Research library

2026-08-17 to 2026-08-21 · Apex paper book

The guardian ratchet: stop giving winners back

A one-way trailing stop for same-day options. It sleeps until a trade is up 15%, then never lets the result go backwards.

VerdictADOPTED · RUNNING

In simple words

What happened

Across the 24 trades: held to expiry lost $696; the guardian on every trade made $8,054. With the exclusion applied — guardian on the two lanes it helped, hold-to-expiry kept on the lane it hurt — the same 24 trades come to +$8,271. 19 of 24 trades (79%) reached the arm threshold, so this is not a rare tail event — it governs most of the book.

Who would pay?

A same-day option that goes your way and then fades is worth zero at the bell — not less, zero. Holding to expiry is a bet that the peak IS the settle, and on a pin day it never is. On 2026-08-21 six positions were all profitable at some point and all six settled at nothing: −$4,571 held, +$778 under the guardian.

How it was tested

Every trade was replayed two ways on the same real prices: held to expiry, and managed by the guardian. The guardian has three states — asleep until the bid touches entry +15% (no stop at all, so early dips are free to work), then a floor at breakeven, then a stop that follows the peak up at a 35% distance and never moves down. In live operation the exit fills at the actual bid that crosses the stop, so slippage is real — one exit filled at $12.00 against a $12.09 stop.

What would prove it wrong?

The lane that already disproves it is kept out: across all eight CALL_UNWIND trades the guardian LOST $217, because that lane’s winners peak in the last five minutes and the trail fires on a mid-path wobble. So CALL_UNWIND rides to expiry, and every trade keeps recording both numbers forever — if the governed lanes start losing to hold-to-expiry out of sample, the policy comes back off.

Leader-collapsed evidence

What each setup actually did

24 positions over 5 sessions, one contract each, marked on the bid; guardian results reconstructed from recorded peaks

LanePositionsHeld to expiryGuardianDifferenceWhyDecision
F0FOLLOW_0DTE — same-day fast money13-$9,942-$1,875+$8,067Peaks early, dies at the bell — the guardian’s home turfADOPTED
PUPUT_UNWIND — the opposing play after a put unwind3-$1,775-$875+$900Same shape, smaller sampleADOPTED
CUCALL_UNWIND — the opposing play after a call unwind8+$11,021+$10,804-$217Winners peak into the settle; the trail clips themREFUSED

What we keep

Learnings

  1. 01

    No stop until +15% is deliberate, not reckless: same-day contracts routinely dip hard before they work (one trade fell −37% and finished +327%). A stop at entry harvests noise and kills the winners that pay for the book.

  2. 02

    Once armed, a breakeven floor appears — but it is a floor on the trigger, not the fill. Entries are booked at the mid of the two-sided quote and every exit fills at the crossing bid, so an armed trade can still book a loss of roughly the spread plus one mark interval (the book marks every two minutes). The asymmetry is real; it is not a guarantee.

  3. 03

    The exit fills at the real bid when it crosses the stop, not at the stop level. Model the slip; do not assume the trigger price.

  4. 04

    A policy that helps one lane can hurt another. The same trail that saved the fast lanes cost money on trend-into-settlement flow, so the exclusion is part of the policy.

  5. 05

    Record the road not taken on every trade, from day one. Counterfactuals cannot be back-filled, and an earlier study on the first 18 of these same trades concluded the OPPOSITE — one new day of data reversed it.

This policy is running. Every trade records both outcomes — managed by the ratchet, and held to expiry — so the decision keeps being re-tested on new data.
Back to all studies