Beta · DeriveAI Pro

FTO-1.5

Same trade. Better exits.

Two people can buy the exact same contract and one makes money while the other loses, purely because of when they sold. FTO-1.5 works out when to sell before you ever buy.

Try DeriveAI free

A real example

A SPY spread that could make $2.74 or lose $0.26. FTO-1.5 chose to hold out for 75% of the profit and bail after losing 25%. That is roughly a 31 to 1 payoff, so it only has to be right one time in twenty to come out ahead. It gave the trade a 20% chance, which sounds bad and is in fact good.

Most people would have taken profits early and set a wider stop. On that trade it would have been the worse call.

How much better

Measured against FTO-1 on the shared fixed +1.0R / −0.5R policy over the July 2026 test block — the same trades, held back from training so neither model had seen them. FTO-1.5 is in shadow evaluation: these are research figures, and no forward or live result exists for it yet.

2.5x

closer to the truth

When it says 40%, it is about two and a half times closer to really meaning 40% than FTO-1 was.

60%

less lost on the average trade

Most option trades lose money. Letting the model choose the exits cut the average loss by roughly 60%.

16%

more on its best picks

On the trades it was most confident about, it returned about 16% more for every dollar put at risk.

Read the full evidence report

It understands spreads.
Debit and credit spreads get their own dedicated models rather than being squeezed through one built for single contracts. It reads the whole position and both legs.
It cannot fool itself.
The model is built so a harder target can never be given better odds than an easier one, and a wider safety net can never be given worse odds than a tighter one. Those would be nonsense, so it is not allowed to say them.
It tells you the bar, not just the number.
Every probability now shows the win rate that trade needs just to break even. A 55% chance sounds strong until you see it needs 60%.
It says when nothing is worth trading.
If none of the nine plans makes money on any candidate, the screen says so rather than serving you the least bad one dressed up as a pick.

FTO-1 is still the better sorter.
Handed a whole chain and asked to rank it best to worst, FTO-1 wins. That is why it is still the model behind the ranking you see by default, and why we did not replace it.
It is for picking, not for trading everything.
Its advantage is concentrated in the trades it likes most. By the halfway point of its own ranking it is about break even. Trade everything it scores and you give the edge back.
The spread results are early.
They come from only 38 separate days of data, which is thin. Debit spreads move from slightly losing to slightly profitable on average. Credit spreads improve but are still slightly negative. Treat both as promising rather than proven.
It is in beta.
It is good enough to use and not finished. We would rather say that than pretend otherwise.

What each slice of its ranking actually returned

Average return as a share of what you put at risk. The edge lives at the top and is gone by the halfway mark.

Top 5%

+0.541R

Top 50%

About break even

Spreads, before and after it chooses the exits

Average return as a share of what you put at risk. Above the line is a profit.

Fixed

Chosen

Debit spreads

Credit spreads

The numbers, in full

For anyone who wants to check the claims above. R means a multiple of what you put at risk, so 0.5R on a $200 position is $100. These are the July fixed-policy comparison figures. They are not comparable with FTO-1's own holdout numbers, which cover a different window and policy surface — the two must not be pooled.

Measure	FTO-1	FTO-1.5
Calibration error How far its stated odds sit from what really happens.	0.0718	0.0284
Log loss Penalty for being confident and wrong.	0.2886	0.2688
Brier score Average error of the probabilities.	0.0843	0.0799
Top 5% return What its most confident picks actually returned.	0.468R	0.541R
Average trade Across 100,908 test candidates, fixed exits against chosen exits.	−0.320R	−0.127R
Ranking power (ROC-AUC) Sorting a whole chain best to worst. FTO-1 still wins this one.	0.8735	0.8553

Spreads, average outcome per trade

Strategy	Fixed exits	Chosen exits
Debit spreads	−0.0128R	+0.0378R
Credit spreads	−0.0287R	−0.0067R

Getting it

FTO-1.5 comes with DeriveAI Pro. Pick it from the model menu in the Optimizer, or just use search, where Pro accounts get it automatically. Spread optimizations use it for everyone, free accounts included.

See FTO-1

Try DeriveAI free