Exit Rules

Position Monitoring & Exit Rules

monitor_open_positions() is called every scan cycle (every 30 s). DC/DCS threshold ordering lives in execution.position_monitor; the engine still owns LTP fetching, order placement, and final close bookkeeping.

Position State Machine

State	When Set	Meaning
OPEN	At entry	Normal monitoring.
PROFIT_LOCK	Peak observed P&L reaches TRAILING_PROFIT_LOCK_PCT of expected profit	Trailing floor is active for DC-family option spreads.
EXPIRY_APPROACH	Near expiry DTE ≤ 2	Position is near expiry and should be watched under tighter operator scrutiny.
CLOSING	Exit condition has fired and close routing is about to run	Close is in progress or awaiting confirmation; repeated monitor actions are skipped.

Box Spread — 3 exit conditions (checked in priority order)

#	Condition	Action	Why
1	days_left < 0	Close immediately	Bot was down over expiry weekend
2	days_left ≤ 2	Close early	Avoid pin risk near expiry
3	net_profit_after_close ≥ 0 AND gross_profit ≥ 70% of max_gross	Close early	Capture profit early, free up margin — but only if closing costs don't wipe the gain

# Profit target calculation (live LTPs fetched each cycle)
current_box_value    = C1_ltp − C2_ltp + P2_ltp − P1_ltp
current_profit_gross = (current_box_value − entry_net_debit) × lot
max_profit_gross     = (K2 − K1 − entry_net_debit) × lot
est_close_cost       = trade["total_cost"]  # use opening cost as proxy for closing cost
net_profit_after_close = current_profit_gross − est_close_cost

if net_profit_after_close > 0 and current_profit_gross >= max_profit_gross * 0.70:
    close_position()                    # 70% gross captured AND net is positive → exit
elif net_profit_after_close <= 0 and current_profit_gross > 0:
    hold_to_expiry()                    # close cost would wipe the profit — hold

Calendar Spread — 1 exit condition

Condition	Action
|z_score| ≤ CAL_EXIT_SIGMA (0.5)	Close — spread reverted to mean

Double Calendar — 4 exit conditions (checked in priority order)

#	Condition	Action	Why
1	dte_near < 0, or dte_near = 0 at/after NEAR_EXPIRY_TIME_STOP_HOUR:NEAR_EXPIRY_TIME_STOP_MIN IST	Time stop — close all 4 legs	Holds through expiry day morning, then exits before the close window.
2	Net P&L ≤ −DC_STOP_PNL_PCT (20%) × |net_debit| (after entry + estimated close costs)	P&L stop — Trigger 1	Max loss guardrail regardless of IV or spot. Fires first — fastest to compute.
3	Entry IV − current IV ≥ DC_STOP_IV_DROP (10 pts) Current IV estimated from live straddle LTPs	IV crush stop — Trigger 2	DC is Vega-long — sustained IV drop kills far legs faster than near theta helps. 10pt drop is the configured regime-change stop.
4	|spot − atm_strike| > DC_STOP_TENT_WIDTH (1.30) × near_straddle_entry Only active when DTE ≥ DC_STOP_TENT_DTE (3)	Tent break — Trigger 3	Spot outside the profit tent means delta is no longer neutral. Tight tent at DTE ≥ 3 only — inside 3 DTE the near gamma is too high to trust the tent shape.
✓	Net P&L ≥ DC_PROFIT_TARGET_PCT (32%) × |net_debit| (net of entry cost + estimated close cost)	Profit target — close all 4 legs	32% of net debit captured — theta is front-loaded so the bulk arrives in the first 2-3 days.

# DC P&L calculation (live LTPs fetched each cycle)
near_pnl    = (near_ce_entry − nc_ltp) + (near_pe_entry − np_ltp)  # profit from decaying short
far_pnl     = (fc_ltp − far_ce_entry) + (fp_ltp − far_pe_entry)    # profit from long legs
gross_pnl   = (near_pnl + far_pnl) × lot_size

entry_cost  = trade["entry_cost"]                                    # stored at execution time
close_cost  = TransactionCosts.dc_close_cost(lot, nc, np, fc, fp)   # estimated live
net_pnl     = gross_pnl − entry_cost − close_cost

# Trigger 1: P&L stop
pnl_stop = −abs(trade["net_debit"]) × DC_STOP_PNL_PCT               # −20% of net debit
if net_pnl <= pnl_stop: close_all_4_legs()

# Trigger 2: IV drop (estimated from straddle LTPs, no extra API call)
cur_iv = (nc_ltp + np_ltp) / (0.8 × spot × sqrt(dte_near / 365)) × 100
if entry_iv − cur_iv >= DC_STOP_IV_DROP: close_all_4_legs()

# Trigger 3: tent break (only when DTE ≥ 3)
tent_half = trade["near_straddle_entry"] × DC_STOP_TENT_WIDTH
if dte_near >= DC_STOP_TENT_DTE and abs(spot − atm_strike) > tent_half: close_all_4_legs()

# Profit target: 32% of net debit, net of costs
target = abs(trade["net_debit"]) × DC_PROFIT_TARGET_PCT
if net_pnl >= target: close_all_4_legs()

Stretched DC (DCS) — same 4 triggers, higher profit target

DCS uses the same P&L, IV, and time-stop triggers as DC, but its tent break is wider: DCS_STOP_TENT_WIDTH = 1.40. Its profit target is 42% of net debit (vs 32%) because the OTM wings take 1-2 extra days to contribute meaningful theta differential, so the position is held slightly longer.

AI Market — 3 exit conditions

#	Condition	Action	Bookkeeping
1	Live option LTP reaches the parsed target price	Profit target — close dry-run leg	Remove from open_trades, close matching ai_market_trades row.
2	Live option LTP reaches the parsed stop price	Stop loss — close dry-run leg	Records realised simulated P&L against the analysis id.
3	End-of-day monitor window	Time close	Prevents the AI single-leg trade from carrying overnight.

AI_MARKET is phase-1 dry-run only. The close path fetches live LTP for valuation but never sends a broker close order.

Master DC (MDC_*) — same base exits as the routed strategy

MDC trades are persisted with an MDC_* strategy key, but exit routing strips the prefix and reuses the same base close logic as the selected strategy. For example: MDC_DC follows the DC exit rules, MDC_DCS and MDC_DCS_SKEW follow the DCS exit rules, MDC_DDC follows DDC rules, and MDC_IC follows IC rules.

Iron Condor (IC) — 3 exit conditions

#	Condition	Action	Calculation
1	Credit captured ≥ IC_PROFIT_TARGET_PCT (50%) × entry credit	Profit target — close all 4 legs	current_pnl = entry_credit - current_credit_value
2	Current loss ≥ IC_STOP_LOSS_CREDIT_X (1.5×) × entry credit	Stop loss — close all 4 legs	current_loss = current_credit_value - entry_credit
3	dte_near < 0, or dte_near = 0 at/after configured time-stop IST	Time stop — close all 4 legs	Uses live LTPs if available; falls back to entry credit on expiry-edge LTP failure.

Backtest reuse of live exit helpers

The backtesting engine never re-implements exit logic. arb_bot/backtest/simulation/exit_evaluator.py calls the exact same evaluate_dc_exit(), evaluate_dcs_exit(), evaluate_ddc_exit(), and evaluate_ic_exit() functions (with DCS_SKEW routed through evaluate_dcs_exit()), preceded by update_calendar_position_state() on every cycle — identical to live monitoring. The human-readable reason string each helper returns is normalized to a stable enum-like code for reporting: PROFIT_TARGET, PNL_STOP, TENT_BREAK, TIME_STOP, IV_STOP, SHORT_STRIKE_BREACH, TRAILING_STOP, plus the engine-only END_OF_RANGE. Because the helpers read Config directly and the engine temporarily patches those class attributes per cycle, parameter sweeps change real exit decisions without any code duplication.