Atlas®

Quantitative Market Intelligence & Execution System

Investment Policy Statement (IPS)
Autonomous Intraday Momentum Breakout Strategy
Robinhood Agentic Trading Platform

Version: 1.0-final

Effective Date: July 6, 2026

Classification: Authoritative Runtime Specification

Not investment advice. This document formalizes deterministic machine-executable rules
for an autonomous trading agent. Edge and expectancy remain unvalidated.

1. Executive Strategy Summary

Strategy Profile

Strategy Name	Atlas Intraday Momentum Breakout v1
Asset Class	US Equities (Long-Only)
Universe	S&P 500 constituents (≥1M ADV, $10+ price)
Timeframe Stack	5-minute (execution) / 30-minute (context) / 1-day (bias)
Entry Window	09:35–11:00 ET (90-min morning window)
Holding Period	Intraday only; hard flatten 15:45–15:50 ET
Max Concurrent Positions	3 (max 2 per sector)
Account Type	Robinhood Cash Account ($500–$1,000); no PDT rule applies
Execution Mode	Fully Autonomous, Risk-Engine Gated

Edge Hypothesis

Stocks exhibiting relative volume ≥2.0x, price above a rising 5-minute EMA20/EMA50 alignment, and a clean breakout above a well-formed consolidation (≥6 bars, range ≤0.75×ATR) during the first 90 minutes of regular trading hours have positive expectancy when:

Higher-timeframe (30m + 1D) trend alignment confirmed
Broad market regime (SPY) is supportive (TREND_UP or RANGE)
Spread, liquidity, and event filters remove untradeable names

Edge is captured via fixed-R risk management: entry at trigger, stop below base, half position sold at +2R profit, remainder trailed under 5m EMA9 with 5¢ offset. All risk bounded per-trade (0.50% equity) and per-day (2.0% equity).

2. Core Mandate & Scope

Trading Constraints

🔒 Non-Negotiable Rules

Long-Only: Short logic entirely disabled. No inverse exposures.
Equities Only: No options, no futures, no crypto, no forex.
Intraday Only: Zero overnight holdings. Automatic flatten 15:45 ET, enforced backstop 15:50.
Regular Hours Only: Extended-hours trading forbidden.
Whole Shares Only: No fractional positions.
Cash Account: Settled-funds (T+1) rules govern. GFV (Good-Faith Violation) prevention via E-13 settled-funds guard.

Broker Compatibility

Strategy verified against Robinhood MCP schemas for place_equity_order, review_equity_order, cancel_equity_order, get_equity_historicals, get_equity_quotes, and related tools. Key compatibility points:

Supported: Market/limit orders, stop-market orders, position queries, historical bars, real-time quotes.
Not Natively Supported: OCO (One-Cancels-Other) brackets, trailing stops, order modify/replace. These are agent-simulated.
Critical Assumption (R-3): Cannot place a second sell order on shares already reserved by a resting stop. TP execution uses cancel-then-place sequence to work around this.

3. System Inputs & Data Dependencies

Broker Account Inputs (Refresh at Each Cycle)

Input	Source	Staleness Limit
Equity, Cash, Buying Power	`get_portfolio(account)`	60 seconds
Open Positions	`get_equity_positions`	60 seconds
Open Orders & Statuses	`get_equity_orders`	30 seconds during QUEUED state
Realized Daily/Weekly P&L	`get_realized_pnl`, `get_pnl_trade_history`	60 seconds / 5 minutes
Tradability & Halt Status	`review_equity_order` alerts	Per order submission

Market Data Inputs

OHLCV (5m, 30m, 1D): get_equity_historicals(interval) for watchlist symbols and SPY.
Real-Time Quotes: get_equity_quotes for last price, bid/ask spread.
Session Clock: risk_engine.py clock (ET, weekday, market-hours flags).
Earnings Calendar: get_earnings_calendar premarket check (once per day).
Event Lockouts: Manual config file config/event_lockouts.json (Fed/macro event calendar).

Derived Indicators

All computed from raw OHLCV data per fixed definitions:

EMA20_5m, EMA50_5m, EMA9_5m: Entry trend and trailing exit.
EMA20_30m, EMA200_1d: Higher-timeframe alignment.
ATR14_5m: Volatility; used for stop distance and filters.
RVOL (Relative Volume): Cumulative volume today / 20-day average; min 2.0×.
Consolidation Range: max(high, last 6 bars) − min(low, last 6 bars); must be ≤0.75×ATR.
SPY Trend, Market Breadth Proxy: Regime classification inputs.

4. Entry Logic & Triggers

Setup Validation (Evaluated on Completed 5m Bars)

A setup is valid only if all of the following hold:

All eligibility gates (E-1 through E-17) pass.
Regime is TREND_UP or RANGE.
5-minute price trend: close_5m > EMA20_5m > EMA50_5m.
30-minute alignment: close_30m > EMA20_30m.
Daily bias: close_1d_prev > EMA200_1d.
Relative volume: RVOL ≥ 2.0.
Volatility band: 0.40% ≤ ATR_pct ≤ 2.50%.
Clean consolidation: ConsolRange(6) ≤ 0.75 × ATR14_5m.
No chasing gaps: GapExt_pct ≤ 3.0%.
Near breakout level: close_5m ≥ 0.995 × BreakoutLevel.

Trigger & Execution

Trigger: last_price > BreakoutLevel (monitored in real-time between bar closes).

Execution upon trigger:

Entry limit: min(last_price, BreakoutLevel × 1.0010) (marketable limit, +0.10% cap).
Stop price: max(base_low − 0.05, entry_limit − 1.2 × ATR14_5m) (whichever is tighter).
Stop distance validation: (entry_limit − stop_price) / entry_limit ≤ 2.0%.
Position size per Section 5 below.
Risk-engine validation: risk_engine.py validate-order must return approved=true.
Pre-trade review: review_equity_order must show no blocking alerts.
Order placement: place_equity_order(type=limit, limit_price=entry_limit, gfd, regular_hours, ref_id=uuid).

Post-Fill (First Fill Event)

Immediately place stop_market(side=sell, stop_price, filled_qty, gfd).
Compute r_unit_usd = filled_qty × (avg_fill − stop_price).
Set TP target: tp_price = avg_fill + 2.0 × (avg_fill − stop_price).

5. Position Sizing Engine

Size computation from risk capital, expressed in whole shares:

risk_dollars    = equity × 0.0050                       # 0.50% per trade
stop_distance   = entry_limit − stop_price            # > 0, enforced by rules
raw_shares      = floor(risk_dollars / stop_distance)

cap_notional    = floor((equity × 0.25) / entry_limit)    # ≤25% per position
cap_bp          = floor((min(cash, buying_power) − 100) / entry_limit)
cap_liquidity   = floor(0.0005 × ADV20)               # ≤0.05% of ADV20
cap_openrisk    = floor(max(0, equity×0.015 − open_risk_usd) / stop_distance)

shares = min(raw_shares, cap_notional, cap_bp, cap_liquidity, cap_openrisk)
shares = shares × regime_size_mod                      # Apply regime multiplier
IF shares < 1: REJECT trade

Key Constraints

Risk per trade: 0.50% of account equity (R-1b: consider 1.0% for accounts < $5k).
Notional cap: 25% of account equity per position.
Buying power: Never exceed min(cash, buying_power) − $100.
Liquidity: Order ≤ 0.05% of 20-day average daily volume.
Portfolio open risk: Total across all positions ≤ 1.5% of equity.
Regime modifier: RANGE regime reduces sizes to 50% (TREND_UP = 100%).
Whole shares only: Fractional shares incompatible with limit and stop orders.

6. Exit Logic & Priority Framework

Exits evaluated at each cycle in strict priority order. A higher-priority exit preempts lower ones. All exits logged with full decision context.

P	Exit	Trigger	Mechanism
1	Take-Profit (Half)	`last_price ≥ tp_price = avg_fill + 2.0R`	Limit sell ceil(qty/2) at max(tp_price, bid)
2	Breakeven Move	Immediately after P1 fills	Cancel stop → new stop_market(avg_fill + $0.01)
3	Trailing Stop	Each evaluation cycle post-TP	Trail = max(prior_trail, min(EMA9_5m, bar_low) − 0.05); re-place if > current_stop + 0.05
4	Regime Exit	Regime becomes TREND_DOWN or HIGH_VOL	Tighten: stop_market(max(current_stop, last_price − 0.5×ATR))
5	Time Stop	Position age ≥ 24 bars (2h) AND unrealized < +1.0R	Full exit via J2 limit sequence
6	EOD Flatten	First cycle at/after 15:45 ET with any position	Cancel all orders → market sell all (backed up by risk loop at 15:50)
7	Risk Halt Flatten	Daily loss limit breached (M3.2)	Emergency flatten (market sell all)

Critical Notes

No pyramiding or averaging down. Every exit logs trigger ID, inputs, order params, fills, and realized P&L.
Scaling: Only one partial exit at +2R. Remainder is trailed.
Unprotected window (OCO workaround): Between stop cancel and TP placement, position is undefended. TP limit placed immediately at max(tp_price, bid) to minimize exposure.

7. Risk Management & Guardrails

Per-Trade Limits (M1)

Max risk: 0.50% of account equity.
Max notional: 25% of account equity per position.
Entry slippage cap: 0.10% (limit-enforced).
Spread cap: 0.10% (if measurable; else fallback to marketable-limit caps).
Liquidity: ADV20 ≥ 1,000,000 shares; order ≤ 0.05% ADV20.
Stop distance: ≤ 2.0% of entry price.

Portfolio Limits (M2)

Max concurrent positions: 3.
Max gross exposure: 75% of equity (3 × 25%).
Max open risk: 1.5% of equity across all positions.
Max sector positions: 2 per sector.
Max new entries: 6 per day, 2 per hour.

Loss Limits & Kill Switches (M3)

Limit	Threshold	Action
Soft Daily	Realized P&L ≤ −1.5%	No NEW entries; manage exits only
Hard Daily	Realized + Unrealized ≤ −2.0% OR equity ≤ 0.98×day_start	FLATTEN ALL, HALTED_RISK, auto-reset next session
Weekly	Realized P&L ≤ −5.0%	FLATTEN ALL, HALTED_RISK; manual reset only
Consecutive Losses	3 full stop-loss exits in a row	No new entries for rest of day
Minimum Equity	Equity < $250	No new entries; account too small to size safely

Execution Modes (M4)

M4.3: Fully Autonomous (ACTIVE)

Every entry must pass risk_engine.py validate-order (approved=true) AND review_equity_order (no blocking alerts). Once approved, orders execute immediately without human intervention. This is the governing mode for this system.

8. Exceptional Conditions & Recovery

The following conditions halt Atlas and require intervention or automated recovery:

Data Quality Issues (N-1)

Trigger: Bar age > 90 seconds past close, or quote fetch fails. Action: Skip evaluations; counter++. At 3 strikes, halt new entries. Auto-recover on fresh data.

Broker API Failure (N-2)

Trigger: Tool errors on account-state calls. Action: Retry 2× with backoff (5s, 15s). Halt at 3 strikes. Recover after 3 consecutive healthy checks ≥60s apart.

Order Ambiguity (N-13)

Trigger: Order status neither filled, cancelled, nor rejected after 2 cancel attempts. Action: HALTED_MANUAL_REVIEW. Operator must investigate and resume.

Duplicate Fills (N-6)

Trigger: Filled qty > ordered qty for an order_id. Action: HALTED_MANUAL_REVIEW; market-sell excess only.

Market Halt / Circuit Breaker (N-10)

Trigger: Tradability=halted or review_equity_order shows halt alerts. Action: Freeze that symbol's orders (or all if market-wide). Native stops remain live. Auto-recover on tradability restore.

Core Principle: Broker-side native stops are the final line of defense and are never cancelled during halts (except during controlled flatten sequences). During exceptional conditions, positions are protected by their resting stops; Atlas manages the agent-simulated layers (trails, entries, exits).

9. Runtime Operations & Scheduling

Task Scheduling (R-10 RESOLVED)

Task	Cadence	Hours (ET)	Purpose
Premarket (Loop A)	Once per day	Before 09:25	Build watchlist, classify regime, no-trade-day check
Entry (trading-bot-intraday)	Every 5 min	09:00–10:59	Dense entry window; evaluate and execute new setups
Manage (trading-bot-manage)	Every 10 min	11:00–15:59	Exit management, trails, EOD flatten (first run ≥15:45)
Risk (trading-bot-risk)	Every 2 min	09:00–15:59	Kill-switch checks, flatten backstop (15:50 trigger)

Key Timing Notes

Entry window (09:35–11:00): Runs of the entry loop before 09:35 skip entry evaluation (E-3 gate). Only management allowed.
EOD flatten: First management-loop cycle at/after 15:45 executes flatten. Risk loop backs up with absolute flatten at 15:50.
Scheduler jitter: Gate times evaluated against the actual clock at run start, not scheduled fire time. Up to ~10 minutes of jitter tolerated.
Signal evaluation: All indicators computed only on completed bars (no mid-bar recalculations).

State Persistence

After every mutation, write to state/*.json:

positions.json: equity, cash, buying_power, open positions (symbol, qty, avg_cost, entry_time).
kill_switch.json: halt status, daily/weekly P&L, trades_today, equity peak.
watchlist.json: today's ranked setups, regime classification, no_trade_day flag.
decision_log.jsonl: structured event logs per cycle (one record per line).

10. Parameter Reference Table

All parameters are fixed for v1; no online tuning allowed.

Parameter	Value	Used In
Entry window (ET)	09:35–11:00	E-3, K6
EMA20_5m / EMA50_5m / EMA9_5m	20 / 50 / 9 bars	Entry trend, trailing exit
EMA20_30m / EMA200_1d	20 / 200 bars	HTF alignment, daily bias
ATR lookback	14 bars (Wilder)	Stop sizing, volatility filter
ATR % filter	0.40% – 2.50%	Entry gate (G1)
RVOL minimum	2.0×	Entry gate (G1)
Consolidation bars / range mult	6 bars / 0.75×ATR	Base definition (G1)
Max gap extension	3.0%	Entry gate (G1)
Marketable-limit cap	+0.10%	Entry pricing (G1)
Stop ATR multiplier	1.2×	Stop distance (G1)
Risk per trade %	0.50%	Position sizing (I)
Max position % of equity	25%	Notional cap (I)
Max portfolio open risk %	1.5%	Aggregate risk (I)
ADV liquidity cap %	0.05%	Order size (I)
Max concurrent positions	3	E-7
Max per sector	2	E-9
Max daily entries / per hour	6 / 2	E-17
Take-profit R multiple	2.0R	Exit P1 (K1)
Daily loss hard stop	−2.0%	M3.2
Weekly loss hard stop	−5.0%	M3.3

11. Critical Assumptions & Resolutions

R-1: Account Type ✓ RESOLVED

Resolution (2026-07-06): Account ••••3261 verified as a CASH account via get_accounts. PDT rule does NOT apply. Governing constraints: settled-funds/GFV guard (E-13), price band ≤25% of equity (E-10), whole shares only.

R-1b: Small-Account Risk Tier (PENDING USER CONFIRMATION)

At $750 equity, 0.50% risk = $3.75/trade. Whole-share sizing will reject many valid setups on $50–$150 stocks. Consider 1.0% risk tier while equity < $5,000. Current: 0.50% remains active.

R-4: Bid/Ask Spread Data (CONDITIONAL)

Assumption: get_equity_quotes returns bid/ask fields. Fallback: If absent, disable spread filter (E-11); rely on marketable-limit price caps for protection.

R-5: Event Lockout Calendar ✓ ACTIVE

Manual config file config/event_lockouts.json defines Fed/macro event windows. Empty file = no lockouts (logged as WARN at startup). Current: Fed events ±30 minutes.

R-6: Execution Mode ✓ RESOLVED

Resolution (2026-07-06): M4.3 (fully autonomous, risk-engine gated) is active. Approval question moot.

R-10: Task Scheduling ✓ RESOLVED

Resolution (2026-07-06): Retimed per Section 9. Entry loop 09:00–10:59 (*/5); manage loop 11:00–15:59 (*/10); risk loop unchanged 09:00–15:59 (*/2).

12. Compliance & Audit Framework

Deterministic Decision Framework

This rulebook contains no discretionary language. Every decision path is either a numeric comparison or an enumerated state transition. All gates are explicit and testable.

Decision Priority Order

Capital Preservation > Determinism > Fail-Safe Operation > Broker Compatibility > Implementability > Backtestability > Execution Realism > Performance.

Structured Logging

Daily JSONL logs (logs/atlas_YYYYMMDD.jsonl) capture:

Timestamp, cycle ID, event type (cycle, setup_pass, order_submit, order_fill, halt, etc.).
Full indicator state (EMA, ATR, RVOL, breakout level, consolidation range).
Eligibility results (passed/failed checks).
Sizing decisions (equity, risk $, stop distance, shares).
Order details (ref_id, order_id, type, fills, partial status).
Exit triggers and realized P&L.
Broker errors and halt reasons.

Logs retain enough data to replay decisions deterministically for backtesting and audit.

Broker Capability Verification

All native broker features referenced were verified against live MCP tool schemas except:

R-2: GFD order longevity across full session.
R-3: RH behavior when a second sell order conflicts with a resting stop.
R-4: Bid/ask presence in quote responses.

⚠ Legal Disclaimer

This document formalizes the operator's deterministic trading rules for machine execution. It is NOT investment advice. Edge and expectancy are unvalidated; the strategy has not been backtested. This rulebook serves as specification only; use is at the operator's sole risk and responsibility. Comply with all applicable securities laws and regulations.