Options paper trading on alphabench is more sophisticated than the equity paper trading path. The agent doesn't just check signals once a day. It watches tick-level micro-structure, generates a structured thesis, validates it against the live order book, and then manages the position with alert-triggered exits throughout the day.
This guide covers the full lifecycle: thesis generation, validation, deployment, monitoring, exit, and review.
1. How the agent generates an options thesis
During a wake, after identifying a directional opportunity in an underlying, the ambient
agent calls analyze_tick_opportunities to characterise the current micro-structure:
- Spread: bid-ask spread as a percentage of the option premium
- BSI (Buy-Sell Imbalance): order flow directional pressure. Positive BSI means net buying; negative means net selling.
- OI direction: is open interest accumulating or unwinding?
- Return windows: 1m, 5m, 15m return momentum
- Noise level: intraday price noise relative to the option premium
From this, the agent calls propose_paper_trade_thesis which generates a structured
thesis document:
Underlying: NIFTY
Direction: bearish
Instrument: NIFTY2660223300PE (weekly, 23300 put)
Strike selection: slightly OTM, 150 points below spot
Quantity: 50
Entry: buy at 38-42 (mid of current bid-ask)
Thesis: OI accumulation on PE side, BSI=-1.2 (net selling pressure),
momentum -0.8% over last 5m bar. Directional bias intact.
Stop: 28 (25% loss on premium, triggered by adverse move)
Target: 65 (70% gain on premium)
Score: 0.88 / 1.0
A score above 0.7 is typically the threshold for proceeding to validation.
2. Validation before deployment
Before deploying, the agent runs validate_tick_thesis, a market impact check that
prevents deploying into unfavourable execution conditions:
| Check | What it catches |
|---|---|
| Spread check | Bid-ask wider than 8% of premium: reject |
| Depth check | Order book too thin for the required quantity |
| Noise-to-stop ratio | Intraday noise > 60% of stop distance: stop fires on noise |
| BSI confirmation | Order flow directional pressure consistent with thesis |
If any check fails, the deployment is rejected with a specific reason and recorded in the missed-alpha ledger. The agent moves on without forcing a bad trade.
3. What options the agent can deploy
| Type | Status | Notes |
|---|---|---|
| Long call (buy_call) | Supported | Full paper execution |
| Long put (buy_put) | Supported | Full paper execution |
| Short straddle / condor | Analysis only | Research and sizing, no paper execution |
| Short naked options | Not supported | Risk profile unsuitable for autonomous execution |
The platform restricts autonomous paper trading to long options only (buy call or buy put). Short options strategies are available for backtesting and analysis through the chat interface but are not deployed autonomously.
The reason: long options have defined maximum loss (the premium). Short options have theoretically unbounded risk, which requires human oversight even in paper mode.
4. Deployment
A thesis that passes validation is deployed via deploy_paper_option_thesis:
- A paper trade record is created in the database with the entry price, quantity, strike, and expiry
- Two alerts are armed automatically:
- Target alert: fires when LTP reaches the target price
- Stop-loss alert: fires when LTP falls below the stop price
- The deployment appears in the Paper tab with live P&L updating on each tick
From the Paper tab you can see:
- Current LTP of the option vs entry price
- Unrealised P&L
- Time since entry
- Whether the stop/target alerts are still armed
5. Alert-triggered exit
When the target or stop alert fires, the agent wakes with:
Wake reason: alert:fired:NIFTY_PE_23300_TARGET
On-fire message: NIFTY 23300PE hit target 65. Entry was 39. +67% gain on premium.
Context: NIFTY2660223300PE | LTP: 65.5 | Time: 11:42 IST
The agent calls the exit tool, which records the exit trade at the current bid price (using the more conservative side of the spread for realistic P&L). The paper position is closed and P&L is finalised.
Three exit types:
| Exit type | What triggers it |
|---|---|
target_hit | LTP reaches or exceeds the target price |
stop_loss | LTP falls below the stop price |
session_end_close | Market close (15:20 IST) with position still open |
Session-end close prevents holding intraday option positions overnight. Overnight IV crush and gap risk are hard to manage autonomously.
6. PnL reconciliation
After any exit, the platform verifies:
- Entry record: the simulated fill at the entry price
- Exit record: the simulated fill at the exit price
- PnL calculation: (exit price - entry price) x quantity x lot size
- Reconciliation check: database total matches the computed P&L
You can see the full reconciliation in the Paper tab under each closed position.
7. Post-exit review
After each exit, the agent calls generate_post_exit_review and records:
- Thesis accuracy: did the price move in the predicted direction?
- Exit quality: was the exit at target / stop / session-end, and was that the right call?
- What to do differently: if it was a stop-loss exit, what signal missed the reversal?
- Updated beliefs: the agent updates its view of the instrument based on this trade
The review is visible in the wake log under the Ambient tab. After several sessions, you can read the review history to understand the agent's learning trajectory.
8. Monitoring without waking
Between alerts, the agent is not active. The paper strategies run on their own. Tick data flows, LTP updates, and the alert system monitors stop/target conditions without requiring any LLM wakes. The agent only wakes when a condition fires.
This makes paper options trading efficient: a position held for 2 hours may only generate 2 agent wakes (entry + exit), consuming minimal token budget while getting the full benefit of real-time tick monitoring.
Next steps
- Ambient Agents: research, deploy, and monitor: the broader autonomous loop that options deployment is part of.
- Paper Trading: equity paper trading for daily-bar strategies.
- Backtest Multi-Leg Options Spreads: validate options strategies in backtest before deploying.