DeriveAI is a ROUTUR, Inc. company.
The models are built and evaluated by DeriveDX Quantitative Labs, ROUTUR's research arm.
The upgraded FTO-1 was validated walk-forward across a full year: 1,207,199 candidate contracts in 12 monthly folds with a 30-day embargo, so no fold saw its own month. Candidate ROC-AUC is 0.7366, up from 0.6880 for the previous deployed model, and log loss improved from 0.5104 to 0.4665. On displayed probabilities the average calibration error is 2.9% (4.2% on the unseen second half of the span) and the worst displayed band sits 7.3 points from reality.
The upgrade also changed what the model is willing to say. A probability is displayed only when that band's recent delivered accuracy supports it: 44.2% of contracts get an exact number and the rest get a probability range, including every day in the regime where no tested configuration produced trustworthy numbers. Ordering above a stated 35%, which previously ran backwards, is now monotone through the 90%+ range. Trade-level economics under the upgraded selection have not been re-measured; the unfiltered historical cohort was not profitable.
FTO-1.5+ remains shadow-only. Its upgraded exit model, validated on the same walk-forward frame, doubled the modeled average outcome per trade against the previous version (+0.0215R versus +0.0107R, with fixed exits at −0.0829R on the same trades) and was better in 8 of 12 months. That average is frictionless, regime-carried and not a profitability claim. The timing and daily-review results remain earlier research evidence measured under the previous selector and are re-measured on the new selection in forward confirmation.
| Term | Meaning in this report |
|---|---|
| Target hit | Take profit reached before stop loss or expiry, under the model’s stated policy. It is not an overall profitable-trade or portfolio-return measure. |
| R | Realized return normalized by the policy’s risk basis. It excludes commissions, slippage, position sizing, capital constraints, and portfolio overlap. |
| Temporal holdout | A date-separated test period excluded from model fitting and calibration. It is historical out-of-sample evidence, not live-trading performance. |
| Forward monitoring | Predictions captured from a frozen artifact for recommendations created after its freeze. A forward metric is publishable only when the predictions were captured before outcome resolution and the cohort has matured. |
FTO-1 estimates whether a single option will reach a +1.0R take profit before a −0.5R stop or expiry. The validation runs across 12 consecutive monthly folds with a 30-day embargo: every month is scored by a model that never saw it, every label begins on forward day 1, and nothing the model reads is dated later than the day before the decision.
| Held-out metric | Result |
|---|---|
| Test candidates / folds | 1,207,199 / 12 monthly |
| Candidate ROC-AUC | 0.7366 |
| Candidate ROC-AUC, prior deployed model | 0.6880 |
| Log loss (prior model 0.5104) | 0.4665 |
| Calibration error on displayed claims | 2.9% |
| Calibration error, unseen second half | 4.2% |
| Worst displayed band | 7.3 points |
| Numeric coverage (range shown otherwise) | 44.2% |
| Stated ~94% band: claims / observed | 12,348 / 88.6% |
| Stated ~25% band: claims / observed | 90,125 / 24.1% |
| High confidence, over two weeks to expiry | 98.5% observed |
| High confidence, three to seven days to expiry | 84% observed |
FTO-1.5+ uses the upgraded FTO-1 to choose the contract, applies the adaptive take-profit/stop-loss policy, and checks from day one onward whether an open trade deserves an early-close review. The exit-policy head is validated on the same 12-month walk-forward frame as FTO-1; timing and daily review keep their earlier research results.
| Single-option outcome measure | Result |
|---|---|
| Adaptive mean R, upgraded exit model | +0.0215R |
| Adaptive mean R, previous exit model | +0.0107R |
| Fixed policy on the same trades | −0.0829R |
| Months better, version to version | 8 of 12 |
| Exit-plan ordering accuracy (previous version 0.7610) | 0.7555 |
| Early-review coverage / average day (earlier research) | 3.90% / day 1.09 |
On 22,241 resolved trades from a one-month research window, measured under the previous contract selector, the timing model first asks whether the trade resolves on day one and estimates a later day only when it does not:
| Timing measure | Earlier estimate | FTO-1.5+ | Improvement |
|---|---|---|---|
| Average distance from resolution day | 0.4528 days | 0.3547 days | 21.7% lower |
| Median timing loss | 0.2264 | 0.1774 | 21.7% lower |
| Within one day | 83.65% | 85.56% | +1.91 points |
The production model now separates ordering from display: a selection score ranks every contract, and a daily-refreshed display layer decides whether an honest probability can be shown or a range should stand in. A fresh cohort must capture those predictions at recommendation time and mature before a forward result can be published.
| Forward requirement | Status |
|---|---|
| Current production artifact | Frozen and versioned |
| Prediction captured before outcome | Required |
| Mature resolved cohort | Pending |
| Fresh out-of-time window | Required |
| Publishable forward performance | Not yet available |
A historical reconstruction is not a substitute for predictions recorded before the market outcome. The next published forward figure will come from the current production artifact and a sufficiently mature cohort.
| Publication check | Current state |
|---|---|
| Historical validation result | Published above |
| Prospective prediction capture | In collection |
| Outcome maturity | Pending |
| Forward metric | Withheld until valid |
No FTO-1.5+ forward claim exists yet. The adaptive-exit result is walk-forward historical evidence, the timing and daily-review results are earlier research evidence measured under the previous selector, and the model remains in shadow evaluation.
A forward figure will appear here only after the current predictions and review suggestions are captured at recommendation time, graded against later outcomes, and evaluated over a fresh mature window.
