How the SST AI Stock Scanner Works
Most trading platforms hide their methodology behind marketing copy and a paywall. This page is the opposite. Every component of the SST AI stock scanner — the agents, the genome, the scoring model, the risk envelopes — documented in plain English, in public, so you can decide for yourself whether the architecture is worth your subscription.
The Six-Agent Multi-Agent Trading Pipeline
The SST AI stock scanner is not a single model spitting out tickers. It is a coordinated team of six autonomous agents, each with a narrow job description, passing structured findings to the next stage like a trading desk would — if that desk never got tired, never had an ego, and never forgot what worked last week. The pipeline is sequential by design: a candidate ticker cannot reach the execution layer unless every agent has signed off on it. That structural friction is the entire point. Most retail signal services skip this step and pay for it in drawdown.
The agents do not share a model. They share a contract. Scout produces scan candidates with raw market data attached. Analyst takes those candidates and computes ranked technical scores, sentiment estimates, and structural statistics. Researcher fetches deeper context — volume profile, float, catalyst confirmation, sector regime. Debunker tries to kill the trade thesis, surfacing dilution history, earnings traps, and any reason the candidate should be suppressed. Monitor watches the active positions and the broader market health in real time. Brains aggregates the entire pipeline output, applies the evolved signal weights, and emits the scored decision.
Scout — Market Intelligence
Continuously scans the IBKR small-cap universe and the broader equities tape for price, volume, and float patterns that match the genome's current criteria. SuperNova Indicator integration catches high-of-day breakouts in real time. Scan interval is configurable; default is five minutes during the trading window.
Analyst — Statistical Layer
Computes multi-timeframe technicals on every Scout candidate. Digital metrics (win rates, ratios, percentages) and analog metrics (trend, momentum, pattern recognition) are combined into a composite opportunity score. Operates at a level no discretionary trader can sustain across a full session.
Researcher — Deep Context
Pulls volume profile, float data, sector positioning, and catalyst verification for every ticker that survives Analyst. Cross-references against the genome's historical signal-performance database to confirm the setup matches a configuration the engine has seen pay off before.
Debunker — Counter-Analysis
The skeptic. Actively tries to disprove every trade thesis. Scans for dilution history, recent share issuance, earnings traps, sympathy-play risk, and known overfit signal combinations. Issues suppression orders that override the rest of the pipeline when a hidden risk is detected.
Monitor — Real-Time Watchdog
Watches every active position alongside the broader market regime. Tracks profit targets, stop levels, time-based exits, sector rotation, VIX shifts, and thread health across the engine. An internal agent watchdog auto-restarts any dead background process within sixty seconds.
Brains — Decision & Learning
The genome lives here. Aggregates the entire upstream pipeline, applies 96+ evolved signal weights, runs the 17-check Quality Gate (with four hard blocks), and emits the final scored decision — including position size and entry parameters. Also the agent that learns from closed trades and updates the weights.
The 96-Signal Self-Evolving Genome
Every static rule-based scanner has the same fatal flaw: the market shifts and the rules do not. A momentum filter that printed money in February prints losses in August because the regime changed and the threshold did not. The SST genome was built to solve exactly this problem. Rather than hard-coding a single set of rules, the engine maintains a population of weighted signals — a self-evolving trading algorithm that adapts continuously from real outcomes.
A signal weight is the engine's current estimate of how informative a specific market observation is right now. Volume surge might carry a weight of 2.5 because it has paid off in recent trades. RSI divergence might sit at 0.2 because the live results have been noisy. Brains aggregates the active weights at scoring time, applies them to the observed signals on each candidate, and produces a composite score. When a trade closes, the genome compares the prediction to the realized outcome and nudges the contributing weights up or down based on whether they helped or hurt. That is the evolution loop.
Crucially, the genome only learns from real fills. Paper trades, backtests, and shadow simulations are used for diagnostic visibility but they do not modify the live weights. This is the lock-except-live discipline: the only data allowed to change the production genome is data that came from actual orders, actual slippage, and actual commissions. Backtest-overfitting is a documented failure mode of every signal-evolution system — this protocol is the structural defense against it.
The genome's generation counter has crossed 1.6 million as of the April 2026 valuation snapshot, with 569,000+ Monte Carlo simulations layered on top and 15.6 million+ raw signals processed. Those numbers are not vanity metrics; they reflect the depth of the population the live trading engine is selecting from each cycle.
8-Factor Prediction Scoring with Monte Carlo Simulation
Every published prediction passes through an eight-factor scoring stack before it surfaces in a watchlist or alert. The factors were selected because each one captures a category of risk or signal that the others cannot. A high genome score alone is not enough to publish a pick. The setup must clear the full eight-factor gate first, including the Monte Carlo simulation step that quantifies probability of profit and expected value across one thousand simulated price paths.
Multi-Signal Confluence
RSI, MACD, moving-average alignment, volume profile, support/resistance, and Bollinger Band geometry must agree. The minimum confluence threshold is three independent factors aligning before the pick advances.
Monte Carlo Simulation
One thousand simulated five-day price paths per candidate, drawn from 90-day historical volatility. Outputs win probability percentage, expected value, and full path distribution. Picks with poor probability geometry are filtered out regardless of confluence.
Earnings Proximity Filter
Any ticker within five trading days of an earnings announcement is auto-excluded. Earnings produce binary outcomes that the genome's price-action signals cannot predict; the only safe stance is to step aside.
Correlation Guard
Picks correlated above 0.7 against each other get flagged. The engine refuses to publish more than three highly-correlated names in the same window. This prevents a single sector move from wiping out the appearance of diversification.
Institutional Flow Detection
Unusual-volume signatures and accumulation patterns are weighted as confirming evidence. Retail-only price moves on thin tape carry less weight than moves with measurable institutional footprint.
Liquidity Score
Each ticker gets a 0–100 liquidity rating. Anything below 30 is filtered out automatically. This is the structural guard against being trapped in a position you cannot exit at a reasonable spread.
Setup Expiration
Every signal carries a three-to-five day validity window. Setups that go stale auto-expire from the watchlist rather than sitting around inviting late, low-quality entries after the move has already played out.
Sentiment Pulse
FinBERT NLP processes ticker-relevant news and social keyword streams. Sentiment is a tie-breaker, not a primary driver — it adjusts confidence on otherwise-qualified setups, not the qualification itself.
Picks that survive all eight factors get published with an A+ through F confidence grade, entry price, stop loss, target price, Monte Carlo win probability, strategy label, and the list of contributing signals. Subscribers see exactly which factors triggered. No black box, no “trust me.”
Per-Asset-Class Risk Envelopes
A penny stock does not behave like a mega-cap. An ETF does not behave like a low-float runner. Most trading platforms ignore this and apply one stop-loss percentage to everything — which means the same setting that is reasonable for SPY is structurally suicidal for a four-dollar biotech, and the setting that fits the biotech is so wide it never triggers on the ETF. The SST engine tunes risk to the liquidity tier, not to a default.
| Profile | Stop Loss | Take Profit | Trailing Stop | Reward/Risk | Score Threshold |
|---|---|---|---|---|---|
| Penny & Small Cap | 3.0% | 9.0% | 3.5% | 3.0 : 1 | 80 |
| Mid Cap | 2.0% | 4.0% | 2.5% | 2.0 : 1 | 75 |
| Large & Mega Cap | 0.8% | 1.5% | 1.0% | 1.88 : 1 | 70 |
| ETF | 0.8% | 1.5% | 1.0% | 1.88 : 1 | 70 |
Penny and small-cap setups get wider envelopes because the underlying volatility demands it and the typical run-up is large enough to justify a 3:1 reward-to-risk geometry. Mid caps get tighter parameters with a 2:1 target. Large caps and ETFs trade on much smaller percentage moves, so the engine compresses both the stop and the target accordingly while accepting a narrower reward-to-risk ratio. Each profile carries its own minimum score threshold, because the bar for risking capital on a thin-float penny is structurally different from the bar for risking capital on a liquid mega-cap.
All orders are LIMIT orders, offset roughly one percent from the current price. Market orders are prohibited platform-wide. The live alpha-testing account is a Roth IRA cash account, which means no shorting, no margin, T+1 settlement, and tax-advantaged compounding when the system is profitable. These constraints are documented because they affect what the engine can and cannot do — subscribers should understand them before evaluating the methodology.
The Two-Engine Architecture
The SST product line is intentionally split into two engines that share the same genome architecture but operate under different scope and different proof gates. This is not a marketing distinction. It is a deliberate sequencing strategy designed to prevent the multi-asset SaaS variant from going live before the equities engine has earned the right to be replicated.
SST Alpha Engine
- Stocks and options on US equities — the engine described on this page
- Running today against a live Roth IRA cash account at IBKR
- Six-agent pipeline + 96-signal genome + 8-factor prediction scoring
- Per-profile risk envelopes (penny / mid / large / ETF)
- Members site publishes signals + scorecard + heatmap from this engine
- Purpose: prove the architecture works on real money before scaling it
SST Omega Engine
- Multi-asset coverage: options, futures, forex, ETFs, mutual funds, crypto
- Multi-user SaaS variant built on the same genome architecture
- Nine specialized agents extending the Alpha pipeline for cross-asset analysis
- Not unlocked for paid customers until Alpha posts verified track record
- Status: parked behind the proof gate — in development, not in production
- Purpose: the second-stage commercial product once the first stage has earned trust
Why split it this way? Because the alternative — rushing a multi-asset SaaS to market before the underlying methodology has live receipts — is exactly the playbook every failed signal service has run. The proof gate is the structural commitment that Omega does not get monetized to subscribers until the Alpha numbers earn it. If you have ever paid for a product that promised cross-asset edge with no live equity-side track record to back it up, you already understand why this sequencing matters.
Transparency Receipts
Every claim made on this methodology page is backed by an auditable artifact. The public can see the proof gallery and the scorecard without paying anything. Members get the full signal detail and prediction history. The architectural commitment is simple: if the engine cannot stand up to public inspection of its own results, it does not deserve a subscription.
AI vs Guru Proof
Timestamped entries on every published trade, head-to-head against well-known alert services. Public access, no paywall.
Members Scorecard
Every closed position, winners and losers, with no editing. Win rate, average gain, average loss, and monthly breakdown.
Live Performance Panel
Paper-trading and live alpha numbers side-by-side on the homepage. Live numbers clearly labeled as small-sample, in-progress alpha testing.
Wins get posted publicly after trades close. Losses get posted on the scorecard. Live picks stay behind the members paywall to prevent front-running — that protection benefits the subscriber, not the platform. None of this is unusual for a serious trading operation. It is unusual for a retail-facing alert service, which is a separate observation entirely.
Disclaimer & Compliance
Strong Stock Trading L.L.C. is an Arizona limited liability company. The SST Alpha Engine and the strongstocktrading.com members site provide EDUCATION and AI-powered market analysis tools. Nothing on this platform constitutes investment advice. Strong Stock Trading L.L.C. is not a registered investment advisor and does not hold itself out as one.
Numbers cited on this page (genome generations, Monte Carlo simulations, live trades, win rate, Sharpe estimate, codebase size, signal counts) are updated periodically. As of June 13, 2026: 340 live trades closed, 26.5% overall win rate, avg win +5.07%, avg loss -3.9%, avg return -1.52% (all trades, inclusive of losses), on a Roth IRA cash account in active alpha testing. Best single pick: GXAI +17.25% (Feb 03, 2026). Category win rates: Penny 36.2%, Small Cap 16.9%, Mid Cap 15.7%. These numbers should not be taken as indicative of future performance under different account sizes, different market regimes, or different brokers.
Now You Know How It Works
No black box. No hype. Decide for yourself whether the architecture deserves your subscription.