Soomario Strategies is a systematic trading platform operated by a single principal, with all production logic running on dedicated infrastructure across three execution venues — Hyperliquid for crypto perpetual futures, Binance USD-M futures for crypto and US equity perpetuals, and Solana spot for tokenized equities. The platform comprises seven strategies designed around a common thesis: most edge in retail-accessible markets is already in public data — observable in liquidation books, unlock calendars, options chains, and price action — and the structural inefficiency lies almost entirely in how few participants apply that data systematically, across many assets, with discipline, in the face of regimes that punish discretion.
The platform addresses two distinct audiences. Six strategies are vault products: capital is contributed to an on-chain or exchange-side vault, the strategy executes against pooled equity, and a 10% performance fee is taken on net profits — no management fee, no lockup, no aggregator. The ninth, Accumulator, is a subscription signal product at $7 per month that delivers daily quantitative buy recommendations to retail investors who execute the trades themselves. Both audiences receive the same standard of analytical rigour. They differ only in custody, capital model, and degree of automation.
This document is a unified credential for the platform. It describes how the strategies think, what they look at, where their respective edges sit, and — equally important — what we do not claim. We do not publish exact parameters. We do not publish backtest summary statistics in the body of this paper. We do not promise returns. What follows is a full description of the analytical vocabulary the platform speaks, the structural reasoning behind each strategy, and the controls that bound risk at every layer.
Vault Products — 10% performance fee, no management fee
| Strategy | Design Principle | Venue | Asset Class | Status |
|---|---|---|---|---|
| Gladius | Mean reversion at exhausted liquidation cascades, multi-signal confluence | Binance USD-M | 14 crypto perpetuals | ● Live · June 2026 |
| Elite | Discretionary vault managed by the founder, with small automated positions alongside | Hyperliquid | Varies — chosen by the manager | Closed to new deposits · vault since March 2026 |
| Libration | Multi-coin momentum swing from one shared account, one position per coin | Hyperliquid | 11 crypto perpetuals | ● Live |
| Hype Bot | The Libration engine concentrated on a single market, one position at a time | Hyperliquid | HYPE perpetual | ● Live · November 2025 |
| Aphelion | Long-bias accumulator exited by a recycling tail trail | Binance USD-M | 19 US equity perpetuals | ● Live · April 2026 |
| Rotation | Long-only unlevered rotation of tokenized US equities, permanent core plus scored slots | Solana spot (Jupiter) | Tokenized US equities | ● Live |
Subscription Product — $7/month, 7-day free trial via Whop
| Product | Design Principle | Delivery | Asset Coverage | Status |
|---|---|---|---|---|
| Accumulator | Quantitative signals to scale DCA buys by oversold opportunity, personalised to budget | Telegram + Web Push | 14 curated equities & crypto + unlimited watchlist | ● Live · February 2026 |
The six vault strategies are decoupled by design. They use different signal categories, target different inefficiencies, and reach their drawdowns at different times. A subscriber to the Accumulator subscription is also a qualified prospect for the vault products — and the vault products' transparency (every position visible on-chain, in real time) is itself the cleanest reference any subscription buyer could ask for.
The data is public. The systematic application is rare.
The platform's commitments are not aspirations — they are constraints encoded into every strategy on the platform. The list below is what we do not change in any individual strategy's design without an explicit reason and a documented review.
Every signal that opens or closes a position is generated by code. The operator's discretion is reserved for parameter review, universe additions and removals, and strategy-level governance — not for individual trade decisions. Discretionary intervention introduces a class of failure that is invisible in backtest and impossible to audit in production. The platform avoids it by structure.
Most retail-accessible automated strategies run the same settings across every asset they touch. One indicator period. One stop band. One leverage tier. This is computationally cheap and produces a product that can be deployed anywhere — and it leaves real edge unclaimed on every asset that doesn't match the implicit reference. Soomario strategies that span multiple markets calibrate per asset. A high-velocity meme-style coin runs on different settings than an institutionally-traded major. The discipline is expensive in research time. It is cheap in capital.
Several strategies place their primary risk controls — stops, defense limits, take-profit triggers — directly on the exchange's matching engine, where they fire whether the bot is alive or not. A stop-loss that depends on the bot's heartbeat has a single point of failure between signal and exit. Whenever the venue allows it (Hyperliquid and Binance both do), our stops live there. The bot's monitoring loop is then a redundancy layer, not a sole one.
Custody is pushed as close to the depositor as each venue allows. The Hyperliquid strategies run on the protocol's native vault contract, where capital is custodied on-chain and every entry, exit, and balance change is verifiable in real time. Rotation is fully self-custodied in a wallet, holding spot tokens with no leverage. The Binance strategies are distributed by copy trading, so follower funds never leave the follower's own account. Investors do not rely on an LP statement, a quarterly report, or our word for anything.
We publish vault addresses, dashboards, infrastructure stack, design principles, and the full categorical description of every analytical input the platform uses. We do not publish which exact indicators a strategy reads, their periods, timeframes or threshold values, the timing of entries and exits, weighting matrices, leverage tiers, DCA percentages, hard-stop bands, or scoring formulas. Sophisticated investors do not need these to evaluate a strategy. Less-sophisticated investors are not protected by them. And the platform's edge is durable only to the extent it is not handed away. The rule is consistent: describe the discipline, not the digits.
Position-size caps, leverage ceilings, no-stacking rules, kill-switch thresholds, capital caps per sleeve — these are not knobs the operator turns optimistically based on recent performance. They are constraints encoded into the system that bound the worst possible outcome on any single trade and on the portfolio in aggregate. A strategy that would be more profitable with looser constraints, but is engineered to refuse to loosen them, is a strategy whose worst-case behaviour is computable in advance.
Every strategy that automates anything important automates it twice. Signal generation is decoupled from execution. Bot-side stops back exchange-side stops. State is reconciled against ground-truth exchange data on every cycle. Two independent components must both fail before a position runs unmanaged. This adds moving parts, and it is worth every line of code that implements it.
It is not novelty-seeking. None of the building blocks the platform uses are new. Momentum oscillators, moving averages, statistical normalisation, liquidation maps, unlock calendars, and trend filters have all existed for years. What is new is not the parts but the discipline of combining them, applying them across many assets, encoding the constraints in production, and refusing to tune them retroactively to flatter recent performance. The edge is operational, not informational.
This section is the platform's vocabulary. It defines every category of analytical input the strategies use, and what each reveals about a market. Which inputs a given strategy reads, and how, is part of what stays private. The categories are not weighted equally across strategies — some strategies use one category as their dominant signal and the others as confirmation; some use four or five categories layered as gates that all must agree before capital deploys.
What this section does not do is publish exact indicator periods, weights, thresholds, or scoring formulas. The categorical description below is sufficient for a sophisticated investor to evaluate whether the platform's analytical stance fits their understanding of the markets it trades. The exact numerical implementation is not.
If you are reading this section to find a parameter table, you will not find one. You will find every category of analysis the platform applies, and the rationale for applying it.
Momentum oscillators are the platform's most widely deployed category. The best-known member of the family, the Relative Strength Index, measures the magnitude of recent gains against recent losses on a 0–100 scale, identifying when an asset is statistically stretched in one direction. Wherever the platform computes an oscillator it uses the standard reference calculation, so that any signal that fires on the live engine fires on a chart the operator can manually re-derive.
Oscillators are read differently depending on what a strategy is built to capture. A mean-reversion design treats an extreme reading as evidence that a move is exhausted. A momentum design treats a decisive transition as a directional re-anchoring after consolidation, not a contrarian fade. An accumulation design treats a deeply oversold reading as one input to a multi-signal filter. Which oscillator each strategy reads, on what horizon and at what levels, is calibrated per asset and kept private; the platform does not run one setting on every market.
Volume tells the platform what level of conviction a price move is happening on. A move on average volume is more likely to be a normal mean-reverting fluctuation; the same percentage move on volume two or three standard deviations above the recent baseline is more likely to be a flow event — capitulation, forced liquidation, or a regime-changing shift in positioning. The platform measures volume anomaly through statistical normalisation, which compares the current bar against the asset's own recent history rather than against a universal scale.
Gladius uses a custom volume-based measure of forced flow as one leg of its confluence requirement. The platform does not use traditional named "money flow" indicators (CMF, MFI) directly; a normalised formulation is more flexible and produces fewer false signals across heterogeneous assets.
The workhorse of this category is the Z-score: the number of standard deviations a current observation sits from the mean of a chosen lookback window. Statistical normalisation is used extensively because it is the most disciplined way to ask: "is what's happening right now actually unusual?" A 5% drop on an asset that routinely moves 8% in a day is a normal move; the same 5% drop on an asset that rarely moves more than 2% is a regime event. Absolute percentage thresholds cannot make this distinction. Normalised readings can.
The platform applies statistical normalisation to volume (as described above), to short-window price excursions, and to longer-horizon price distributions. A longer-horizon reading can require genuine statistical overextension before capital deploys — filtering out cases where price looks weak in the short run but sits well within the asset's longer-term distribution. Normalisation is most informative on crypto, where the wider volatility regime makes a raw percentage move hard to read.
The platform reads trend direction through two distinct lenses. Long-period moving averages define an asset's secular position — is current price extended below the long trend (an accumulation opportunity) or above it (a distribution context)? The platform's accumulation products use moving-average extension as a gate; they will not signal an accumulation unless the asset is meaningfully below its long-period reference.
Volatility-adaptive trend filters draw a moving stop line that flips when price breaks through it decisively. On their own they whipsaw in sideways markets, which is why the platform never uses a trend filter as a sole signal — but as a confirmation layer on top of momentum and volume conditions, it provides directional clarity. Trend direction is a context filter, not a primary entry signal anywhere on the platform.
In perpetual futures markets, open interest reveals the size of leveraged positioning at every venue, and funding rates reveal which side of the book is paying the other to hold their positions. These are direct readings of crowdedness — they tell the platform whether longs or shorts are the structurally fragile side of any given market.
Gladius uses the long/short open-interest skew as a secondary input — a short-heavy book is structurally favourable for accumulation entries, since shorts covering against a bounce supply the buying pressure that translates a DCA fill into realised P&L. It is not a primary signal; it is a confirmation layer that flag when a primary signal is structurally well-supported.
This is the platform's most distinctive analytical category, and it is the one that several strategies are explicitly built around. Liquidation density data — derived from aggregated open-interest readings across multiple major centralised exchanges — maps the price levels at which clusters of leveraged positions would be forcibly closed if hit. These clusters are not theoretical; they are real positions held by real traders, and when price reaches the cluster, the resulting forced flow creates a predictable price magnet.
The platform computes a Long Max Pain level (the price below current market where leveraged longs are most concentrated) and a Short Max Pain level (the corresponding cluster above). Gladius uses the Long Max Pain level as the primary anchor for its entries. Rotation's crypto sleeve uses the same anchor to place its entries. Gladius is named for the broader thesis that exhausted cascades reverse; it reads the underlying liquidation data both as an entry anchor and to inform asset selection.
The data itself is publicly observable on services like Coinglass. The platform's edge is in computing the zones from primary exchange feeds rather than scraping aggregators, in tracking its universe in parallel, and in trading the zones as a portfolio rather than as anecdotal individual setups.
The Alternative.me Fear & Greed index is a daily synthetic reading of broad crypto market sentiment, computed from volatility, momentum, social media volume, dominance, and survey inputs. It is publicly available, updated daily, and meaningfully predictive of regime behaviour even when it is poor at predicting next-day price.
Rotation displays the index on its dashboard for context; no current strategy trades on it.
Distinct from a simple moving-average crossover, price-vs-trend extension measures the magnitude by which current price sits below (or above) its long-period reference line. The reading is normalised to give an apples-to-apples score across assets with different volatility profiles. A volatile crypto asset sitting 8% below its long-period average is a different signal from a low-volatility large-cap equity sitting 8% below its own long-period reference.
Aphelion combines extension below a long-period average with a short-term oversold reading into a single score, and only a sufficiently strong score fires a buy. The Accumulator product uses extension below the long-term trend as one of the weighted inputs into its 0–100 strength score. Rotation's Fastest Horse equity sleeve uses moving-average context as part of its technical timing layer to identify favourable entries on otherwise-fundamentally-screened candidates. Across all uses, the principle is consistent: extension is a measure of how far the asset is from its trend, not just which side of the trend it sits on.
Volatility is not a signal the platform trades on directly — it is a context every strategy adapts to. The implementations vary. Gladius calibrates its signal settings and hard-stop bands per coin against each asset's measured volatility profile; a high-velocity meme-style asset gets faster, tighter parameters than an institutionally-traded major. Libration sets its trailing exits per coin, so wider-ranging assets get wider trails. Rotation's crypto core sleeve spaces its DCA levels by each asset's own volatility rather than by fixed percentages.
The platform does not trade volatility expansion or compression as a primary signal. It trades the underlying inefficiencies through the lens of each asset's volatility, which is why per-asset tuning is one of the platform's core philosophical commitments. A single set of parameters across heterogeneous volatility regimes is, by definition, suboptimal on every asset. The platform refuses that shortcut.
A signal that fires on a single horizon — say, a short-term oscillator extreme — is more vulnerable to noise than the same signal that fires when shorter, medium, and longer timeframes all agree. The platform reads multi-timeframe confluence in two ways. First, by requiring agreement across horizons before acting: Gladius requires its momentum read to agree across horizons, so a short-horizon extreme that the longer horizon does not confirm is not traded. Second, by separating cadences within a strategy: Rotation keeps its rebalancing cadence deliberately slower than its price, technical and risk checks, so a single noisy session cannot rewrite the portfolio. Confluence is treated as evidence that a signal is durable; absence of confluence is treated as a reason to wait.
The platform treats certain calendar events as primary signals on the strategies that are designed to exploit them. Token unlocks — the scheduled release of previously-locked supply to teams, investors, or ecosystems — are publicly disclosed in every project's tokenomics, but most participants either ignore them or trade them anecdotally. The platform maintains a curated unlock calendar weighted by recipient class (team and investor unlocks score higher than ecosystem ones), proximity, and size relative to circulating supply. It is used as contextual risk input on asset selection rather than as a standalone entry trigger.
Calendar events are treated as deterministic context the strategy plans around, not as lagging indicators.
This is the most distinctive analytical category on the platform, and it is used by exactly one sleeve of one strategy — by design. Rotation's Fastest Horse equity rotation engine is built on a deliberate premise: the highest-return equity allocation over a three-to-five-year window is rarely the stock with the best recent momentum or the cheapest current multiple — it is the stock with the largest credibly-modelled upside to a long-horizon thesis target.
The dominant scoring axis is therefore "visionary upside" — sourced from credible thesis-driven research (specialist analysts publishing multi-year price targets attached to a documented thesis, profit-growth aggregates from Wall Street consensus over multi-year windows). Visionary targets are the heaviest score axis specifically because they are the hardest to fake. Wall Street consensus drifts with the news cycle. Technical indicators flip on any multi-day swing. A genuine three-to-five-year price target — sourced, dated, attached to a published thesis — changes slowly and reflects real conviction. The category is mentioned here because a sophisticated reader of this whitepaper would notice its absence from a purely technical analytical toolkit and reasonably ask whether the platform handles long-horizon equity exposure at all. It does. It does so on exactly one sleeve, with an explicit buffer that prevents thrashing on small score changes. This sleeve's exact scoring weights, buffer, and slot count are not published.
Every strategy on the platform reconciles its internal position state against ground-truth exchange data on every cycle. This is not an analytical input to a trading decision — it is the substrate that makes any other analytical input trustworthy. In algorithmic trading, the bug that destroys an account is rarely a bad signal. It is the bug where the bot believes it has a position the exchange does not recognise, or believes it does not have one when it does.
Aphelion reconciles its in-universe positions against Binance every 60-second tick. Gladius reconciles against exchange ground truth every cycle and refuses to act on a stale read. Rotation sizes every sell from the on-chain balance rather than its internal ledger, so the chain is always the authority on what is actually held. State integrity is a category because we treat it as one. It deserves the prominence.
The thirteen categories above are the full vocabulary of the platform. No strategy uses all of them; no category is used by every strategy. What every strategy shares is the principle that multiple categorical signals must align before capital deploys — single-category signals are noise, and the discipline of demanding agreement across categories is what converts public observations into a tradeable edge.
The next nine chapters describe each strategy on the platform in turn — its identity, status, venue, asset universe, the structural inefficiency it targets, the analytical categories from Section III that it actually uses, the lifecycle by which positions are opened and closed, the risk controls bounding worst-case behaviour, and the honest limits we acknowledge for each.
The chapters are deliberately uniform in structure to make cross-strategy comparison easy. Reading them in sequence is the fastest way to build an integrated picture of the platform; reading any one in isolation is sufficient to evaluate that strategy on its own terms.
Each strategy is a complete answer to a different question. Together they form a portfolio whose drawdowns are not synchronised.
The order of the chapters is roughly thematic rather than chronological — Gladius, Elite, Libration and the Hype Bot (the crypto perpetual strategies), then Aphelion (the equity accumulator), then Rotation (the unlevered tokenized-equity book), and finally Accumulator (the subscription product, treated separately because its mechanics and audience differ).
Gladius was previously published as Max Pain, and briefly as Zones. Those names described overlapping work on the same thesis. They are now consolidated under Gladius, which is the only live implementation.
In leveraged perpetual markets, forced liquidation flow has a finite supply at any price zone. Once it is exhausted, the asymmetry that drove the cascade disappears and price reverts. When over-leveraged longs cluster at a level and that level breaks, forced selling accelerates the move briefly and then runs out — long-side liquidations finish, organic flow re-balances, and the asset reverts off its overshoot. The same logic applies symmetrically to upside squeezes. Most participants either chase the cascade direction or trade reversals anecdotally on a single chart. Gladius demands confluence across momentum, volume anomaly, and trend reversal, applied across its universe every bar.
Gladius combines momentum oscillators read across more than one horizon, a volume-anomaly liquidation proxy (a normalised measure of price excursion against recent baseline — high readings mark forced-flow events rather than organic positioning), and trend direction confirmation. The indicators are combined as a confluence requirement rather than a weighted sum: each alone produces too many false signals to trade, and the intersection is rare. The rare bars are the set the strategy trades. The exact settings are private; every trade is published.
The single most important architectural decision is that the backtest and the live bot import the same signal code. Backtest and live cannot compute a signal differently, because there is only one implementation. This was a deliberate response to an earlier generation of the strategy in which locally-computed indicator math produced live trades that did not match the backtest they were sized against.
A confirmed signal opens a position anchored to the long max-pain level, scaling in rather than committing everything at a single price. Profit is taken in stages, with a trailing exit on the remainder. Every position carries a hard stop placed on the exchange at entry. There is no averaging into losers beyond the pre-declared entry plan, and no stacking of same-direction positions on one asset.
Hard stops are rested on the exchange matching engine, so they fire whether the bot is alive or not. A reconciliation loop checks internal state against exchange ground truth every cycle and refuses to act on a stale read. The backtest engine models intrabar stop hits explicitly — an earlier trailing-stop-only model reported a 0.00% average loss precisely because it never modelled the hard stop being hit inside a bar, and that flattered the results substantially.
The live record is short and currently negative. As of the most recent dashboard reading, Gladius has closed 32 positions since late June 2026 at a 56.2% win rate, a profit factor of 0.91, and a cumulative realised return of −15.6%. A win rate above 50% with a profit factor below 1.0 is the signature of a strategy whose average loss exceeds its average win — which is exactly what the figures show. These numbers are published as they stand, and the live dashboard is public and continuously updated.
The walk-forward gate exists to make this measurable rather than arguable. On deliberately un-tuned baseline parameters the gate declined to certify any asset for live capital, which is the gate performing its function. Signal evaluation is on bar close, not high-frequency. Parameters are not stable indefinitely. The strategy is not foolproof.
Closed to new deposits; no longer offered as a service. AUREUS is moving its allocation from Elite into Libration.
Elite is a Hyperliquid vault traded by Soomario's founder. He decides what the vault holds, how large each position is and how much leverage it uses, and he trades by hand on the vault's behalf. A few small automated positions run alongside the discretionary book. The founder keeps his own capital in the vault, so his results move with the depositors'.
A deposit buys a pro-rata share of the vault, held in Hyperliquid's native vault contract. Gains and losses are shared in proportion to each stake. The manager takes 10% of profits and nothing else: no management, deposit or withdrawal fees. Each deposit is locked for one day, after which it can be withdrawn at any time. Every position and trade is public on Hyperliquid, and the live dashboard shows the vault's value, time-weighted returns, drawdown, positions and recent trades.
Not systematic — most of the book depends on one person's judgement, so there is no backtest that describes what Elite will do. Not market-neutral — the vault takes directional, leveraged positions and can be concentrated. Not a balanced portfolio — markets, direction and size change with the manager's view. Past results, shown on the live dashboard, do not predict future ones.
A simple momentum signal is the most widely known kind of signal in retail trading, and on any single chart it is close to worthless — it fires constantly, and most fires are noise. What changes its character is breadth and discipline: evaluating the same rule across a basket of genuinely volatile assets, on completed bars only, with one position per asset, a stop that never moves against the position, and a portfolio-level halt that stops the book adding risk on a bad day. The edge is not the indicator. It is the refusal to take the same signal discretionarily.
A single proprietary momentum signal, applied identically across the basket. Longs and shorts use deliberately different conditions, since the two directions do not behave identically in crypto perpetuals. Entries are evaluated only on completed bars, with a per-coin guard preventing a signal firing twice inside the same bar. Universe membership is refreshed periodically on a volatility-and-liquidity filter, which is a maintenance tool rather than a trading signal. The exact settings are private; every trade is published.
One position per coin, no pyramiding. Each position is sized as a fixed fraction of current equity, which means the book scales with the account rather than with conviction. Two exits run in parallel: a tight trailing take-profit that locks in gains once a trade is working, and a wide hard stop resting natively on the exchange that is never moved against the position. Maximum concurrency is a derived number — leverage divided by notional fraction — rather than an independent knob, so the book cannot quietly become over-leveraged by adding names.
The hard stop sits far inside the liquidation buffer at the operating leverage, so the stop fires long before margin is at risk. There is no daily drawdown halt: the hard stops cap each loss, and accounts run through the engine carry one catastrophe floor: if the account falls 60% below its starting balance, the engine closes its positions and stops until it has been reviewed. All positions run under isolated margin from one shared account, so a single asset cannot consume the whole book's margin.
The published equity curve is flow-neutral: it plots a starting baseline plus cumulative realised profit plus open unrealised profit, never raw wallet value. Deposits and withdrawals therefore cannot appear as performance. This matters more than it sounds — raw-wallet curves are the single most common way a self-reported track record flatters itself, and the property holds if and when the strategy is wrapped as a pooled vault.
The strategy is validated in backtest and walk-forward, but the open question on any tight-trailing system is whether real slippage preserves expectancy — a trailing stop this tight is exactly the kind of parameter that survives on paper and degrades in execution. It is not market-neutral and not a hedge; in a coordinated sell-off, correlated long positions can stop out together. It runs a small, deliberately volatile universe, which is the source of both its opportunity and its variance.
The Hype Bot runs the same engine as Libration against a single market. Where Libration spreads one rule across a basket, this book concentrates it on the one asset the platform has the longest continuous read on. The signal is deliberately ordinary — a momentum signal of the kind every retail chart can display, and on any single chart that kind of signal is close to worthless. What is uncommon is the refusal to act on anything else: no intra-bar entries, no averaging down, no second position, and no override when the chart looks tempting.
A proprietary momentum signal on HYPE, evaluated only on completed bars, with a guard preventing a signal firing twice inside the same bar. Longs and shorts use deliberately different conditions, since the two directions do not behave identically in crypto perpetuals. Nothing else is consulted. The exact settings are private; every trade is published.
One position at a time, no pyramiding. A signal arriving while a position is open is skipped, not acted on — the book waits for the current trade to resolve rather than reversing into it. Two exits run in parallel: a tight trailing take-profit that follows a winning trade, and a wide hard stop resting natively on the exchange that is never moved against the position.
When the trailing take-profit begins to follow price matters more than any other setting in the book. This is not a detail — it is the edge. Tested across the full HYPE history at one-minute resolution, a trail that engages too eagerly turns a clearly profitable rule into a net loser on identical entries, because a trade that ticks marginally into profit in its first minutes is stopped out on the first wobble, booking a rounding error that fees turn into a loss. Same signal, same trail width, same hard stop. The only difference is patience; the exact rule is private.
The hard stop rests on the exchange matching engine, so it fires whether the bot is alive or not, and it sits far inside the liquidation buffer at the operating leverage. There is no daily drawdown halt; accounts run through the engine carry one catastrophe floor: if the account falls 60% below its starting balance, the engine closes its positions and stops until it has been reviewed. Because a single-asset book concentrates its whole allocation in one position, a hard stop costs materially more of the account here than the same stop does inside the multi-coin vault — the sizing is set with that asymmetry in mind rather than inherited from the basket.
The published record is simulated. The trade history shown on the dashboard comes from the strategy tester on both sides of the go-live marker; the segment after November 2025 records when real capital began following the signals, not a set of realised fills. The tester prices each bar from a limited number of points, which under-triggers a stop as tight as this one — at full one-minute resolution the same rules return materially less per trade than the tester reports. The figures published are the optimistic end of the range and are labelled as such.
There is no diversification: when HYPE is untradeable the book has nothing else to do. HYPE also rose roughly 450% across the period covered, and a long-biased engine on a trending asset flatters itself — this record has not been tested through a sustained downtrend in its only market. And like every tight-trail system on the platform, it wins often and small while losing rarely and large; that shape survives only while the win rate holds, and inverts quickly if it slips.
The signals Aphelion trades are visible on every charting platform. What is uncommon is acting on every confirmation across nineteen assets, every session, with identical sizing — and then holding the exit rule fixed while the position is still working. The structural inefficiency it exploits is the gap between what is statistically oversold and what most participants find emotionally tradeable. Most retail traders avoid these setups because they look terrible on the chart. Most institutional accumulators do not run leveraged perpetuals on single-name equities.
A weighted score of a momentum oscillator and price-versus-trend extension below a long-period average, the same template for every stock. A minimum score gates whether a signal fires at all; deeper and longer-awaited dips buy more. Below that threshold nothing happens, however tempting the chart looks. The exact weights and settings are private; every trade is published.
Confirmed signals buy in scaled increments against an accruing budget. Positions are then held through normal volatility and exited by a recycling tail trail: the strategy takes profit only after a large run, and then on part of the position, leaving the remainder working and starting a fresh cycle from there. The average entry is never rewritten by this process, so the position's true cost basis stays honest across repeated trims. The strategy is accumulate-and-trim, not predict-the-top.
Earlier published versions of Aphelion described a pre-positioned limit-buy “defense layer” that would add to a position approaching its liquidation zone. That mechanism has been removed. Monte Carlo testing showed it amplified losses on volatile names rather than damping them: it added size precisely where the distribution's left tail was fattest. The current configuration runs no defense layer and no take-profit ladder — only the recycling tail trail. This is a case where the honest result contradicted the more attractive-sounding design, and the design lost.
A per-trade cap expressed as a fraction of equity, a hard daily notional ceiling that stops the book adding size after a set amount of entries, and a two-stage kill switch on daily realised loss and on aggregate unrealised loss. A cooldown period blocks re-entry on any asset following a liquidation event. Margin is isolated per position, so no single name can consume the account.
Long-bias by construction: in a generalised sell-off where every name falls together, Aphelion accumulates more aggressively, which concentrates exposure to systemic moves. That is the design and the risk. The live track record is short. It is not market-timing — the engine does not predict tops or bottoms, and long sideways stretches produce slow accumulation without exciting moves. Leveraged perpetuals on single-name equities carry liquidation risk that a spot position does not.
Tokenized equities put US single-name exposure on-chain, tradeable spot, around a self-custodied wallet rather than a brokerage relationship. That removes the two things that make leveraged equity strategies fragile — margin and liquidation — and leaves a simpler question: which names deserve the capital now. Rotation answers it with a scoring model rather than a view, holds a permanent core in its highest-conviction name, and rotates the remainder across a small set of scored mega-caps.
Candidates are ranked on three weighted layers: long-horizon visionary upside drawn from thesis-driven research, near-term analyst consensus refreshed regularly, and technical entry timing from momentum and trend indicators. The visionary layer carries the most weight because it is the hardest to fake and the slowest to change. The composite ranks the universe; the top names fill a fixed number of rotation slots alongside the permanent core. Membership and weights are reviewed on a fixed schedule, with a buffer that prevents churn when two candidates are close.
Earlier versions of Rotation ran a third sleeve of event-driven crypto shorts alongside the equity and crypto-long sleeves. That sleeve has been removed. Six days of live data showed it a net drag: the stop-losses hit while only the shallow grid trims paid, and a pause gate fired repeatedly during precisely the regime the sleeve existed to exploit — closing it during its own best conditions. The freed allocation was redistributed to the long sleeves. Rotation is now long-only.
Orders route through Jupiter's aggregated swap API — buys swap stablecoin into the token, sells swap back. Three safety rails bound execution. Token mints are resolved only from an explicit configured mapping or a verified-token registry with an exact symbol match; an ambiguous or unverified symbol is refused rather than guessed, because trading a wrong mint is unrecoverable. Orders are confined to regular market hours by default, since tokenized equities track their underlying's NAV and off-hours on-chain liquidity is thin. Quotes above a maximum price-impact threshold are rejected outright. Sells size from the on-chain balance rather than the internal ledger, so the chain is always the authority on what is actually held.
No leverage anywhere in the sleeve, so there is no liquidation risk and no margin call — the worst case is the drawdown of the underlying equities themselves. Position sizing is capped per name, with a ceiling on the permanent core so conviction cannot become concentration by drift.
Rotation's current record starts on 8 September 2026, when the book moved back to Solana; earlier records were reset and are not cited. As of 17 October 2026 it was down 3.1% against SPY −0.7% and QQQ −1.2% over the same window, with a 7.3% maximum drawdown. Six weeks is far too short to judge, and the record has not yet been tested by a sustained drawdown in its underlying names. Rotation is not yet open to outside capital. Annualised figures derived from a window this short are not meaningful and are not claimed. Rotation is net long by construction and is not a hedge against anything. Tokenized equity wrappers additionally carry issuer and redemption risk that direct share ownership does not.
Accumulator is the only Soomario product that is not a vault. Subscribers receive daily quantitative buy recommendations and execute the trades themselves on whatever brokerage or exchange they prefer. The product retains no custody of subscriber funds, executes nothing on subscribers' behalf, and is positioned as market commentary and educational content — not as an investment adviser. The audience is retail investors who want to time their dollar-cost-averaging purchases with quantitative discipline rather than buying fixed amounts on fixed schedules.
Pure dollar-cost averaging works mechanically — volatility smooths out, emotional timing mistakes are avoided, compound growth does the heavy lifting. But it leaves money on the table. It buys the same amount whether the asset is 50% above its long-term average or 30% below. A mathematically better approach — buy more when oversold, less or nothing when overextended — is widely understood but almost never implemented, because doing so correctly requires reliable price data across multiple asset classes, consistent calculation of technical indicators, and a disciplined scoring model that rolls indicators into a single actionable number, delivered daily without the user having to open a charting platform. Most retail investors either do not have time, do not have the skill, or do not trust themselves to follow through. The result is a market of disciplined investors executing an undisciplined version of a disciplined strategy.
Accumulator computes three indicators per asset for each daily signal. Momentum oscillators detect short-term oversold conditions, calibrated separately for stocks and crypto. Price-vs-trend extension detects mean-reversion opportunities when price is significantly below its long-term trend. Statistical anomaly detection flags statistical outliers, carrying more weight on crypto, where the wider volatility regime makes normalisation more informative, than on stocks. The three indicators are combined using per-asset weights into a 0–100 strength score. Signals below a threshold are suppressed; signals above fire through the notification pipeline. Signal strength then maps to a buy multiplier between 1.0× and 5.0×, which scales the user's accumulated daily-buy allowance — a stronger signal recommends a larger buy.
Subscribers receive daily signals for 14 curated assets ("Mag14") spanning AI infrastructure, semiconductors, EV, and major cryptocurrencies, plus unlimited personal watchlist tickers scored using the same framework and delivered as private DMs. During onboarding, each subscriber selects a monthly budget tier; every signal computes the recommended buy amount at that tier, so the same quantitative opportunity is rendered as a different dollar amount for each subscriber. Signals arrive as branded PNG cards rendered with Pillow — sector badges, price, momentum reading, trend deviation, signal strength with strength pills, and a gold-trimmed callout with the personalised buy amount. Two-channel delivery (Telegram + web push) ensures signal delivery even when one channel is unavailable.
A scheduled daily job fetches OHLC data from Yahoo Finance with CoinGecko fallback, computes indicators for all Mag14 assets and every ticker on any user's watchlist, scores each asset, renders a personalised PNG per subscriber at their own budget tier, dispatches notifications, and logs every signal to an audit table with full reasoning — allowing dashboard-side performance attribution. User isolation is enforced at the database query layer; personal watchlist signals never surface on another subscriber's dashboard. Telegram bindings are established through 10-minute HMAC-signed deep links to prevent account hijacking.
Not a trading bot — the product recommends; the user executes. Slippage and order timing on the user's brokerage are outside the product's control. Not a hedge — DCA assumes long-term mean reversion and growth; in a persistent bear market, DCA losses compound just as DCA gains compound in a bull market. Not personalised investment advice — the product does not know a subscriber's full financial picture and is structured as market commentary, not as an investment adviser. Not infallible — every signal is logged and trackable; some will resolve favourably and some will not. The product is transparent about this.
Accumulator is the front door to the broader Soomario ecosystem. The signal logic it uses — momentum, trend extension, statistical normalisation, weighted scoring — is a subset of the analytical toolkit the vault products use. A subscriber who finds the product useful, follows the signals over a quarter, and develops a sense of how the platform thinks is a qualified prospect for the higher-ticket vault products. The $7/month price point is deliberately below the threshold where users evaluate financial products carefully; it functions as an introduction to the discipline and a continuous demonstration of how the platform applies it.
Risk management on the platform is the same framework reading differently for each strategy. The constraints below are universal — encoded in different forms across the six vault strategies and (where applicable) in the subscription product's signal-pipeline integrity controls.
Every leveraged position on the platform has a hard stop except Aphelion's, which deliberately runs without one on isolated margin (see Aphelion above). Each stop is sized to the asset's volatility and the strategy's risk budget for that position. Where the venue supports it, the stop is a reduce-only stop-market trigger placed directly on the exchange's matching engine — Hyperliquid and Binance for the perpetual strategies. Stops on the exchange fire whether the bot is alive or not. Where the strategy's design also calls for it (Gladius, Libration), an independent bot-side monitoring loop checks the same thresholds at a fixed interval and force-closes any position that has crossed without the exchange-side stop having fired. Two independent components must both fail before a position runs unmanaged.
Earlier versions of Aphelion pre-positioned a passive limit buy above the liquidation price. That layer has been removed: Monte Carlo testing showed it added size where losses were fattest. No strategy on the platform currently runs a pre-positioned defense layer.
Aphelion's account-level kill switch halts new entries if intraday losses exceed a threshold or if total unrealised P&L drops below a defined percentage. Engine-run accounts carry one catastrophe floor: if the account falls 60% below its starting balance, the engine closes its positions and stops until it has been reviewed. Gladius' reconciliation guard refuses to act when live equity (inclusive of all unrealised P&L) falls below peak by a defined amount, closing all grids at market and preventing new positions until operator reset. Each is sticky — once tripped, manual review is required to resume. Existing positions continue to be managed by their trail and stop logic; only new entries are blocked.
Position-size caps, leverage ceilings, no-stacking rules, per-name allocation caps in Rotation, and a sizing basis set per book rather than inherited — a single-asset book concentrates its whole allocation in one position, so the same percentage stop costs more of the account than it does inside a basket — all are encoded as constants the strategy code cannot override based on optimistic signal readings. A strategy that would be more profitable with looser constraints, but is engineered to refuse to loosen them, has a worst-case behaviour that is computable in advance. Gladius caps leverage per venue limits. Aphelion at 2x. Elite's leverage is set position by position by its manager, within Hyperliquid's limits. Rotation's short sleeve cannot grow into a meaningful share of the portfolio even if every short signal looked great that week.
The six vault strategies are decoupled by design. They use different signal categories, target different inefficiencies, and reach their drawdowns at different times. A coordinated risk-off across all crypto might pull Gladius, Elite, and Libration into drawdown simultaneously — but it is the regime in which Aphelion's accumulation levels fill and Rotation's unlevered sprts are typically most productive. A sustained risk-on rally is the exact regime in which Elite's directional momentum capture and Aphelion's long-bias accumulator perform best. The strategies are not chosen by AUM contribution; they are chosen by inefficiency targeted, and the resulting portfolio is what diversification across edges (rather than across assets) actually looks like.
The bug that destroys an algorithmic trading account is rarely a bad signal. It is the bug where the bot believes it has a position the exchange does not recognise, or believes it does not have one when it does. Every strategy on the platform reconciles internal state against ground-truth exchange data on every cycle — Aphelion every five minutes, Gladius against exchange ground truth every cycle, Rotation with a force-refreshed read before any state-changing close. State integrity is treated as a category of risk control with the same prominence as position-level stops. It is the substrate that makes any other control trustworthy.
| Execution venues | Hyperliquid (crypto perpetuals); Binance USD-M futures (crypto and US equity perpetuals, plus copy trading); Solana spot routed via Jupiter (tokenized equities) |
| Hosting | Railway — dedicated containers per strategy, persistent volumes, Singapore region for execution-sensitive workloads |
| Languages | Python 3.13 across all production strategies; JavaScript (Accumulator dashboard, signal-card rendering pipeline) |
| Frameworks | Flask for HTTP/dashboard layer; APScheduler for cron and interval cycles; gunicorn for production serving; Hyperliquid Python SDK |
| Persistence | SQLite on persistent volumes for state, positions, fills, and signal audit logs; JSON for ephemeral state files |
| Front-end | React via Babel CDN (no build step), Tailwind utility CSS, PWA-installable; marketing site on GitHub Pages |
| Auth & payments | Whop OAuth 2.0 with PKCE; Whop subscriptions; constant-time admin key comparison |
| Notifications | Telegram Bot API; Web Push VAPID; Discord channels for live trade broadcast |
| Image rendering | Pillow with bundled Poppins typography for Accumulator signal cards |
| Monitoring | Per-strategy live dashboards on soomariostrategies.com subdomains; on-chain vault verification for HL-vault strategies; structured JSONL logs |
The platform uses CEX-primary data feeds for price, candles, funding, and open interest — Binance, Bybit, and OKX bulk endpoints — because they are higher-throughput, free, and designed for steady-state monitoring. Hyperliquid is used for state queries, vault data, and execution only, never for steady-state monitoring (this is a deliberate choice against rate-limit pressure). Yahoo Finance and CoinGecko provide equity and crypto OHLC for accumulator-style strategies. DeFiLlama provides emissions, protocol metrics, and tokenomic context for universe construction. Coinglass-equivalent computations of liquidation density are performed natively by the platform from primary exchange OI feeds rather than scraped from aggregators. The Alternative.me Fear & Greed index provides daily regime context for the strategies that use it.
Every live vault product publishes its on-chain address. Vault performance, deposit history, position state, and realised P&L are visible on Hyperliquid's public vault page in real time. There is no LP statement, no quarterly attestation, no aggregated performance summary — the system is its own audit. The dashboards at soomariostrategies.com refresh on the order of seconds and show the same state any depositor could reconstruct from the public chain themselves. Every entry, exit, and risk-control trigger is logged and, for several strategies, broadcast to a public Telegram channel as it happens.
The vault contracts are Hyperliquid native — depositor capital is custodied at the protocol level. The operator does not custody depositor funds. Withdrawals are processed through the protocol's standard withdrawal flow on the depositor's own schedule. There are no lockups, no minimum hold periods, no deposit or withdrawal fees imposed by the operator. Hyperliquid's standard 0.02% taker fee applies to all trades and is paid to the protocol, not to the operator. Funding payments flow through the vault as normal perpetual mechanics.
The platform is operated by a single principal — MLO — based in Incheon, South Korea. Background includes a B.S. in Finance (Metropolitan State University of Denver, 2008), former FINRA Series 7/63 licenses, an M.Ed. in Curriculum & ELL (Concordia, 2019), and several years of self-directed quantitative trading research preceding the launch of Soomario Strategies in late 2025. Strategy logic is designed by the operator; implementation is AI-assisted, with all production code reviewed before deployment. There is no team. There is no fund administrator. The platform's operational simplicity is a deliberate constraint — it is part of why the worst-case behaviours described in this document are computable in advance.
The platform runs on Railway. A Railway outage suspends all execution while the outage persists; positions managed by exchange-side stops continue to be protected by those stops; positions reliant on bot-side monitoring would fall back to the exchange-side hard stops described in Section V. Multi-source price feeds include defined fallbacks (Yahoo → CoinGecko, Binance → Bybit → OKX), but a coordinated outage across primary and fallback sources would suspend signal generation. Telegram and web push are best-effort delivery; the Accumulator product runs two-channel redundancy and audits delivery, but neither channel offers SLA guarantees. Database backups for the subscription product are not yet automated as of this document; current mitigation is that database state is rebuildable from source signal logs in minutes.
The operator is a single person. Operational risk includes the operator being unavailable due to illness, travel, or other circumstance. The strategies are designed to run unattended for extended periods (exchange-side stops, automated reconciliation, kill switches), but materially novel events — exchange policy changes, protocol upgrades, regulatory developments — may require operator action that cannot occur during an unavailability window. This is an honest limit.
The vault products are custodied by Hyperliquid's native vault contract. The platform does not control this contract; smart-contract risk in the protocol's vault architecture is borne by depositors in the same way it would be by any direct Hyperliquid user. The operator does not custody funds, and the platform's operational design specifically avoids any architecture in which depositor funds could be mishandled by a bug in platform code — the strategies place orders against the exchange/protocol's API; they do not move depositor balances directly.
The Accumulator subscription is structured as market commentary and educational content, with explicit "Not financial advice. Trade at your own risk." disclaimers on every signal. This positioning aligns with established frameworks for financial newsletters and signal services in jurisdictions including the United States (where publishers are protected under the Lowe v. SEC precedent for impersonal, non-fiduciary market commentary), the United Kingdom, the European Union, and South Korea. The vault products operate as systematic, on-chain perpetual trading vaults — a category for which regulatory frameworks vary significantly by jurisdiction. Depositors are responsible for their own jurisdictional compliance regarding leveraged perpetual futures trading.
Past performance does not guarantee future results. Soomario Strategies' vault products trade leveraged perpetual futures, leveraged options, and equity-perpetual instruments, all of which carry substantial risk of loss including the possibility of losing the entire allocated capital. The Accumulator subscription product provides quantitative market commentary; it does not execute trades or hold subscriber funds. This document describes the platform as currently implemented and is provided for informational purposes only. Nothing in this document constitutes financial, legal, tax, or investment advice. Conduct your own research and consult appropriate professionals before allocating capital to any vault product or acting on any signal.
| Hyperliquid | On-chain perpetual exchange. Elite, Libration and the Hype Bot execute here. Native vault contract provides depositor self-custody. Standard 0.02% taker fee. |
| Binance USD-M | Centralised futures venue. Gladius and Aphelion execute here; Aphelion is distributed as a copy-trading lead account, so follower capital stays in the follower's own account. TradFi perpetuals provide the US equity universe. |
| Solana · Jupiter | On-chain spot. Rotation swaps stablecoin against tokenized US equities through Jupiter's aggregated routing, from a self-custodied wallet. No leverage, no liquidation. |
| Bybit · OKX | Used as data sources only — bulk price, candle, funding, and OI feeds for steady-state monitoring of the broader crypto universe. Not used for execution. |
Wilder, J. W. (1978). New Concepts in Technical Trading Systems. The original definition of the Relative Strength Index and the recursive moving average smoothing method.
Hyperliquid Documentation. hyperliquid.gitbook.io/hyperliquid-docs. Vault contract architecture, perpetual mechanics, HIP-3 specification.
DeFiLlama. defillama.com. Tokenomic and emissions data; protocol metrics used in universe construction.
Alternative.me Fear & Greed Index. alternative.me/crypto/fear-and-greed-index. Daily synthetic crypto market sentiment reading.
Lowe v. SEC, 472 U.S. 181 (1985). Foundational US precedent on First Amendment protection for non-fiduciary market commentary by financial publishers.
Platform · soomariostrategies.com
Twitter / X · @SoomarioStrat
Operator · MLO · Incheon, South Korea
Subscription · Soomario Accumulator — $7/month, 7-day free trial via Whop
This document represents the Soomario Strategies platform as of September 2026. Strategy implementations evolve as markets evolve, and this revision retires three strategies (Alpha, Zones, Farms) that earlier versions described as live. Where individual strategy whitepapers conflict with this document on operational detail, the individual whitepapers — kept current per strategy, including the separate AUREUS whitepaper covering the tokenized fund — are authoritative. The principles in this document do not change between revisions.
SOOMARIO STRATEGIES