FTO-1.6 · New

FTO-1.6

Know your odds and your exit before you enter

Covered calls, cash secured puts, condors, spreads: FTO-1.6 reads your whole position as one risk object and tells you the probability it reaches your profit target before your stop, plus the exit plan that gives it the best expected value. Benchmarked on 74,160 graded historical outcomes, it ranks winners above losers 74% of the time and beat our trained machine-learning models without using any training data of its own.

Try DeriveAI free

What changes for you

5

strategies get odds for the first time

Covered calls, cash secured puts, put selling, iron condors and poor man's covered calls previously had no model behind them at all. Now every structure speaks the same 0 to 100% language as the rest of the product.

9

exit plans scored on every candidate

Nine take-profit and stop combinations are evaluated for expected value, and the best one is attached to the trade. You do not just get the odds; you get the plan that earns them.

0

rows of training data

The model is analytic mathematics over the position itself. No black box, no retraining drift, and it answers instantly without slowing a single job.

An illustrative trade

What you see before you commit

Say the Optimizer surfaces a credit spread. FTO-1.6 hands you two things about it: the probability and the plan. The numbers below are illustrative, not a live quote.

The odds

68%

The chance this position touches its take-profit before its stop, within its lifetime. Expiring without reaching either counts as a miss, exactly how outcomes are graded.

Vertical spread probabilities are calibrated on graded production outcomes. When the model says 70%, about 70% happens.

The plan

Nine exits scored, one recommended

Take profit +25% · stop -50%

Small wins, frequent stops

Take profit +50% · stop -100%Recommended

Recommended: best expected value

Take profit +75% · stop -150%

Bigger target, slower resolution

Expected value under a deliberately conservative rule. The recommendation is decision support, not an automatic close.

Evidence that the method works

Shown one winner and one loser, who picks the winner?

The cleanest way to read a ranking score. Across 74,160 graded vertical spread outcomes held out from calibration, each approach was shown pairs of one winning and one losing trade. These figures are vertical spreads, the family with production-graded outcomes; they are the proof of method.

FTO-1.6, closed form, no training data

74%

Our trained machine-learning spread models

58%

A coin flip

50%

The analytic model beat the trained models on their own held-out data.

Vertical spreads, pairwise ranking accuracy on held-out data. Full tables below.

When it says 70%, seventy happens

Stated probability against what actually happened, credit spreads on the held-out window. Honesty you can check bucket by bucket: the average gap between said and delivered is 2.4 percentage points.

Model said

What happened

Low

Mid

High

Mid-range predictions track outcomes within about one point; the model runs slightly optimistic at the top of the range and we say so. Stability held across more than a year of consecutive out-of-sample months and multiple volatility regimes.

The negative result worth publishing

We tested the obvious shortcut. It fails.

The tempting way to price a multi-leg trade is to score each leg on its own and combine the answers. We tested that properly, several ways, across all 74,160 graded outcomes.

On credit spreads it is worse than flipping a coin. A credit structure profits when its legs lose value, so a leg's own profit probability points the wrong direction, and no combination rule or recalibration could fix it.

Reading the position as one object, which is what FTO-1.6 does, beat every shortcut by more than ten ranking points on both families. We did the rigorous thing, published that the shortcut fails, and built the harder correct thing instead.

FTO-1.6 on credit spreads

70%

A coin flip

50%

Leg math on credit spreads

43%

Every structure, one language

What each strategy gets from FTO-1.6. For single options and verticals the learned models keep the ranking; FTO-1.6 presents their output in the same schema and adds the path analytics they never had.

Strategy	Status	What you get
Cash secured put	New coverage	Odds, calibrated level, exit plan
Put selling	New coverage	Odds, calibrated level, exit plan
Covered call	New coverage	Odds level and exit plan
Iron condor	New coverage	Odds level and exit plan, no ranking
Poor man's covered call	New coverage	Odds and exit plan, unmeasured so far
Debit spread	Learned head retained	FTO-1.5+ ranking plus path analytics
Credit spread	Learned head retained	FTO-1.5+ ranking plus path analytics
Single option	Learned head retained	FTO-1 ranking plus path analytics

The five new strategies, measured honestly

They had no graded history, so we built one: 12,267 real structures assembled from historical market data and graded against real price paths under the same rules production uses. Here is what that supports today, and what it does not yet.

What you get today

A calibrated probability level: after fitting, stated probabilities land within 2 to 4 points of delivered rates.
A recommended take-profit and stop on every candidate, chosen by expected value.
The same risk language as every other structure in the product.

What we do not claim yet

A winners-versus-losers ranking for covered calls and iron condors. Ranking power there is near coin flip, so the product does not reorder them and no per-trade probability is displayed.
Any number for poor man's covered calls. Twenty-one measurable structures is not evidence, so it stays unmeasured until it is.
Production validation. FTO-1.6 launches observing quietly alongside the live models; that record starts accruing now.

The numbers, in full

Everything above, with nothing rounded away. All figures out-of-sample on a held-out window the calibration never saw.

Head to head, same held-out rows

Approach	Debit AUC	Credit AUC
FTO-1.6, analytic Closed-form position mathematics, no training data.	0.743	0.697
Trained ML spread models Our production spread models.	0.582	0.628
Best leg-probability composition The shortcut from the negative result.	0.640	0.576
Random baseline Coin flip.	0.496	0.506

Probability accuracy

Measure	Debit spreads	Credit spreads
Graded outcomes in test	15,939	13,032
Brier skill over base rate How much sharper than always guessing the average.	+10.5%	+9.1%
Calibration error Average gap between stated and delivered.	0.081	0.024

Collateral structure backtest

Strategy	Structures	Calibration error after fit
Put selling and cash secured puts Real ranking signal.	6,054	0.024
Covered call Level only.	5,939	0.044
Iron condor Level only, never reordered.	253	0.015
Poor man's covered call Unmeasured.	21	n/a

Where you meet it

Choose a covered call, cash secured put, put selling, poor man's covered call or iron condor in the Optimizer and FTO-1.6 is the model doing the scoring, exit plan attached. On single options and verticals the learned models keep the ranking and FTO-1.6 adds its path analytics alongside.

How FTO-1.5+ chooses exits

Performance measured on 74,160 graded vertical spread outcomes spanning more than a year of entries across dozens of underlyings, evaluated out-of-sample on a held-out window the calibration never saw. Coverage for cash secured puts, put selling, covered calls, iron condors and PMCC is calibrated against a historical structure backtest and is provided as risk analytics, not as a prediction of individual trade outcomes. Past performance of any model does not guarantee future results. Nothing here is investment advice.

Try DeriveAI free