risk-management

name	risk-management
description	Invalidation discipline, exit rules, and portfolio heat — how to manage risk on open paper positions.

Risk management

Position sizing prevents ruin. Risk management prevents death-by-a-thousand-cuts. Separate concerns.

Invalidation discipline

Every thesis has an invalidation field. It is a commitment, not a suggestion:

• If invalidation triggers during market hours: close the position or scale down. Don't rationalize.
• If invalidation triggers overnight (gap-through): close at open unless a new thesis can be filed.
• If you want to hold past invalidation, you must file a NEW thesis with updated invalidation. Don't silently move the goalpost.

Use close_thesis(thesis_id, "triggered", ...) to mark an invalidated thesis even if you haven't yet exited the trade — future /enter calls for the same ticker won't see it as a live thesis.

Exit taxonomy

Trades close for one of four reasons. Tag them correctly on close_trade:

• WIN: target hit or thesis fully played out, >25bps positive.
• LOSS: stop hit or invalidation triggered, net negative.
• SCRATCH: exit within ±25bps; usually thesis-change or time-stop.
• OPEN: still live.

Time-stop rule of thumb: if the thesis had a "2-4 weeks" timeframe and 4 weeks pass with no target hit and no invalidation, SCRATCH the position. Capital re-deployed > dead money.

Portfolio heat

Before opening a new position, mentally sum current heat:

heat = Σ  max(0, unrealized_loss_to_stop) over open positions
       (= per-position R1 contribution if stops were hit simultaneously)

The order guard caps R1 per trade at 2.5% equity and R2 caps concurrent opens at 6 (2026-05-22 threshold raises; src/trading_agent/sizing.py is canonical). Worst-case heat ≈ 15% equity if every stop hits at once — and short option legs now contribute assignment-stress heat too. Live with that number — if it feels too high, keep open count down or widen position stops.

The "don't touch it" list

Paper account or not, these reflexes kill real accounts:

1. Averaging down into a loss without a new thesis. Hook allows it if sizing passes, but it almost always compounds the mistake. File a new thesis first, explaining why the second entry is different from the first.
2. Moving stops further away. If the original stop is hit, that's the trade. Don't escape-hatch by widening.
3. Revenge trading. After a loss, the /research step (with search_past_trades) is mandatory, not optional.
4. Selling premium on the way down. ALL SELL-to-open option legs are currently hard-blocked inside the order tools (R_short_option_open_blocked) until multi-leg combos get atomic sizing; R5b/R5c define the collateral/stop rules that will govern short legs when the block lifts. Don't invent workarounds — every short leg charges assignment-stress heat.
5. Holding single-leg long options through earnings. IV crush is a known lesson (see earnings-play skill).

What the hook does NOT catch

The PreToolUse hook protects entry. Exits are up to you:

• No auto-stop-loss on the broker side (Moomoo paper API is "fire a limit, manage manually").
• No trailing stop. Use /eod-review to flag positions that have run past targets without a plan.
• No correlation check beyond GICS sector bucket.

When a position goes wrong

1. Is invalidation triggered? → close, close_thesis(..., "triggered"), then immediately append a post_mortem note with the cause.
2. Is invalidation not triggered but the position is underwater? → hold per plan. Check your journal — a "2-4 week" thesis that's 3 days old is not failing yet.
3. Have you re-read the original thesis today? If yes, and you still believe it, hold. If no, do it before acting.