-loss, targets, and a re-entry zone. Each multiplier is calibrated against historical reality, not academic theory.

6
Independent multipliers
  • Gap momentum×curve
  • Dollar volume×liquidity
  • Float rotation×supply
  • Self-calibrationlive
Chapter 03 · Pre-Market Watch

Three-and-a-half hours before the bell.

At six in the morning Eastern, while the rest of the market is still asleep, Prismo scores yesterday's movers for continuation probability and surfaces three priority bands — HOT, WARM, MONITOR — well before the first opening tick. The edge is temporal: by the time the floor opens, you've already chosen.

06:00 ET
Daily pre-market scan
HOT
≥ 70
WARM
45–69
MONITOR
< 45
Chapter 04 · Action Signals

Four signals, six factors each.

Score, expected gain, risk-to-reward, volume confirmation, ML confidence, ticker follow-through history. Each combines into a single verdict — Strong Buy, Buy, Hold, or Watch — that you can act on without reading the fine print. The fine print is, of course, available below.

Strong Buy≥ 13
Buy10–12
Hold7–9
Watch< 7
Chapter 05 · Accuracy

Every call is tracked. No exceptions.

Every prediction is reconciled against the next day's close — per ticker, per signal, per gap-size bucket. The accuracy dashboard exposes the win rate, average error, and signal-by-signal breakdown so you can see exactly where the model excels and where it doesn't.

71.5%
Direction-hit rate
414
Trades validated
$8,212
Backtested P&L
Chapter 06 · The Pipeline

Four cron jobs. No human in the loop.

Pre-market watch at six. Main scan at eight. Intraday top-mover refresh at two. Accuracy reconciliation at four-thirty. Five days a week, every week, US holidays excluded. By the time you wake up, the day's verdict is already typeset.

  1. 06:00
    Pre-market watch
    Continuation scoring
  2. 08:00
    Main scan
    12,000+ tickers ingested
  3. 14:00
    Top-mover refresh
    Intraday outliers
  4. 16:30
    Accuracy tracking
    Yesterday reconciled
01 / 06
A Day In Four Acts

How it works

From raw market data at dawn to a verdict before the bell.

Act I · Ingest

Twelve thousand tickers, before sunrise.

Polygon-grade aggregates flow into the engine before the floor opens. Volume profiles, gap analysis against the prior close, momentum signatures across the prior twenty trading days. Every US-listed equity above the liquidity floor enters the funnel.

12,000+
Tickers ingested daily
Volume profileslive
Gap analysisvs prev close
Momentum window20d
Act II · Quantify

Twenty-one features through one tree.

Each candidate is reduced to a 21-dimensional vector — gap percent, relative volume, ATR ratio, sector rotation, follow-through history, and seventeen others — and passed through a gradient-boosted classifier. The output is a probability, between zero and one, that today becomes tomorrow's headline.

21
Engineered features
Core
5 features
Volume
5 features
Momentum
5 features
Intel
6 features
Act III · Project

The probability becomes a trade plan.

An expected price. A stop-loss tuned to the volatility regime. Two targets — the realistic one and the stretch. A re-entry zone if the morning fades and the afternoon recovers. Every number is a working hypothesis, calibrated against the history of stocks that looked exactly like this one.

  • Expected pricecomputed
  • Stop lossATR-tuned
  • Target 1conservative
  • Target 2stretch
  • Re-entrypullback zone
Act IV · Reconcile

Every verdict, eventually, is graded.

After the close, every alert is reconciled against the day it predicted. The label — accurate, inaccurate — feeds Saturday's retraining batch. The model that wakes up Monday morning is, by definition, the most-informed version of itself that has ever existed.

Saturday
Weekly retrain cadence
414
Trades labeled
71.5%
Validated win rate
01 / 04
The Signal Set

Twenty-one features.
Four categories.

Every signal earns its place by predictive lift on out-of-sample tests. The primary feature, the catalyst that defines a gap-and-go:

Primary signal 46.3%
Gap-Up %

Opening overshoot · the catalyst

Core 5
  • 01Gap-Up %
    opening overshoot
  • 02Relative Volume
    vs 20-day avg
  • 0352W Proximity
    distance from year high
  • 04Price Level
    liquidity band
  • 05Market Cap
    size & float bias
Volume 5
  • 06Vol Acceleration
    5d vs 20d trend
  • 07Float Rotation
    vol vs share float
  • 08ATR Gap Ratio
    gap vs daily range
  • 09Sector Gap Count
    peer rotation
  • 10Sector Avg Gap
    sector momentum
Momentum 5
  • 115-Day Change
    short trajectory
  • 1210-Day Trend
    2-week bias
  • 13Prev Day Range
    yesterday's spread
  • 1452W Low Dist.
    runway above floor
  • 1552W Breakout
    above year high
Intelligence 6
  • 16Follow-Through
    ticker hit rate
  • 17Alert Count
    prior appearances
  • 18Avg Day Change
    typical alerted move
  • 19Day-of-Week
    weekly seasonality
  • 20Monday Flag
    post-weekend regime
  • 21Friday Flag
    pre-weekend de-risk
Under The Hood

The whole machine, in motion.

From dawn data to a verdict before the bell — every piece of the engine, in one continuous reel. Hover to pause.

01
The Scan

Twelve thousand tickers, before sunrise.

Every trading morning, Prismo pulls grouped daily bars for 12,000+ US equities via Polygon.io and compares them against the prior session to find gap-ups — stocks that opened materially higher than yesterday's close.

An automatic fallback chain tries three data strategies in sequence: all-tickers snapshot, gainers endpoint, grouped daily comparison. Scans run reliably regardless of API tier.

Polygon.io 12,000+ symbols
02
The Score

A hybrid scorer, tilting toward the model.

Each candidate is scored 0–100 by a hybrid engine: a rule-based static scorer provides the baseline, while a Gradient Boosted Trees classifier overlays probability-weighted predictions. As the model proves itself, its influence ramps from 0% to 70% via an adaptive alpha.

Bonus points for volume-gap synergy, dollar volume, 52-week breakouts, and sector-wide momentum.

Adaptive alpha 0 → 0.7 Hybrid · GBT
03
Self-Learning

A model that upgrades itself.

After every close, the accuracy tracker reconciles each alert against actual closing prices and writes a label — win or loss — onto a 21-feature vector. Saturday's batch retrains on the freshly accumulated ground truth.

The model class auto-progresses with sample size: Logistic Regression at 50, Gradient Boosted Trees at 200, an Ensemble at 500+.

Currently 414 labeled · 80.3% acc Saturday retrain
04
Prediction Engine

Six multipliers, institutionally calibrated.

Expected price isn't a guess — it's a function of institutional-grade momentum curves. Decades of market data say 5–10% gaps follow through ~55% of the time; 10–20% gaps, ~65% with +8–15% avg continuation.

Six multipliers refine each forecast: volume confirmation, dollar volume, float rotation, ATR-relative gap, ticker history, and score-based confidence.

Self-calibrating 6 multipliers
05
Top Indicators

The features doing the heavy lifting.

Out of twenty-one engineered features, six explain most of the model's lift. Each weight below is the model's relative reliance on that feature.

  • Gap-Up %46.3%
  • Price Level13.7%
  • ATR-Relative Gap12.4%
  • Prev Day Range11.3%
  • Volume Acceleration6.4%
  • Relative Volume4.8%
06
Stop-Loss

Tuned to the stock's own volatility.

A stop-loss isn't a fixed percentage — it's set from ATR (Average True Range): at least 1.5× the daily range below entry, capped at 8%. The stop respects the volatility regime the stock is already living in, instead of imposing one.

Wide-range stocks get wide stops. Quiet stocks get tight ones. Either way, the trade isn't stopped out by ordinary noise.

≥ 1.5× ATR · capped 8% Volatility-aware
07
Target Levels

A realistic close, and a stretch.

Two targets per trade: a conservative target at the expected close from the momentum curve, and a stretch target at 1.8× the expected gain — capped at the gap-range upside.

As the engine accumulates outcomes, self-calibration blends 60% model with 40% historical actuals per score-bucket. Predictions get sharper without manual tuning.

60/40 blend Two-stage
08
Re-Entry Zone

The second-chance price level.

If a stock gaps up $10 and pulls back $4, that 40% retracement is statistically the cleanest re-entry. Momentum stocks that hold their re-entry zones typically resume the move; the ones that don't, weren't really momentum trades.

The re-entry is published with every alert — so a missed open isn't a missed trade.

40% gap retracement Tactical
09
The Pipeline

Four cron jobs, five days a week.

06:00 — pre-market continuation watch. 08:00 — main 12,000-ticker scan. 14:00 — intraday top-mover refresh. 16:30 — accuracy tracking against the close.

Mon–Fri, US holidays excluded. By the time you wake up, the day's verdict is typeset.

Zero manual intervention Mon–Fri
10
By The Numbers

A platform that grades itself.

Every prediction is reconciled. Every win and loss is logged. Every weekend, the model is rebuilt against the new ground truth. Below: the current state of play.

ML accuracy
80.3%
Validated win rate
71.5%
Trades reconciled
414
Backtested P&L
$8,212
01 / 10

Plans

Prismo Free
$0
1 alerts
Prismo Starter
$59
3 alerts Telegram
Popular
Prismo Pro
$99
15 alerts Telegram
Prismo Elite
$159
Unlimited alerts Telegram WhatsApp
Compare plans →

Frequently asked questions

What exactly does Surgepulse Prismo do?
Prismo scans 12,000+ US stocks every trading morning using machine learning. It identifies stocks that are gapping up with unusual volume — the #1 predictor of intraday surges. You get ranked alerts with expected price targets, stop-loss levels, and AI confidence scores before the market opens.
How accurate are the predictions?
Our ML model achieves 80.3% accuracy on out-of-sample data (data the model never saw during training). Across 414 backtested trades over 28 trading days, the system produced a 71.5% win rate with a 2.6:1 win/loss ratio. The ML filter alone boosts win rate to 84% by filtering out low-confidence signals.
Is this financial advice?
No. Surgepulse Prismo is an informational tool, not a financial advisor. All alerts, predictions, and scores are for educational purposes only. Past performance does not guarantee future results. You are solely responsible for your own trading decisions. Always consult a qualified financial advisor.
How does the AI learn and improve?
Every trading day creates a feedback loop. Morning: the scanner generates alerts with 21 engineered features. Afternoon: actual closing prices label each alert as a win or loss. Weekly: the Gradient Boosted Trees model retrains on all accumulated data. The model auto-upgrades from Logistic Regression (50 samples) to GBT (200+) to a full Ensemble (500+) as data grows.
What data source do you use?
All market data comes from Polygon.io — the same institutional-grade data provider used by hedge funds and fintech platforms. We pull grouped daily OHLCV bars for every US equity, compute relative volume, gap analysis, momentum indicators, and sector rotation signals from this data.
What stocks does Prismo focus on?
Mid and small-cap US equities — $300M to $10B market cap, priced between $2 and $500. This is the sweet spot where surges are most explosive and most detectable. Large-caps move too slowly; micro-caps are too illiquid. The scanner filters for stocks with genuine institutional interest via dollar volume analysis.
What's the difference between plans?
Starter ($29/mo) gives you 5 alerts per scan with scores and signal breakdowns. Pro ($49/mo) unlocks 10 alerts, AI confidence scores, and Telegram notifications. Elite ($79/mo) gives unlimited alerts plus WhatsApp and priority support. All plans include expected price targets, stop-loss levels, and historical accuracy data.
Can I get alerts on my phone?
Yes. Pro plans include Telegram alerts — you'll get a message with your top surge candidates every morning before market open. Elite plans add WhatsApp notifications. You can set a minimum score threshold so you only get notified about high-conviction signals.

The next surge is forming.

71.5% win rate. 414 validated trades. Start detecting surges today.

Create Account