Zeus AI Engine
Institutional Liquidity & Market Structure Intelligence
Official User Manual | Version 1.5.1
Developed by Plazo Sullivan Roche Capital (PSRC)
Platform: TradingView | Pine Script v6Category: Institutional-style Market Structure & Liquidity Signal IndicatorPrimary timeframes: 5-minute and 15-minuteStatus: Research and Human Forward Validation
1. Introduction
Zeus Engine is an institutional-style trading intelligence indicator designed to identify potential buying and selling opportunities at significant market reference levels.
Rather than relying on a conventional moving-average crossover or isolated oscillator signal, Zeus evaluates how price behaves around concentrations of historically significant liquidity, session boundaries, and volume-weighted valuation levels.
The system seeks to answer five questions:
Where is price interacting with meaningful market reference levels?
Are multiple independent reference levels concentrated in the same price area?
Has price rejected that area and subsequently confirmed a structural break?
What do higher-timeframe direction, market regime, volume pressure, and momentum suggest?
Where are the suggested entry, structural invalidation level, and potential profit objectives?
When Zeus identifies a qualifying setup, it displays a graded BUY or SELL signal with an entry reference, suggested stop-loss, and three take-profit objectives.
An institutional context dashboard provides a separate assessment of prevailing market conditions.
Zeus is a machine learning framework, not an automated strategy. It does not place trades, manage positions, access brokerage accounts, or guarantee trade outcomes.
DOWNLOAD THE ZEUS AI ENGINE
Download the White Paper
Also vailable on:
Ninjatrader, cAlgo cBot and MT5
QuantConnect (Lean Engine)
Python Stack (andas, scikit-learn, XGBoost/LightGBM, and PyTorch)
ZEUS AI Confirmation Gateway v1.0 - Python
Download the Plazo Sullivan Roche Capital | Python Integration for Zeus Engine v1.5.1
For superior entries, you can deploy our Python application that connects Zeus Engine's TradingView signals to ChatGPT, Claude, or Grok for an AI-assisted second opinion before your human team considers acting on them.
How it Works:
Capabilities
Module | Included |
ChatGPT / OpenAI API | |
Claude / Anthropic API | |
Grok / xAI API | |
Multiple AI reviewers | |
Zeus v1.5.1 alert parsing | |
Configurable minimum signal grade | |
Signal freshness validation | |
Price and structural target checks | |
SQLite review history | |
Duplicate-signal protection | |
Automatic trade execution | Intentionally excluded |
Independent live market-data verification | Not yet included |
Important: The AI models review the Zeus alert snapshot. They do not independently observe the current chart, live liquidity, or broker order book. Therefore, their verdict is an advisory assessment of the supplied evidence, not independent market confirmation.
Quick installation
Extract the ZIP package and open a terminal in its folder.
1. Configure your preferred AI provider.
For ChatGPT/OpenAI:
export OPENAI_API_KEY="YOUR_API_KEY"
export ZEUS_OPENAI_MODEL="YOUR_AVAILABLE_MODEL_ID"
export ZEUS_PROVIDERS="openai"
For Claude, use ANTHROPIC_API_KEY, ZEUS_CLAUDE_MODEL, and ZEUS_PROVIDERS="claude".
For Grok, use XAI_API_KEY, ZEUS_GROK_MODEL, and ZEUS_PROVIDERS="grok".
To request all three reviewers:
export ZEUS_PROVIDERS="openai,claude,grok"
2. Set your private webhook token.
export ZEUS_WEBHOOK_TOKEN="$(python3 -c 'import secrets;print(secrets.token_urlsafe(32))')"
3. Launch the gateway.
python3 zeus_ai_gateway.py serve
4. Connect TradingView.
Create one Zeus alert using Any alert() function call, and configure its webhook destination to:
You'll need an HTTPS reverse proxy or secure tunnel forwarding requests to the local Python server. TradingView webhooks have a three-second response limit, so Zeus queues reviews and processes the model requests separately.
The package README includes the full setup, local dry-run commands, and security considerations.
Verification performed
All 12 automated local tests passed, including BUY/SELL parsing, directional SL/TP validation, JSON wrappers, stale-signal handling, minimum-grade filtering, database deduplication, and dry-run review behavior.
The implementation uses the documented OpenAI Responses API, Anthropic Messages API, and xAI Chat Completions endpoint.
2. The Institutional Philosophy Behind Zeus
Professional trading research often focuses on market liquidity, price acceptance, structural transitions, and the relationship between price and volume-weighted value.
Zeus uses observable price and volume information to model several of these concepts.
2.1 Liquidity reference levels
Previous daily, weekly, and monthly highs and lows are prominent market reference prices.
These areas can attract breakout activity, profit-taking, stop orders, hedging flows, and discretionary interest.
Zeus monitors:
Previous Daily High (PDH) and Previous Daily Low (PDL)
Previous Weekly High (PWH) and Previous Weekly Low (PWL)
Previous Monthly High (PMH) and Previous Monthly Low (PML)
Completed Asian session high and low
Completed London session high and low
Daily VWAP deviation bands at ±1 and ±2 standard deviations
Weekly VWAP deviation bands at ±1 and ±2 standard deviations
The prior daily, weekly, and monthly levels use completed periods rather than developing higher-timeframe highs and lows.
2.2 Liquidity clustering
A reference level becomes more interesting when several distinct categories of levels occur close together.
For example, a previous weekly low may sit near an Asian range low and the daily VWAP lower second-deviation band.
Zeus measures this proximity using an ATR-adjusted clustering tolerance.
The default tolerance is 0.15 × ATR(14).
The engine assigns greater weight to certain longer-horizon levels and second-deviation VWAP bands. Related levels from the same reference family are not counted repeatedly as independent families within a cluster.
The resulting confluence score contributes to the signal grade.
Important: A higher confluence score means stronger agreement among Zeus's predefined reference-level rules. It does not establish a statistically higher probability of success.
2.3 Price rejection
A level touch alone is insufficient to produce a confirmed Zeus signal.
The engine first seeks a qualifying price rejection.
For a potential BUY, price must interact with a reference level, close back above it, and finish sufficiently high within the candle's range.
For a potential SELL, the inverse applies.
The source currently requires the close to be within the upper 40% of the candle for bullish rejection, or the lower 40% for bearish rejection, subject to additional price-extension limits.
This captures both liquidity sweeps and certain ordinary level-rejection patterns. Consequently, not every Zeus signal should be interpreted as proof of a stop hunt or institutional liquidity sweep.
2.4 Market Structure Shift (MSS)
After detecting rejection, Zeus waits for directional confirmation.
For a BUY, price must close above the previously frozen structural high reference with a qualifying bullish candle body.
For a SELL, price must close below the frozen structural low reference with a qualifying bearish candle body.
By default:
Structural lookback: 5 bars
Minimum displacement body: 0.20 × ATR
Maximum confirmation window: 12 bars
Signal cooldown: 8 bars
The structural break reference is frozen when the rejection event is registered, helping prevent later changes to the confirmation threshold.
The current MSS implementation uses a historical lookback break. It does not perform a complete higher-high/higher-low swing-sequence analysis or independently verify institutional participation.
The signal is generated only after the confirmation candle closes.
3. Quick Start Guide
Step 1 — Install the indicator
Open TradingView, access Zeus Engine [Plazo Sullivan Roche Capital], and add it to a standard candlestick chart.
If using a supplied Pine Script source file, paste it into the Pine Editor, save, and select Add to chart.
Step 2 — Choose your market
Begin with a liquid, actively traded instrument.
Suggested research markets:
Gold: XAUUSD
US technology index: USTEC / NAS100 equivalents
US broad-market index: US500 / S&P 500 equivalents
Major FX: EURUSD, GBPUSD, USDJPY
Crypto: BTCUSD, BTCUSDT, ETHUSD
Energy: Brent or WTI crude oil
Ticker names and prices vary by TradingView data provider and broker.
Step 3 — Select a timeframe
Recommended starting point: M15 for primary signal assessment and M5 for more frequent intraday observations.
Both have demonstrated functional signal behavior during preliminary chart reviews, but neither has a verified profitable forward-test record.
Zeus supports standard intraday chart timeframes up to 60 minutes. H1 is supported but not yet established as a preferred operating timeframe.
Step 4 — Use the default settings
Start with the standard v1.5.1 settings. Avoid optimizing individual parameters before accumulating a meaningful prospective sample.
Step 5 — Identify a signal
A qualified signal displays:
BUY or SELL
Grade: A+, A, B, or C
Signal classification
Confluence-family count
Entry reference
Stop-loss
TP1, TP2, and TP3
Step 6 — Review the setup
Before considering a trade, check the proposed stop distance, nearby obstacles, spread, volatility, scheduled economic events, and the institutional dashboard.
Step 7 — Create an alert
In TradingView:
Select Create Alert.
Choose Zeus Engine as the indicator.
Select Any alert() function call.
Configure your preferred notification method.
Save the alert.
The dynamic unified alert includes direction, grade, Entry, SL, TP1–TP3, confluence, signal classification, target types, and institutional context.
Step 8 — Record the observation
During forward testing, save the alert timestamp, ticker, timeframe, grade, structural levels, subsequent price movement, and actual human execution decisions separately.
Zeus indicates a setup; the human trading team decides whether and how to act.
4. Signal Grades Explained
Zeus grades each confirmed signal using the number of distinct clustered reference families and their accumulated weights.
Grade | Current qualification | Interpretation |
A+ | At least 3 independent reference families and score ≥8 | Highest rule-based confluence |
A | At least 2 independent families, unless already A+ | Multiple-family confluence |
B | Lower family count with score ≥3 | Stronger single-family reference |
C | Remaining valid confirmed signals | Basic qualifying setup |
Every grade still requires the indicator's rejection and structural confirmation rules.
An A+ signal is not equivalent to a guaranteed or historically calibrated high-probability trade.
Initial historical research has not established a consistent monotonic relationship between grades and trading outcomes. All four grades should therefore be tracked during forward validation.
5. Understanding BUY and SELL Signals
BUY signal
A BUY appears after the engine identifies an eligible bullish level rejection and a confirmed bullish structural break.
The indicator plots:
ENTRY: Confirmation candle closing price.
SL: Based on the retained rejection extreme, the confirmation candle extreme, and an ATR buffer.
TP1–TP3: Structural reference objectives or explicitly marked projected reward/risk objectives.
SELL signal
A SELL follows the equivalent bearish sequence: rejection of an eligible level, confirmed bearish structural break, and valid directional stop distance.
The entry reference is the signal candle's closing price.
Actual execution may occur at a different price, particularly in fast markets. The displayed levels are a research trade plan, not guaranteed executable prices.
6. Structural Stops and Three Take-Profit Objectives
Suggested Stop Loss
Zeus uses an ATR-buffered structural invalidation model.
The default stop buffer is 0.12 × ATR.
A proposed signal must satisfy the minimum stop distance and cannot exceed the configured maximum of 4 × ATR.
The stop calculation is based on the event's retained rejection extreme and the confirmation candle's extreme. It does not necessarily capture every intervening candle extreme in a complex multi-bar rejection.
Traders should independently assess whether the suggested stop properly protects the full structure.
Take Profit 1, 2 and 3
Zeus has two target modes.
RR-aware — Default
The system searches for qualifying structural objectives beyond minimum reward/risk thresholds:
TP1: minimum 1R
TP2: minimum 2R
TP3: minimum 3R
If a suitable structural target is unavailable, Zeus substitutes a projected R-based objective.
Structure only
The system prioritizes the next eligible structural reference levels without requiring the 1R/2R/3R minimums. Where sufficient structural targets are unavailable, a projected fallback may still appear.
Reading target labels
S = Structural objective
P = Projected objective
R = Distance from entry expressed relative to the proposed entry-to-stop risk
A label such as TP1 1.25R S means that the displayed target is a known structural reference located approximately 1.25 times the proposed stop distance from entry.
Zeus also identifies a nearby known structural obstacle. A projected 2R or 3R target should not be mistaken for an unobstructed price objective.
Historical research across several instruments suggests that shorter fixed-R targets can improve some first-touch observations relative to structural TP1. However, consistent positive net expectancy has not yet been demonstrated.
7. Institutional Intelligence Dashboard
The dashboard can be displayed in any of the four chart corners.
Its purpose is to provide context independently from the BUY/SELL trigger.
Institutional Daily Bias
Zeus uses completed higher-timeframe information.
Two selectable frameworks are available:
H4 EMA200 — Default: Requires directional agreement between the completed H4 close relative to H4 EMA200 and the completed daily close relative to Daily EMA50.
H4 EMA20 + D EMA50: Uses H4 EMA20 and Daily EMA50 agreement.
Outputs: BULLISH, BEARISH, or MIXED.
Markov Market Regime
Zeus classifies observed market conditions into five states:
UPTREND
DOWNTREND
RANGE
HIGH VOLATILITY
TRANSITION
The state classifier uses directional movement, EMA relationships, ADX, and ATR-relative volatility.
A decayed first-order transition matrix tracks transitions between these states. The dashboard incorporates a minimum effective transition-weight requirement before providing a directional Markov vote.
Its persistence and entropy information describe the historical transition behavior of the state model. They are not predictions of trade profitability.
Volume Pressure
Zeus estimates directional volume pressure using candle closing location within the bar's range, scaled by the chart's reported volume.
The result is smoothed and compared against a volume baseline.
Outputs are bullish, bearish, or neutral.
This is a candle-volume proxy, not exchange-verified bid/ask delta, institutional order flow, or a direct measurement of institutional positions.
On CFD and FX charts, volume may represent tick activity rather than centralized exchange volume.
Momentum
Momentum combines:
RSI(14)
MACD(12, 26, 9) histogram
A directional vote requires agreement between the relevant RSI threshold and MACD histogram direction.
Final Institutional Appraisal
Zeus combines the four directional domains using these predefined weights:
Component | Weight |
Institutional Daily Bias | 30% |
Markov Regime | 25% |
Volume Pressure | 25% |
Momentum | 20% |
The signed votes produce a BULLISH, BEARISH, or NEUTRAL appraisal and an alignment score ranging from 0 to 100.
The alignment score is not a forecast probability. A reading of 100/100 means full weighted directional alignment among the voting domains, not a 100% chance of a successful trade.
The dashboard is informational and does not automatically veto a confirmed BUY or SELL signal.
8. Signal Classification
Each Zeus signal is also assigned one of three research classifications.
CONTINUATION: The signal direction agrees with the frozen pre-rejection local trend measured from EMA20/EMA50 relationships and slope.
REVERSAL CANDIDATE: The signal direction opposes that pre-rejection local trend.
UNCLASSIFIED: The preceding trend information is insufficient or inconclusive.
This classification is independent from the signal grade.
It is an EMA-based diagnostic model and does not yet constitute a complete structural reversal/continuation taxonomy.
9. Recommended Timeframes and Assets
These are forward-testing priorities, not proven performance rankings.
Asset class | Suggested starting timeframe | Research considerations |
US indices | M15, then M5 | Cash-market open, volatility, news |
Gold | M15 and M5 | London/New York liquidity and abrupt volatility |
Brent / WTI | M15 and M5 | Energy releases, contract/feed characteristics |
Major FX | M5 and M15 | London and New York overlap |
BTC / ETH | M15 and M5 | Weekend liquidity and 24/7 trading |
For initial human research, M15 offers fewer and potentially easier-to-review signals; M5 provides more granular observations.
Do not assume the same parameters have equal predictive value across markets.
Crypto's continuous trading hours, exchange-specific volume, and weekend liquidity deserve separate analysis.
Higher timeframes beyond H1 are outside the current signal engine's supported operating range.
10. Recommended Default Settings
Use these as the initial, frozen Zeus v1.5.1 baseline.
Setting | Default |
Previous daily / weekly / monthly H/L | Enabled |
Asian / London ranges | Enabled |
Daily / weekly VWAP deviation levels | Enabled |
Cluster proximity | 0.15 ATR |
ATR lookback | 14 |
Structural break lookback | 5 bars |
Bars permitted after rejection | 12 |
Minimum displacement body | 0.20 ATR |
Stop buffer | 0.12 ATR |
Maximum stop distance | 4 ATR |
Signal cooldown | 8 bars |
Target mode | RR-aware |
TP minimums | 1R / 2R / 3R |
Display last signals | 2 |
Projected level length | 30 bars |
Daily bias framework | H4 EMA200 |
Markov decay | 0.995 |
Minimum Markov transition weight | 20 |
Signal classification | Enabled |
Outcome observation horizon | 100 bars |
Episode gap | 20 bars |
Observation sample | All loaded history |
Validation population | Episode-first |
Round-trip cost | 0 ticks initially; configure before evaluating net performance |
Unified alerts | Enabled |
Minimum alerted grade | B |
Dashboard view | Executive |
Session settings
The source defaults to the America/New_York timezone with:
Asian range: 19:00–00:00 New York time
London range: 02:00–05:00 New York time
These are configurable research windows, not universal definitions of the complete geographical trading sessions.
TradingView's IANA timezone handling accommodates seasonal timezone changes, but the intended windows should be reviewed against each instrument's actual session schedule and data feed.
Signal display settings
Choose to display only the latest 1, 2, or 3 signals.
This controls historical trade-plan drawings and labels on the chart. It does not limit the historical signals included in Zeus's statistical calculations.
11. Research Dashboard and Forward-Testing Metrics
Zeus v1.5.1 includes an observational outcome engine.
It compares the existing structural TP1 against four fixed reward/risk objectives:
0.50R, 0.75R, 1.00R, and 1.50R.
All experiments preserve the same original entry reference and stop distance.
The indicator records:
TP-first and SL-first observations
Ambiguous events where TP and SL are both touched in one candle
Expired and unresolved observations
Mean R for decisive outcomes
Maximum Favorable Excursion (MFE)
Maximum Adverse Excursion (MAE)
Bars to resolution
Grade-based and classification-based observations
Episode-first and overlapping signal counts
Estimated round-trip trading costs expressed in ticks
Cost-adjusted observational results and scenario ranges
Independent episode analysis
The default episode gap is 20 bars.
The first signal following a sufficiently long signal-free interval is designated episode-first. Subsequent signals inside the specified gap are classified as overlapping.
This reduces duplicate-signal distortion, but it does not guarantee statistical independence.
Observation samples
All loaded history: Reconstructs historical results using the bars available on the chart.
From selected date: Reconstructs observations beginning at the chosen date.
Realtime only: Tracks signals observed during the active realtime execution of the script.
Realtime-only observations are not permanently retained when TradingView reloads or recalculates the indicator.
For a durable forward-test record, use external alert capture or a manually maintained log.
Cost assumptions
The default round-trip cost is zero ticks. This indicates that the displayed result is gross, not that execution is free.
For more realistic analysis, estimate the combined effect of spread, round-trip commissions converted into price ticks, and expected slippage.
Actual costs differ by broker, instrument, volatility, order type, and trade size.
Important measurement restrictions
Historical first-touch observations are not executed trades.
Zeus cannot determine which level was hit first when both SL and TP are touched inside a candle. Such observations are marked ambiguous, with separate scenario assumptions.
Resolved-only mean R excludes ambiguous and expired signals. The expanded dashboard provides additional diagnostics, but its standard-error estimates should not be treated as statistically robust confidence intervals.
12. Best Practices for Human Traders
Respect confirmation
Act only on a completed signal. A level interaction or dashboard directional vote is not itself a confirmed Zeus entry.
Examine the full structural context
Review the originating liquidity level, nearby obstacles, trend conditions, stop distance, and scheduled market catalysts.
Avoid treating every grade as a proven probability
A+ is a confluence designation, not a statistically verified win-rate category.
Compare the proposed entry with obtainable execution prices
The chart entry corresponds to the signal-close reference. A trader may receive a different fill after an alert.
Maintain independent risk limits
Zeus is not a portfolio risk engine. The user must manage position size, cumulative exposure, correlated positions, drawdown, and trading suspension rules independently.
For a controlled forward test, a fixed small notional risk budget or paper-trading approach makes observations more comparable across signals.
Watch economic releases
Central-bank announcements, employment data, inflation releases, oil inventories, and other relevant catalysts can abruptly change liquidity and volatility.
Zeus does not include an automatic live economic-calendar blackout.
Do not force dashboard agreement
A technically valid reversal candidate can appear while the daily bias or Markov regime remains opposed.
Record the disagreement as information. Whether opposition should become a formal signal filter must be established through testing.
Preserve settings during the research period
Do not repeatedly tune ATR multipliers, confirmation thresholds, session windows or target objectives to fit recently observed results.
Use stable configurations and a predefined evaluation window.
Use normal candlestick data
Standard time-based candlesticks are recommended. Synthetic chart styles such as Heikin Ashi and Renko can alter price-derived levels and invalidate assumptions about actual executable prices.
13. Forward-Testing Protocol
PSRC recommends a disciplined human validation process.
For each qualified event, record:
Instrument, market data source, timeframe, date and time.
Signal direction, grade, classification and confluence count.
Entry, SL, TP1–TP3 and target-source classifications.
Daily bias, Markov regime, volume and momentum at signal time.
Alert receipt time and actual executable market price.
Estimated and observed trading costs.
Whether SL or each experimental TP was touched first.
Maximum favorable/adverse movement in R.
Whether the event is episode-first or overlapping.
Any skipped entry and the reason for skipping it.
Evaluate each market and timeframe separately before combining results.
Pay particular attention to unresolved observations, session effects, and the difference between historical chart reconstruction and genuinely prospective records.
Do not declare a validated trading edge from attractive screenshots, isolated profitable signals, or historical gross mean-R figures alone.
14. Understanding the Limits of Institutional Analysis
Zeus is described as an institutional-style indicator because it incorporates market-reference, liquidity, VWAP, structural-confirmation, market-regime, and volume-pressure concepts frequently used in professional market analysis.
It does not claim to possess bank order books, broker-level client positioning, proprietary institutional transaction feeds, genuine exchange-wide order flow for decentralized markets, or verified knowledge of large market participants' intentions.
Its analytical components are transparent mathematical and price-based approximations.
The word institutional describes the design philosophy, not a claim of exclusive institutional information.
15. Risk Disclosure
Trading leveraged instruments, CFDs, cryptocurrencies, commodities, equities and foreign exchange involves substantial financial risk.
Zeus Engine provides analytical signals for research and educational decision support. Signals, stops, and targets are hypothetical planning references and are not personalized investment advice or guarantees of future performance.
Historical reconstruction, indicator first-touch experiments, and preliminary visual forward observations are not substitutes for independently verified live or paper-trading results.
Users remain responsible for all investment and trading decisions, risk management, execution, regulatory compliance, and resulting financial outcomes.
ZEUS ENGINE v1.5.1
Institutional Liquidity. Market Structure. Quantitative Intelligence.
Developed by Plazo Sullivan Roche Capital
Website: https://www.plazosullivan.com
Research release | Forward validation in progress | No guaranteed performance claims.
Comments