Automated, defined-risk S&P 500 index-options trading software — offering brief
QuantGlide Apex is the most sophisticated program QuantGlide has ever released: a machine-learned S&P 500 index-options system that, across a five-year simulation, compounded $200,000 into $2.46 million net of modeled commissions — at a 2.39 Sharpe ratio, and without ever losing more than 12.7% from any peak.
It trades SPX/SPXW options through three autonomous intraday sleeves. Every position is a fully wing-protected, defined-risk structure: the maximum possible loss is fixed and known at the moment of entry, positions are cash-settled against the official index print, and nothing is ever managed, rolled, or stopped intraday — the structure itself is the risk control. There is no scenario, however extreme, in which a position loses more than the number printed on it at entry.
Each sleeve is gated by its own machine-learned model, retrained on a rolling walk-forward schedule with no look-ahead. The program is delivered as software: it trades the client's own brokerage account, under the client's control, sized by the client's own risk dial — and its models pass on roughly two of every three trading windows they evaluate. It does not need to be watched, fed, or second-guessed. It waits, and when its models find the edge, it takes it.
Simulated performance, 2021-10-14 → 2026-08-13, $200,000 initial, standard risk dial. See §IV and the disclosures in §VIII.
I created Apex to serve the needs of the most serious, well-capitalized investors. To that end, I will NOT be offering any of the cheap entry points to QuantGlide Apex that exist for the Flagship algorithm. No signals will be sold, no Discord alerts will be sent, and the features and exact mechanics remain strictly undisclosed and confidential.
QuantGlide Apex has a significantly higher MAR ratio than the Flagship, and a much more normal risk profile compared to the skewed risk profile of the Flagship. The Flagship earns its edge the way every premium seller does: its spreads are defined-risk, but each win is a small credit and the rare loss is several times a typical day's gain, because the loss only comes when the index runs through its short strikes. That is a skewed return profile — a quiet, steady climb, with drawdowns that arrive suddenly, sometimes in a single session. Apex is shaped the opposite way: each trade's defined maximum loss is on the order of what it stands to make — about 1.4× the credit on the late-day sleeve and about 1.8× on the midday sleeve, where a thin credit can push it to two or three times — and because most losses stop well short of that maximum, realized wins and losses run comparable in size. Drawdowns therefore build gradually out of losing sequences and recover the same way. Two edges, two opposite shapes — held together, each covers the other's weakness.
Apex is meant to be a complement to the Flagship algorithm, not a replacement for it. By adding another uncorrelated source of edge, investors can smooth out their portfolio with the power of diversification. That is why I designed Apex to be as different from the Flagship as possible: it trades the opposite part of the options chain — the strikes at and near the money, rather than the tails — and it leverages machine learning to extract its edge, rather than static rules built around tail-avoidance. It also trades both 0DTE and 1DTE options, unlike the Flagship which focuses only on trading 0DTE options.
While the track record shown here is simulated, I am highly confident in the process used to develop the algorithm — the same process that built the Flagship you know and love, including a comprehensive suite of parameter-robustness tests and the use of validation and out-of-sample data sets to keep overfitting to an absolute minimum.
Running Apex requires a minimum account size of $75K, and it may not be suitable for most investors. For those who are fortunate enough to be able to run it, this is a serious algorithm designed to achieve serious results.
The program expresses one thesis three ways: short-dated S&P 500 index options are systematically rich at specific points in the trading day, and that richness can be harvested selectively with a robust ML model that is retrained periodically and adaptive with market conditions. By creating 3 independent sleeves and fusing them into one strategy, we find optimal trading opportunities and use each sleeve's model to decide whether a given entry time is suitable for opening a position. Position size is stepped, not continuous: every entry is 1× or 2× the base unit, set by a predefined sizing-tier rule on each of two sleeves that reads that sleeve's own recent record — the model's score decides whether to trade, not how much. The rule steps size up after a losing stretch rather than down, which is exactly why the size of any position is hard-capped at 2 times the base unit: the cap is what keeps that rule from becoming runaway sizing or a martingale.
A same-day-expiry structure taken mid-morning when a trend-quality signal clears its adaptive threshold. Settles at that afternoon's official print — no overnight risk.
A second same-day sleeve at midday, gated by its own independently trained model over a wider feature set. Settles the same afternoon.
A four-leg, wing-protected structure opened near the close and cash-settled at the next session's official print. The single overnight exposure in the program; both tails are capped.
Defined risk, always. There are no naked options, no stop-loss dependencies, and no intraday management to fail. The worst case of every position is purchased up front. While many traders believe more management leads to better results, we have found that excessive management actually dilutes alpha by incurring large fixed costs (spreads, commissions) for no real expected gain in edge. For that reason, we stick to the classic hold until expiry approach that has already shown its merits in the Flagship QuantGlide algorithm.
Selectivity over frequency. Each sleeve trades only when its model clears its threshold. A trading window is one sleeve's chance to trade on one session, and across the simulation the program took roughly one window in three — 32% of the windows it scored, and between 32% and 39% in every full calendar year after 2022, when only the late-day sleeve was yet trading. It is in the market on about two sessions in three, usually with a single position: two sleeves stacked on about one session in five, and all three on about one in forty. This isn't a bug — since we are trading ATM spreads, the edge simply is not there most of the time. Most systems manufacture trades to justify their fee; this one earns its keep by refusing to trade without an edge.
Volatility-adaptive structure. The late-day sleeve sets its wing widths from the market's own expected move, so its defined risk per unit stays proportionate across calm and stressed tapes. The two same-day sleeves use fixed 50-point wings; there it is the credit collected that rises and falls with volatility, which means their maximum loss per unit is largest exactly when premium is thinnest. In every sleeve the risk dial sets how many units are traded from that day's defined maximum, so the dollars at risk per entry stay proportionate to the allocation whatever the structure's width.
Each sleeve is driven by a regularized regression model over engineered features of the current session and option surface. Models are refit on a fixed rolling window and cadence; at every decision point the model has seen only information available at that moment. There is no discretionary override — the system's decisions are the strategy.
Scheduled events. The program does not stand aside for the economic calendar. It carries one calendar rule, on one sleeve: Sleeve C opens no new overnight position on the afternoon of a Federal Reserve decision day, because research found the premium on offer after an announcement too thin for the move that tends to follow it. Everything else trades as normal. Sleeves A and B take decision days on their ordinary signals, and a Sleeve C position opened the afternoon before settles at the decision day's close — so the program is routinely in the market through the announcement itself: on 30 of the 34 decision days since the simulation's first trade, 54 structures in all, each defined-risk like every other position. No other scheduled release is filtered.
Development followed a research ledger discipline: every candidate change was adopted or rejected on pre-registered criteria against the full simulation frame, with fills and fees modeled, and rejected variants are retained in the ledger with their verdicts. The production configuration is frozen; changes require re-verification against the champion's full decision log. Every possible precaution was taken to mitigate overfitting, including measuring robustness of the edge across time, and using a validation set and out-of-sample period in order to avoid building the model around random noise.
A note on disclosure. This brief intentionally omits the information that would allow reconstruction: exact decision times, strike- and width-selection rules, the feature set, model hyper-parameters, thresholds, and sizing-tier conditions. Clients allocate to the program's process and controls, which we document fully; the mechanics remain proprietary.
The record below is a single, fully specified simulation of the production configuration over October 14, 2021 → August 13, 2026, at $200,000 initial capital and the standard risk dial, with Interactive Brokers index-option commissions modeled ($16,722 paid in simulation). The starting capital is chosen so that the risk dial, not the one-unit minimum, sets position size from the first trade — the same code on $100,000 runs its first two years at the floor and prints a deeper 2022 drawdown (§VI). It is not a curve fit to look good, it is the production code run over five years of history — through the 2022 bear market, the 2023 melt-up, the 2024 vol events, and the 2025–26 regime — one configuration, never re-tuned mid-frame. Note what the frame contains: a 68.1% compound annual growth rate earned at a beta to the S&P 500 of just 0.13. This is what a genuinely independent return stream looks like. The flat opening segment is real: the walk-forward design deploys no capital until its first training window matures in May 2022 — the program would not trade before its models were ready, and it never has.
Every calendar year in the frame finished positive — through a bear market, a melt-up, and two distinct volatility regimes — with the weakest full year at +34.8% and the strongest at +167.2%. Consistency at this level, at 0.21 beta, is the whole argument for the program: it is not a bet on the market going anywhere.
| Year | Return | Ending equity |
|---|---|---|
| 2022 capital at work from May | +1.1% | $202,261 |
| 2023 | +34.8% | $272,580 |
| 2024 | +86.0% | $506,957 |
| 2025 | +167.2% | $1,354,442 |
| 2026 through Aug 13 | +81.9% | $2,463,270 |
| Measure | Value | Measure | Value |
|---|---|---|---|
| Net profit (total) | +1,131.6% | Annualized volatility | 17.3% |
| CAGR | 68.1% | Beta to S&P 500 | 0.13 |
| Sharpe / Sortino | 2.39 / 2.65 | Annualized alpha | 0.41 |
| Probabilistic Sharpe | 99.6% | Information ratio | 1.69 |
| Max drawdown | −12.7% | Portfolio turnover | 1.78%/day |
| Longest drawdown recovery | 150 days | Orders placed | 8,132 |
| Commissions modeled | $16,722 | Structures filled | 1,027 of 1,030 signaled |
Rebuilt from the simulation's own order log: one entry per sleeve per session traded, the structure exactly as the simulation executed it at its modeled fills, settled per unit at the official index print, before sizing. Three of the 1,030 signals in the frame went unfilled in simulation and are excluded. Generic platform "win rate" statistics count individual option legs and misstate combination strategies; the figures below are the correct per-trade record.
| Sleeve | Trades | Win rate | Avg win | Avg loss | Profit factor | Expectancy / trade | Typical max loss / unit | Max loss ÷ max gain |
|---|---|---|---|---|---|---|---|---|
| A — morning same-day | 188 | 62.8% | +$1,266 | −$1,332 | 1.60 | +$299 | $2,650 | 1.1× (0.5–2.0×) |
| B — midday same-day | 460 | 62.8% | +$961 | −$1,094 | 1.49 | +$197 | $3,250 | 1.9× (0.8–3.4×) |
| C — late-day carry | 379 | 62.5% | +$1,226 | −$1,175 | 1.74 | +$326 | $3,700 | 1.4× (1.0–1.7×) |
| Program | 1,027 | 62.7% | +$1,114 | −$1,167 | 1.61 | +$263 | $3,300 | 1.5× (0.8–2.8×) |
Max loss ÷ max gain is the structure's defined maximum loss divided by the credit collected; the median is shown with the 10th–90th percentile range. Max gain is always the credit.
Here is the edge in one line: an at-the-money structure like these settles in profit when the index finishes inside a band roughly one credit wide on either side of the strike, and at the volatility the market is pricing, that is close to a coin flip — a broker's probability-of-profit readout on one of these entries typically shows 45–55%. Across the 1,027 structures the simulation filled, it won 62.7% of them, every loss capped by construction. Note what the table does and does not say. The defined risk is not one-to-one with the reward: the maximum loss runs about 1.4× the credit on the late-day sleeve and about 1.8× on the midday sleeve, where 50-point wings sit against a credit that is often only 15–20 points — a midday entry showing a 2:1 or higher max-loss-to-max-gain ratio at your broker is ordinary, not an error. What keeps realized wins and losses comparable in size — +$1,114 against −$1,167 — is that the average loss stops at about 40% of its defined maximum. Sixty-nine structures (6.7%) did settle at their full maximum; the largest single-structure loss in the log was $4,390 per unit, a late-day structure taking its whole defined maximum in February 2024. The win rate is not bought with tail risk — it is the edge itself. Averages are per unit structure at settlement; account results scale these by position size.
Before any capital traded, the production deployment was verified against the research champion: its full five-year decision log — every day, every sleeve, every structure — was proven byte-identical to the research implementation's before adoption. We have a high degree of confidence that the strategy the production software executes is identical to the one that was backtested.
An eight-session paper phase (Aug 24 – Sep 1, 2026) then reconciled the production pipeline against the champion day by day. On every session with valid data, the production system's decisions matched the simulation's. A second reconciliation over the first live weeks (September 2–18, 2026: thirteen sessions, three sleeves) compared 36 decisions and found 33 identical; the three that differed traced to a live-only calendar entry, a cloud data backfill that moved one model refit boundary, and a last-day data artefact — none to the production software. On every one of the ten live structures with a simulated counterpart at the same strike, the live fill was equal to or better than the fill the simulation had assumed.
Per-trade risk is capped by construction. The maximum loss of every structure is fixed at entry. Two sleeves carry no overnight exposure at all; the late-day sleeve holds one night with both tails wing-protected. Thus, the only instruments traded are iron butterflies on the 0DTE/1DTE with the short strikes directly at the money, and the long strikes at some variable width away. While the exact width of the spread is not fixed like it is for the Flagship, the total amount of risk taken every day is governed by the Risk Dial regardless of how wide or narrow the chosen spread width is that day — with one exception that matters at small sizes. The software never trades fewer than one unit per entry. The record in §IV is simulated from $200,000 so that the dial sets size from the first trade; below roughly $170,000 of allocation at the standard dial, one unit already carries more risk than the dial implies: at the $75,000 minimum, a late-day structure's defined maximum is about 5.5% of the allocation rather than 2.5%, and the sizing tiers multiply that unit. In order to trade this strategy, a minimum of $75K in capital is required to be able to survive normal drawdowns, and clients near the minimum should read the stress numbers below at that effective dial.
Drawdowns are expected and sized for. The simulation's worst peak-to-trough decline was 12.7%, from December 2024 to March 2025, recovered in 103 days; the longest any drawdown took to recover was 150 days, for a shallower 7.5% dip in 2022. Allocation intensity is set per client by a risk dial, and the full dial curve has been measured on the production configuration itself — same five-year frame, same code, only the dial changed:
| Risk dial | Simulated max drawdown | Cumulative return | CAGR | Note |
|---|---|---|---|---|
| Standard (2.5%) | 12.7% longest recovery 150 days | +1,131.6% | 68.1% | Production setting; the record in §IV |
| 2× (5%) | 29.2% longest recovery 150 days | +16,537% | 187.9% | Measured Sep 2026; Sharpe 3.15 |
| 4× (10%) | 55.2% longest recovery 235 days | +903,160% | 557.7% | Measured; offered to no client |
Each row compounds the same $200,000 start with nothing withdrawn. Deployment above the standard dial is by approval only. What the curve shows is that drawdown scales roughly in proportion to the dial — twice the dial produced 2.3 times the drawdown, four times the dial 4.3 times.
Unlike the Flagship strategy, the max drawdown is dependent less on any given trade and more on a sequence of bad trades and a bad trade on a day where the risk was dialed up. Further, unlike the Flagship, each trade's defined maximum loss is of the same order as its maximum gain — typically 1.4× the credit on the late-day sleeve and 1.8× on the midday sleeve (§IV), never the several-to-one of a far-out-of-the-money credit spread — and the edge comes from the win rate being significantly higher than what the market prices in. That is, the market prices these at-the-money structures at roughly even odds, and the simulation won 63% of them. This strategy was specifically created as a diversifier and complement to the Flagship strategy, and is in no way meant to replace or supersede it.
Stress numbers. On days that meet a sizing-tier rule, position size steps up to 2× the base unit. A maximum-size position taking its full defined loss in a single session is a realistic tail event — about $8,400 per base unit on the late-day sleeve. At a $170,000 allocation that is 5% in one day; at the $75,000 minimum, where the one-unit floor applies, it is about 11%; and a session on which the late-day and midday sleeves both take a doubled maximum is about $15,600 per base unit, roughly 21% at the minimum. The simulation's 12.7% drawdown already contains such exposures at the standard dial — 69 of its 1,027 structures settled at their full defined maximum — but it does not contain the floor's amplification: the same code run from $100,000, where the first two years sit at the floor, prints a 15.0% drawdown in June 2022. Size allocations with drawdowns in mind.
The program is licensed as software. It connects to your own brokerage account and is managed from your QuantGlide dashboard. No discretionary asset management is performed, and QuantGlide never requests, receives, or holds your brokerage username, password, or other broker login credentials. What QuantGlide does hold, for each connected account, is the access token your authorization issues — a credential that lets the software transmit orders in that account until you revoke it, and that confers no other access. It cannot be used to deposit, withdraw, or transfer cash or securities. Those tokens are held encrypted at rest, and you can revoke them at any time at your brokerage or in Trade It.
Apex runs at the same brokerages as the Flagship, over the same connection paths: tastytrade through our direct integration, and every other supported brokerage through our connectivity partner, Trade It. If you already run the Flagship, we can reuse any accounts you already have connected — and your existing QuantGlide dashboard login. For an existing customer, deployment is a configuration step, not an onboarding.
Running the Apex algorithm will NOT affect revenue share or billing for the Flagship algorithm in any way, or affect its trade execution. You can run Apex in the exact same brokerage account, or in any other supported Margin brokerage account with options level 3. Preparing a suitable brokerage account, and acquiring the options trading level the system requires, is solely your responsibility.
If either connection path is ever interrupted, your license term is extended day for day for the time the software could not trade. The terms sheet spells out how that works.
Your assets never leave your custody, every order is visible at your broker in real time, and you can stop the software at any moment. The dashboard carries an off switch, which stops it opening any new position from the next signal onward. It does not recall an order already working in the market — that keeps repricing until it fills or is cancelled at that book's scheduled cut-off — so to stop a working order, revoke the connection outside the dashboard, at your brokerage directly or in Trade It if that is where you linked the account. Neither an off switch nor a revocation closes a position that is already open: only you can do that, in your own account.
Your installation is configured with its own capital allocation and risk dial (§VI), set with you at deployment. One operating limit applies to every installation:
There is no account value below which the software stops trading on its own. The risk dial and the off switch are the only controls, and both are yours.
Tell us the allocation you would license and your quote arrives by email within a minute: the full fee schedule, your number, and Ben's direct line for anything you want to talk through first. The commercial terms are in the terms sheet you will countersign.
Request a quote at www.quantglide.com/apex, or call Ben at (281) 849-3448.
Prefer to talk first? Call or text Ben at (281) 849-3448, or write to ben.margaryan@quantglide.com.
Simulated performance. All performance shown is simulated. Simulated results have inherent limitations: they are prepared with the benefit of hindsight, do not reflect how economic and market conditions might have affected decisions, and are not actual trading. Simulated fills may differ from achievable fills. Commissions are modeled; no representation is made that any account will achieve similar results.
Fees not reflected. Performance shown is net of modeled brokerage commissions and gross of the annual software license fee, which is payable in advance regardless of trading results. The fee's effect on returns is proportionally larger at smaller allocations; evaluate the program net of the fee at the allocation you intend to run.
Risk of loss. Options trading involves risk of loss and is not suitable for all investors. Every structure's maximum loss is defined at entry, but losing sequences will occur and drawdowns larger than the simulated maximum are possible. Past performance, simulated or actual, does not guarantee future results.
Third-party connectivity. Except on tastytrade, the link between the software and the client's brokerage account is provided by Trade It, an independent third party. Where a brokerage's linking flow requires login credentials, those credentials are supplied to and held by Trade It, not by QuantGlide, and their handling is governed by Trade It's own terms and privacy policy. An interruption of that link can stop the software from trading a connected account until connectivity is re-established or the client connects an account at another supported brokerage.
QuantGlide
QuantGlide is a research and automation tool for traders. It does not provide investment advice, solicit capital, or operate as a broker-dealer or investment advisor. Trading options, especially on SPX, involves substantial risk and is not suitable for every investor. You can lose your entire initial investment. Figures not explicitly labelled as live are hypothetical, backtested results; simulated results have inherent limitations and may differ materially from actual results. Past performance, whether actual or simulated, is not indicative of future results.
© QuantGlide. All rights reserved.