Top — Opportunities

Scored

—

Showing

—

Long

—

Short

—

Avg Score

—

Avg Confidence

—

Top Gainer

—

#	Symbol	Name	Price	Open	Change	Volume	RVOL	Trades	Dir	Score	Conf	Rec	Exp Move	Mkt Cap	$100 Target	Rationale
Loading opportunities...

Market MoversReal-time gainers, losers & most active from FMP

🟢 Top Gainers

🔴 Top Losers

🔥 Most Active

← Back to Dashboard

—

📈 Price History

📊 Price & Volume

🤖 AI Score & Signals

📋 Fundamentals

📐 Feature Readout

📐 CAN SLIM Analysis (N · L · M factors)

📊 Trade Activity Summary

✂️ Stock Split

👤 Insider Trades (90 days)

🏛️ Congress Trades (90 days)

🏦 Institutional Holdings (13F filings)

📰 Recent Documents & Catalysts

AlertsScore jumps, new catalysts, trading events

Unread only

Loading alerts...

System OverviewPipeline health & data stats

SettingsYour dashboard filters — find the best opportunities

🤖 Signal Filters

Direction

Min Score

Min Confidence %

Min Expected Move (basis points)

100bp = 1% move. Set to 50+ to see only meaningful opportunities.

Max Results

📋 Stock Filters

Price Range ($)

Market Cap Range

Exclude ETFs & Funds

Sectors (leave empty = all)

Ctrl+click to select multiple. Ctrl+click selected to deselect.

Exclude Tickers

Search and click to add. These tickers will be hidden from your dashboard.

📊 Volume & Momentum

Min Volume (absolute)

Min Volume vs Average (x ratio)

Unusual volume = institutional interest or catalyst confirmation.

Min Change %

e.g. 1 = only stocks up ≥1% · -5 = exclude stocks down >5% · 0 = all

Only show stocks already moving by this % or more.

📐 Technical Position

52-Week High Proximity

Stocks near highs often have momentum and breakout potential.

52-Week Low Proximity

Stocks near lows may be bounce candidates (contrarian).

Moving Average Position

MA position indicates trend. Above = bullish, below = bearish.

🛡️ Risk Gates

Min Dollar Volume (price × volume)

Ensures enough liquidity to enter/exit positions.

Max Spread %

Exclude penny stocks (under $1)

🔔 Notifications

Score Jump Threshold

New entrant alerts

Halt/SSR alerts

Filters apply to your dashboard view. Scoring engine runs globally.

My ProfileAccount details

👤 Personal Information

First Name

Last Name

Country

Receive communications

🔒 Change Password

Current Password

New Password (min 8 chars)

Confirm New Password

📋 Account Info

Manage UsersAdmin only

Create New User

Edit User —

First Name

Last Name

Country

Role

Status

Receive communications

New Password (leave blank to keep current)

Auto TradingPaper Trading Engine Performance

✅ Circuit breakers normal — trading active

📊 Open Positions

📋 Today's Trades

🚫 Today's Rejections

Candidates that the engine considered but DIDN'T buy. Each cycle examines high-scoring tickers and runs them through a gate funnel (score, confidence, volume, uptrend, etc.). When a candidate is rejected, the reason is recorded here. Use this to verify your gates are actually working — if you enabled the uptrend filter but see zero uptrend rejections, something's off.

📈 Performance History

How to read this: P&L is realized profit/loss on closed trades only — a +$4,514 day means you walked away with $4,514 more than you spent. Notional is total dollars committed to entries that day (not your loss exposure). Click any date row to see individual trades.

📝 Recent Alpaca Orders

All orders placed via Alpaca — buys, sells, stops, and cancellations. Use filters to narrow down.

Trading InfoHow the auto-trading engine works

🏗️ Architecture Overview

The auto-trading engine is a fully automated system that reads the AI scoring pipeline and executes paper trades via the Alpaca Trading API. It runs as a background worker during market hours (9:30 AM – 4:00 PM ET), checking for opportunities every 3 minutes. A pre-market scan at 9:00–9:29 AM builds a watchlist of high-momentum stocks before the bell. Every trade passes through multiple independent safety layers before execution.

Pre-Market (9:00–9:29 AM)
  Pre-Market Scan ── top movers by RVOL × momentum → watchlist (10 symbols)
       ↓
Score Engine (All Scored Tickers)
       ↓
  Entry Scan ─── watchlist-first, then by score ≥ 45, confidence ≥ 40%
  ├── Gate 1: No ETFs / passive funds
  ├── Gate 2: AI catalyst required (analyses_count ≥ 1)
  ├── Gate 3: Expected move ≥ 30bp
  └── Gate 4: Directional signal ≥ -0.05 (not bearish AI read)
       ↓
  Risk Manager ← validates every trade (ABSOLUTE VETO)
  ├── Exposure check (80% portfolio max)
  ├── Single position cap (15% max)
  ├── Sector concentration (30% max)
  ├── Risk-based sizing (1% portfolio risk per trade)
  └── Circuit breaker check
       ↓
  Alpaca Order ── market buy via paper trading API
       ↓
  Stop Loss ───── GTC stop placed immediately ($25 max loss)
       ↓
  Monitor Loop ── every 3 minutes during market hours
  ├── Circuit Breaker (equity vs day/week/peak)
  ├── Continuous Trailing Stop (GTC, ratchets upward)
  └── Position sync
       ↓
  EOD Close ───── 3:50 PM ET (no overnight risk)

📥 Entry Rules

The engine scans all scored opportunities each cycle and selects up to 3 candidates that meet these criteria:

✅ Score ≥ 45 (out of 100)

✅ AI Confidence ≥ 40%

✅ Direction = LONG only (no shorting)

✅ No halt_risk flag

✅ Not an ETF, fund, or index product

✅ AI catalyst required — at least 1 GPT analysis in last 3h

✅ Expected move ≥ 30bp

✅ Directional signal ≥ -0.05 (AI must not be clearly bearish)

✅ Not already held or traded today

✅ Daily budget remaining ≥ position size

Pre-market boost: Symbols on the 9:00 AM watchlist (RVOL ≥ 1.1, trade count ≥ 100, positive momentum) are sorted to the top of the candidate list at open.

No new entries in the last 30 minutes before market close.

💰 Budget Controls

💵 Max daily notional: —

📊 Position size: — per trade

🔢 Max positions: — concurrent

📈 Max daily trades: —

🕐 Trading window: —

💲 Commission per side: — (round trip = —)

Loading…

🛑 Stop Loss & Trailing Stops

Initial Stop Loss: Every position gets a GTC (Good Till Cancelled) stop order immediately after fill. Max loss per position: —. Stop persists across sessions until triggered or cancelled.

Trailing Stop (two-tier, time-aware):

🌅 Morning session (before — ET): stop sits — below the position's high-water mark

🌇 Afternoon session (from — ET): stop tightens to — below high-water mark

📈 Stop only ratchets up as new highs are made — never lowers

🔒 Trailing stop only activates once the position is profitable (no trailing on losing trades — those use the hard stop loss only)

🔄 Re-evaluated every cycle (~3 min) — stop orders are GTC

Loading…

Overnight recovery: At each day's first cycle, the engine checks all open DB trades. If a stop order expired (e.g., due to a previous bug), it automatically re-places a fresh GTC stop at the original stop price.

🚨 Circuit Breakers

Circuit breakers operate independently of the scoring engine. They monitor actual equity drawdowns and override everything:

⚠️ Daily DD ≥ 2%: Reduce all position sizes by 50%

🛑 Daily DD ≥ 3%: Close ALL positions, halt rest of day

⚠️ Weekly DD ≥ 5%: Reduce all position sizes by 50%

🛑 Weekly DD ≥ 7%: Close ALL positions, halt rest of week

🚨 Peak DD ≥ 10%: Close ALL, halt ALL trading until manual reset

The peak drawdown halt requires admin manual reset. All circuit breaker events write a Critical alert to the portal Alerts page.

🛡️ Risk Manager

Every trade proposal must pass the risk manager before execution. It has absolute veto power.

🏦 Max total exposure: 80% of portfolio (20% cash minimum)

📊 Max single position: 15% of portfolio

🏢 Max sector exposure: 30% in one sector

⚖️ Risk-based sizing: Max 1% of portfolio at risk per trade

💵 Buying power: Verified before every order

🚫 Minimum position: $100 (rejects tiny trades)

Rejected trades are logged with the reason. All trade entries, closures, and rejections also write to the portal Alerts page under Info/Warning severity.

📊 Data Pipeline

The scoring engine uses two paths to find opportunities:

Path 1 — News-driven: Alpaca + Tiingo news feeds → GPT-4.5 analysis → sentiment/catalyst extraction → feature engineering → score

Path 2 — Market-driven: Tiingo IEX live movers (top 200) → any stock moving ≥2% with volume ≥100K gets scored on market data alone (price velocity, volume spike, ATR expansion, RVOL, trade count)

The Market Anomaly Scanner also creates synthetic documents for big movers every 3 minutes, which then flow through GPT for deeper analysis.

⏰ Daily Schedule

Pre-market: News ingestion, anomaly scanning active

9:00–9:29 AM ET: Pre-market scan — builds watchlist of high-RVOL, high-trade-count movers

9:30 AM ET: Market opens — engine initializes circuit breakers, picks watchlist candidates first

9:30 AM – 2:30 PM ET: Active entry window — engine submits new BUY orders

2:30 PM ET: No more new entries — late-day setups have insufficient runway before EOD close

2:00 PM ET: Trailing stops tighten from morning (15%) to afternoon (8%) percent-from-peak

2:30 PM – 3:44 PM: Position monitoring only (stop loss, trailing stop, stall timeout)

3:45 PM ET: Auto-close all open positions (EOD) — flatten everything before close

After 4:02 PM: Engine idle, no trading until next morning

⚠️ Important Notes

• This is paper trading — no real money at risk. After 30 days of validated performance, the system can be switched to live trading.

• All configuration (budget, stop loss, thresholds) is in appsettings.json under AutoTrading, CircuitBreakers, and RiskLimits sections.

• The engine is always long, never short. The correct response to high volatility is reducing allocation, not reversing direction.

• Every trade is logged in the trades database table with full audit trail: entry/exit prices, P&L, score at time of entry, stop loss adjustments, and exit reason.

• All significant engine events (entries, closures, EOD, errors, circuit breakers) are written to the portal Alerts page for persistent visibility beyond the DigitalOcean log stream.

Stock ScreenerReal-time anti-manipulation filters

Min Volume Min Dollar Vol ($) Min RVOL Min Trade Count Max VWAP Dev % Exclude ETFs Limit

Set filters above and click Run Scan.

Trade Settings Auto-trading engine parameters — changes apply within 3 minutes

⛔ Emergency Stop New Buys

Halts new entries for the rest of today. Existing positions continue to be monitored normally (stop loss, trailing stop, EOD close). Auto-resets at start of next trading day. Use when you see something in the market going wrong.

💰 Budget & Sizing

Max Daily Notional ($)

Total dollars the engine can deploy per trading day

Min Position Size ($) — floor for dynamic sizing

Smallest single trade. Dynamic sizing will increase this up to daily budget.

Stop Loss Amount ($)

Max dollar loss before the stop order fires. GTC order placed immediately after fill.

Max Open Positions

Engine will not enter new positions if this many are already open.

🚦 Entry Gates

Minimum Score

Combined score (0–100) must exceed this threshold to qualify for entry.

Minimum Confidence %

AI + market confidence must exceed this. Higher = fewer but more reliable trades.

Profit Target Per Trade ($)

Dynamic sizing targets this dollar profit at the expected move. Position = target ÷ move%.

Dynamic Position Sizing

Auto-size positions to hit profit target. When off, always uses min position size.

📈 Trailing Stop

How it works: Once a position gains Trigger %, a trailing stop activates. The stop locks in Lock Fraction of current gains and ratchets up every Ratchet % of additional gain.

Trigger %

Position must gain this % before trailing stop activates.

Lock Fraction

Fraction of current gain to lock in (0.5 = 50% of gains protected).

Ratchet %

Re-raise stop every additional N% of gain above last raise.

Example:

📡 Path 2 — Market Scan Thresholds — pre-market watchlist + intraday momentum scan

Min Price ($)

Stocks below this are excluded from the scan.

Min Volume (shares)

Minimum shares traded. Filters illiquid movers.

Min Dollar Volume ($)

price × shares — ensures real capital is involved.

Min RVOL (× avg)

Relative volume vs 30-day avg. 1.1 = 10% above normal.

Intraday Min Change% (absolute)

Intraday scan requires |change%| ≥ this. Pre-market: positive only.

Min Trade Count

Entry gate: < this = likely block trades, not real participation.

Max Watchlist Size

Pre-market scan keeps top N by RVOL × change%.

Min Market Score (Path 2)

Gate 4e — market_score must clear this for entry. Lower = more candidates pass. 0–100, typical: 20–35. Default: 25.

Stall Timeout (minutes)

Close any position whose price hasn't moved up in the last N minutes (rolling window). If the highest price in the window never exceeds the price at the window's start, the stock is dead money and gets sold. 0 = disabled. Default: 60.

Commission Per Side ($)

Trade fee charged on each BUY and each SELL. Round-trip cost = this × 2. Set to 0 for Alpaca (paper or live, both commission-free). Set to actual broker fees if modeling live trading at a fee-charging broker. Affects: net P&L reporting, hard stop-loss trigger threshold (so total loss including fees stays at the configured stop amount).

Trailing Stop — Start Hour (ET)

The hour (Eastern Time) at which the engine switches from MORNING trailing-stop tolerance (looser) to AFTERNOON tolerance (tighter). Trailing is active in BOTH sessions, just with different room. Default: 14 (2:00 PM ET) — let trades breathe through morning wiggles, tighten into the EOD close to lock in gains.

Trailing Stop — Morning % From Peak

Used BEFORE the start hour above. Looser tolerance (larger %) lets early-day winners run through normal intraday wiggles. Default: 15% — tolerates the typical morning pullback (e.g. PMAX hit +13.6% peak, pulled back to +7.9%, then continued to +35% — 15% from peak survives that wiggle). Lower if you want more aggressive morning protection; raise if early winners keep getting stopped on noise.

Trailing Stop — Afternoon % From Peak

Used AFTER the start hour above. Tighter tolerance (smaller %) locks in gains as the EOD close approaches. Default: 8% — works well for liquid intraday movers in the protect-profits phase. For low-volatility names, consider 5%; for very volatile small-caps, 10-12%.

📈 Pre-Buy Uptrend Filter

Before placing each buy, looks at recent 1-minute bars and checks if the stock is actually trending up RIGHT NOW. Catches the "bought at the top of a 5% pop" pattern where a stock looked great on the scan but had already topped out by entry time. Three criteria must all be positive: overall window slope, last-third slope, and end-to-end price change over the recent window. If any are negative, the buy is skipped and the next candidate is considered. Skipped buys appear in the gate funnel as uptrend_check_failed:<reason>.

Enable uptrend filter — skip buys that aren't currently trending up

Window (minutes)

How far back to look when measuring the trend. Default 45 — long enough to capture meaningful direction, short enough to be recent. Below 15 min the slope is too noisy. Above 60 min you lose responsiveness to current direction.

Strict trend mode — adds two filters on top of the basic trend checks. (1) Current price must be above the mean of the last 10 bars — blocks buys "at the top" where price has stabilized after a peak. (2) Recent slope as % of price must exceed 0.05% per bar (~1% over 20 bars) — blocks "barely positive" noise. Default ON since 5/15 after 13 of 20 trades immediately went red despite passing basic trend checks. Require bar data — block the buy if recent bars are missing. As of 5/15, the engine now fetches bars ON DEMAND at check time: first tries Polygon, then falls back to Alpaca if Polygon is empty/fails. Only rejects if BOTH providers return insufficient data (meaning the symbol genuinely isn't trading right now — illiquid, halted, or pre-IPO). Recommended ON — protects against buying into a vacuum.

📊 VWAP Filter — day-level direction anchor

Volume-weighted average price from today's market open. Stocks trading above their day's VWAP are in buying territory; below is selling territory. Catches the "buying the dead-cat bounce" pattern where the 30-min trend filter passes during a brief upward bounce inside a larger day-level downtrend. Added 5/18 after 16 of 17 trades lost despite passing trend filters — diagnosis showed stocks were rejected by trend filter early, then briefly passed on transient bounces, then crashed. Rejections appear in the gate funnel as vwap_check_failed:below_vwap.

Enable VWAP filter — require current price ≥ today's VWAP at buy time Require VWAP data — if fewer than 3 bars exist since today's market open (very early morning before bars accumulate), block the buy instead of allowing it through. Recommended OFF for early-morning trading; ON if you only want to trade after VWAP is well-established.

📈 EMA + 5-Min Breakout Filter — Pass 2 entry quality

Identifies stocks currently trending up across multiple horizons, not just stocks that already moved. Requires three sub-checks to all pass: above 9 EMA (short-term trend), above 20 EMA (longer trend), and within X% of the last 5-min breakout high (mid-move, not late-move chase). Tighter proximity = fewer signals, higher quality. Added 5/19 after VWAP alone was insufficient — diagnosis showed engine was buying stocks already topping out. Rejections appear as ema_breakout_failed:*.

Enable EMA + breakout filter

Breakout proximity %

Require EMA data — need at least 20 bars for the 20-EMA calculation. ON blocks when bars are sparse (early morning); OFF allows through. Recommended ON.

🎯 Spread-Aware Entry Pricing — Pass 2 entry quality

Replaces the old "ask × 1.015" (1.5% above ask) entry buffer, which made the engine a liquidity taker at almost any price. New logic: limit = ask + min(buffer%, half-spread), with rejection of wide-spread quotes and quick cancellation if price runs while pending. A no-fill is not a loss; a forced fill near the top is.

Enable spread-aware entry pricing

Buffer above ask (%)

Max spread (% — reject if wider)

Cancel unfilled after (seconds)

Abort if price moves > (%) while pending

🛑 Daily Limits — circuit breakers

Hard caps applied per day. The auto-halt uses the same "Stop New Buys" toggle mechanism — when triggered, it persists for the rest of the day and resets overnight. Max daily trades caps the count of buy orders sent to Alpaca (not gate-funnel rejections).

Auto-halt at realized loss of ($, set 0 to disable)

Max daily trades

Max candidates per cycle (0 = unlimited, default)

📚 CAN SLIM Fundamental Filter — William O'Neil method, pre-qualification

Daily-refreshed (8 AM ET) cache of fundamental data per ticker. Scan loop reads from cache (no API calls during trading). Enabling shrinks the tradeable universe substantially — most small-cap momentum names (sub-$5 with no earnings) will fail. Each criterion independently toggleable. Rejections appear as canslim_failed:c_fail|a_fail|s_fail|i_fail in the Rejections panel (highlighted purple).

⚠ First-time setup required: The ticker_fundamentals table must exist before CAN SLIM can be enabled. If you've just deployed the CAN SLIM build, click the button below to create it. Safe to call repeatedly (no-op if already exists).

📊 Populate cache now: The auto-refresh runs once daily at 8 AM ET. To populate immediately (e.g., right after first deploy so you don't have to wait until tomorrow), click below. Takes 5-15 minutes for the full ticker universe. Safe to leave the browser tab open — the refresh runs server-side and continues even if you close it.

🔬 Diagnose CAN SLIM data quality: If a refresh completed but 0 rows have full data, FMP probably isn't returning what we expect. Use these to see exactly what went wrong.

Enable CAN SLIM filter (default OFF — expect ~80-90% universe shrinkage when enabled)

Required criteria (uncheck to make optional):

C — Current quarterly EPS growth ≥ minimum %, year-over-year (most recent Q vs same Q one year ago) A — Annual EPS growth ≥ minimum % average over last 3 years, AND positive each year S — Supply (float ≤ max). Smaller float = more upside potential on demand spikes I — Institutional sponsorship (# of 13F holders ≥ min). Real quality stocks have institutional buyers. ⚠ If 100% of rows show i_nodata in diagnostics, FMP's institutional endpoint isn't returning data for your plan — uncheck this until the endpoint path is corrected (use "Test FMP for one symbol" diagnostic to find working candidate). Require data — if a ticker has no cache row (new ticker, refresh worker hasn't reached it, FMP returned empty for it), BLOCK the buy instead of allowing through. Default OFF — permissive on missing data so new listings can still trade.

Thresholds (used by daily refresh worker):

C: Min quarterly EPS growth % (YoY)

A: Min annual EPS growth % avg (3y)

S: Max float (shares — 50,000,000 = 50M)

I: Min institutional holders (count)

Max tickers to refresh (0 = no cap = all active tickers, default; ordered by market cap DESC if capped)

Min market cap to refresh ($) (0 = no filter; set to 100,000,000 to skip OTC/micro-caps and drop 12k universe to ~2-3k real names)

Delay between tickers (ms) (FMP Premium = 750 calls/min; 600ms+ keeps us under cap)

Fetch institutional ownership data — disabled by default since FMP's endpoint returns 0% success on this plan (per dashboard). Saves ~25% API quota when off. Turn on once a working endpoint path is identified.

🗑️ Data Retention — auto-purge old rows

Daily purge at 3 AM ET. The scores table grows ~4 GB/month if left unchecked — every 3-min cycle writes ~500 rows with multiple JSONB columns. Defaults below are conservative for an active operator; tighten further if disk is constrained. Set any value to 0 to disable purge for that table. Tables NOT shown (trades, outcomes, tickers, fundamentals, users) are never purged — they're either small or contain irreplaceable P&L history.

Scores (historical rankings — main bloat source) days

Market bars (1-min OHLCV) days

AI analysis (GPT outputs) days

Documents (news articles) days

Trade activity (insider/congress) days

📧 Email Notifications

Sent to ALL active users with role='admin'. Requires SMTP credentials configured server-side via Email:* environment variables. All four toggles default off — opt in to whichever you want. Each event fires at most once per intent (e.g. one email per buy whether it fills or cancels).

On Buy — single email per buy, fired when the limit order fills OR is canceled On Sell — fired on every position close EXCEPT end-of-day (those go in the daily report instead) On Critical Alert — circuit breakers, hard stops, system failures (severity = Critical only) End-of-Day Report — daily P&L summary + per-trade table at ~3:50 PM ET

Rank formula:
RVOL × |change%| descending
Best conviction = highest relative volume AND biggest move.

Changes take effect within 3 minutes — no restart required

System AlertsAPI errors, service health & operational issues

🔴 Active Issues

🔧 API Diagnostics

📋 Recent System Events

🐛 DebugEngine state, scoring health & gate funnel — for diagnosing "why no trades"

⚙️ Engine Cycle Health

💰 Trading Status

🚧 Gate Funnel

🔬 Raw JSON

InformationUnderstanding AI signals & metrics

📊 Score (0–100)

The Score is a composite number from 0 to 100 that represents the overall strength of an AI-detected trading opportunity. Higher scores indicate stronger signals with more supporting evidence.

It combines three components:

Directional Signal (35%)AI sentiment analysis — how bullish or bearish the detected catalysts are. Derived from GPT's evaluation of news, earnings, and events.

Catalyst Signal (30%)Impact momentum — how significant and recent the catalysts are. High-impact events like earnings surprises or FDA decisions score higher than routine news.

Market Signal (25%)Technical confirmation from price velocity, volume spikes, and volatility expansion. Validates that the market is reacting to the catalyst.

Risk Penalty (-10%)Deductions for low liquidity, wide spreads, unreliable sources, halt risk, or contradictory signals.

60+ Strong 30 Moderate <30 Weak

🎯 Confidence (0–100%)

Confidence measures how reliable the score is — not how big the move will be, but how much you should trust the signal. A high score with low confidence means the signal is strong but uncertain.

Confidence is calculated from:

AI ConfidenceGPT's own assessment of how certain it is about the event interpretation. Ambiguous or speculative catalysts get lower AI confidence.

Liquidity ScoreHow easily the stock can be traded. Thin volume or wide spreads reduce confidence because execution is uncertain.

Source ReliabilityQuality of the data source. Major news wires and SEC filings are weighted higher than social media or unverified rumors.

50%+ High — act with conviction 25–49% Moderate — use caution <25% Low — monitor only

🧭 Direction

LONG indicates the AI expects the price to rise. SHORT indicates the AI expects the price to fall. Direction is determined by the sign of the raw score before taking its absolute value.

📏 Expected Move (basis points)

Estimated price movement in basis points (100bp = 1%). This is a heuristic based on catalyst impact, price velocity, and volume — not a prediction. Use it to gauge the magnitude of the opportunity, not as a price target.

💡 Recommendation

The Recommendation synthesizes Score, Confidence, Direction, and signal quality into a single actionable label. This is decision support — not investment advice.

Strong Buy ★★★

Score ≥ 50, Confidence ≥ 35%, AI catalyst confirmed, volume spike confirms. Highest conviction signal.

Strong Buy ★★

Score ≥ 50, Confidence ≥ 35%, no contradictory signals. Strong but without volume confirmation.

Buy

Score ≥ 30, Confidence ≥ 20%. Solid signal with reasonable conviction.

Buy — Mixed Signals

Meets Buy criteria but contradictory signals present. Proceed with caution.

Speculative Buy

Score ≥ 15 but low confidence, no AI catalyst, or thin data. Higher risk.

Watch

Score too low or conflicting signals too strong to act on. Monitor for developments.

Short-side equivalents (Strong Short, Short, Speculative Short) use the same thresholds for bearish direction.

📡 All Data Points Tracked

Complete inventory of every data source, feature, and signal that feeds into opportunity scoring. Data flows through the pipeline: Sources → Ingestion → AI Analysis → Feature Engineering → Scoring → Ranking.

📈 Market Data (Real-Time)

Source: Tiingo + FMP • Refresh: every 60 seconds during market hours

Last Price	Current traded price
Bid / Ask	Best bid and ask (when available)
OHLCV Bars	Open/High/Low/Close/Volume — 1m, 5m, 15m, 1h, daily
Volume	Current session volume
Avg Volume (30d)	30-day average daily volume — basis for RVOL calculation
RVOL	Relative Volume = current vol ÷ avg vol. ≥2x = high conviction (green), 1–2x = elevated (yellow)
Trade Count	Number of individual trades (from Alpaca bars `n` field). Distinguishes real broad participation from a few large block trades
VWAP	Volume-Weighted Average Price — from Alpaca bars `vw` field. Price deviation from VWAP flags pumps or dumps
Day High / Low	Intraday high and low prices
Previous Close	Prior session closing price
Change / Change %	Dollar and percentage change from previous close
Market Cap	Total market capitalization
P/E Ratio	Price-to-earnings ratio (trailing)
52-Week High / Low	Annual price range

🌐 Ticker Universe (Reference)

Source: FMP Screener • Refresh: daily via Universe Seeder

Symbol	Ticker symbol (e.g., AAPL)
Exchange	NYSE, NASDAQ, AMEX
Company Name	Full entity name
Sector / Industry	GICS sector and industry classification
Is Active	Whether ticker is currently tradeable

🤖 AI Analysis (GPT-4.5) (Per Document)

Source: GPT-4.5 structured extraction • Trigger: each new document ingested

Sentiment Score	-100 (extreme bearish) to +100 (extreme bullish)
Impact Level	0–100 significance rating for intraday effect
Bullish Probability	0.0–1.0 likelihood of upward movement
Bearish Probability	0.0–1.0 likelihood of downward movement
AI Confidence	0.0–1.0 model's certainty in its own assessment
Event Type	earnings, guidance, m&a, product, legal, macro, analyst, insider, layoffs, partnership, security_incident, regulatory, rumor, other
Time Sensitivity	immediate, intraday, or multi_day
Summary	≤600 char AI-written explanation of the event
Key Drivers	Up to 10 short driver strings explaining the signal
Risk Flags	halt_risk, low_liquidity, source_unreliable, contradictory, pump_risk, legal_uncertainty
Tickers Mentioned	All symbols referenced in the document
Numbers Mentioned	Key figures (EPS, revenue, guidance) with context
Model Version	GPT model used (e.g., gpt-4.5)
Prompt Version	Versioned prompt template for reproducibility

🔧 Computed Features (Per Scoring Cycle)

Computed every 1–5 minutes during market hours from raw data

vol_spike_ratio	Current 15m volume / avg 15m volume (30d) — detects unusual activity
rvol	Relative volume = current session vol ÷ 30-day avg vol. Live version of vol_spike_ratio
trade_count	Number of individual trades from Alpaca bars. <100 on high volume = suspicious block trades
dollar_volume	price × volume — real capital involved. Screener requires ≥$1M for high-conviction entries
vwap	Volume-weighted average price from stored bars. Baseline for detecting artificial price moves
vwap_dev_pct	% deviation of current price from VWAP. >±5% without volume = pump/dump signal
price_velocity	(last_price − price_15m_ago) / price_15m_ago — momentum
atr_expansion	ATR_15m_today / ATR_15m_30d — volatility regime change
gap_pct	(premarket_open − prev_close) / prev_close — gap impact
news_density_60m	Count of documents in last 60 minutes — catalyst frequency
sentiment_weighted	avg(sentiment × impact) — AI-weighted signal strength
bull_bear_skew	bullish_prob − bearish_prob — directional tilt
impact_momentum	sum(impact_level last 60m) − prior 60m — catalyst acceleration
source_reliability	Weighted score by source type (SEC filings > news > social)
liquidity_score	Function of spread, volume, and price — risk gating

📐 CAN SLIM Methodology (N · L · M factors implemented)

CAN SLIM is William O'Neil's growth stock methodology. It identifies stocks with strong fundamentals, institutional backing, and market leadership. This platform implements the three technical factors most relevant to intraday trading: N, L, and M.

Near New High

IBD Rule: Buy stocks breaking out near their 52-week high, not extended laggards. Leaders typically break out from bases within 15% of their 52-week high.

Score 100	Within 5% of 52W high
Score 85	Within 10% of 52W high
Score 70	Within 15% — IBD threshold
Score 40	Within 25% of 52W high
Score 0–20	>35% below high — laggard

Score impact: N ≥ 70 → up to +8 pts boost

Leader vs Laggard

IBD Rule: Buy the top 1–2 stocks in a strong industry group. An RS Rating of 80+ means the stock outperformed 80% of all stocks over the past 12 months.

80–100	Leader — top 20% of universe
60–79	Average — mid-range performer
0–59	Laggard — avoid or short

How computed: Each ticker's position in its 52-week range is percentile-ranked against the full scored universe each cycle. Stored as 0–100.

Market Direction

IBD Rule: Don't fight the market. 3 out of 4 stocks follow the general market trend. Only buy when the market is in a confirmed uptrend.

↑ Uptrend	SPY price ≥ 50-day MA
↓ Downtrend	SPY price < 50-day MA

Score impact: Downtrend → −8 pts penalty applied to every score. Checked every 5 min using SPY stored daily bars.

Full CAN SLIM also includes: C (Current earnings ≥25% growth), A (Annual EPS growth 3+ years), S (Supply/demand — float + institutional buying), I (Institutional sponsorship). These require quarterly earnings history ingestion and are planned for a future phase. The IBD verdict shown in each ticker's expand panel reads ✓ IBD-quality setup when N ≥ 70, L ≥ 80, and M = uptrend all align simultaneously.

📰 Documents & News (Textual Intelligence)

Source: Tiingo News API • Refresh: every 30–120 seconds

Title	Headline text
Source	Vendor/publisher identifier
Source Type	news, ir, earnings, filing, social, internal
Published At	Original publication timestamp
Body Text	Full article/release content (cleaned)
Content Hash	SHA-256 for deduplication
URL	Link to original source
Language	Document language (default: en)

👤 Insider Trades (SEC Form 4)

Source: FMP API • Refresh: every 15 minutes

Reporting Name	Name of insider (officer, director, 10% owner)
Title / Role	CEO, CFO, Director, 10% Owner, etc.
Action	Buy (Acquisition) or Sell (Disposition)
Shares Transacted	Number of shares bought/sold
Price	Transaction price per share
Total Value	shares × price — dollar value of trade
Securities Owned	Post-transaction holding count
Transaction Date	When the trade occurred
Filing Date	When Form 4 was filed with SEC
SEC Filing URL	Link to original Form 4 filing

Scoring Impact: Insider buying aligned with signal direction → +10 point boost; confidence +10%

🏛️ Congress Trades (STOCK Act)

Source: FMP API (Senate + House) • Refresh: every ~60 minutes

Member Name	Full name of congress member
Chamber	Senate or House of Representatives
Office / District	State and district of the member
Transaction Type	purchase, sale, sale_partial, sale_full, exchange
Amount Range	e.g., "$1,001 – $15,000" (STOCK Act range format)
Estimated Value	Midpoint of reported range
Asset Description	Description of the security traded
Date Received	When disclosure was received by ethics office
Capital Gains >$200	Whether trade had capital gains over $200
Filing URL	Link to STOCK Act disclosure

Scoring Impact: Congress buying aligned with signal → +12 point boost (highest of all trade activity); confidence +12%. Stronger than insider trades due to well-documented informational advantage.

🏦 Institutional Holdings (SEC EDGAR 13F)

Source: SEC EDGAR — free, no API key • Refresh: every ~60 minutes

Fund Name	Investment manager / fund name
CIK	SEC Central Index Key of the filer
Issuer Name	Name of held company (as filed)
CUSIP	9-character security identifier
Title of Class	e.g., COM, COM CL A, CALL, PUT
Shares Held	Number of shares in position
Market Value ($)	Dollar value of position (post-2023: exact dollars)
Investment Discretion	SOLE, SHARED, or DEFINED
Put / Call	Options exposure indicator (when applicable)
Filing Date	When 13F was filed with SEC
Period of Report	Quarter-end date the filing covers
Accession Number	SEC filing identifier for audit trail

Scoring Impact: Institutional accumulation aligned with signal → +5 point boost. CUSIP→ticker mappings are self-learned and improve over time.
Note: 13F filings are quarterly with ~45 day reporting lag.

✂️ Stock Splits (Detection & Penalty)

Source: FMP API • Refresh: every 2 hours • Window: ±90 days

Symbol	Affected ticker
Split Date	Effective date of split
Ratio	e.g., 4:1, 10:1, 1:10 (reverse)
Forward / Reverse	Whether numerator > denominator
Days Since/Until	Age or countdown to split date

Scoring Impact: Splits distort price/volume signals causing false positives. Penalty: −20 pts within 7 days, −10 pts within 14 days, −5 pts within 30 days. Upcoming splits: −10 pts. Confidence: −15% within 14 days.

🏆 Scoring Outputs (Per Ticker Per Cycle)

Generated every 1–5 minutes, stored in scores table

Score	0–100 composite opportunity strength
Confidence	0.0–1.0 signal reliability
Direction	long or short
Expected Move (bp)	Estimated price movement in basis points
Horizon (minutes)	30 (immediate), 120 (intraday), or 390 (full day)
Rationale	Human-readable explanation of the signal
Supporting Signals	List of evidence supporting the direction
Contradicting Signals	List of evidence opposing the direction
Recommendation	Strong Buy/Buy/Speculative/Watch/Short variants
Inputs Snapshot	All feature + document IDs for reproducibility

📊 Outcomes & Calibration (Backtesting)

Labeled overnight for weight tuning • Stored in outcomes table

Realized Move (bp)	Actual price movement over the predicted horizon
Max Adverse (bp)	Maximum adverse excursion (worst drawdown)
Max Favorable (bp)	Maximum favorable excursion (best peak)
Realized Horizon (min)	Actual time to peak move

🛡️ Risk Filters (Hard Gates)

Applied before ranking — tickers failing these are excluded from scoring

Min Dollar Volume	Minimum $ volume over last 20 minutes
Max Spread %	Maximum bid-ask spread percentage
Halt / SSR Detection	Trading halt or Short Sale Restriction active
Price Floor	Minimum price threshold (avoid penny stocks)
Blacklist	Manually excluded tickers

🧠 Self-Learning Maps (Auto-Growing)

Mapping tables that improve automatically as data flows through the system

CUSIP → Ticker	Maps 9-char CUSIPs from 13F filings to our symbols. Learned from each filing processed.
Issuer Name → Ticker	Maps 13F issuer names (e.g., "APPLE INC") to symbols. Uses 4-strategy fuzzy matching.
Company Aliases	Handles name variants by stripping suffixes (INC, CORP, LTD, etc.) and matching by prefix/contains.

📋 Data Source Summary

Source	Provider	Cost	Refresh	Data Type
Quotes & Bars	Tiingo + Alpaca	Included	60 sec	Real-time prices, OHLCV bars
News	Tiingo + Alpaca	Included	30–120 sec	Headlines + full text
Ticker Universe	Alpaca + FMP	Included	Daily	Active tradeable assets, seeding
Snapshots & Movers	Alpaca	Included	60 sec	Latest trade, quote, minute bar
Profiles & Fundamentals	FMP	Included	Daily	Sector, P/E, market cap
Insider Trades	FMP	Included	15 min	SEC Form 4
Congress Trades	FMP	Included	~60 min	STOCK Act (Senate + House)
Stock Splits	FMP	Included	2 hours	Calendar ±90 days
Institutional 13F	SEC EDGAR	Free	~60 min	13F-HR holdings (XML)
AI Analysis	OpenAI GPT-4.5	Per-token	Per document	Structured extraction