Beta Docs

What Beta is

Beta is a hedge fund rebuilt as a token. A small fee on every trade of the token is collected, converted to stable margin, and traded by a systematic strategy on a perpetual-futures venue. When the desk realises a profit, 100% of it is distributed to the people holding the token, proportional to their holdings.

There is no fund administrator, no accreditation, no minimum, and no lockup. If you hold the token, you own a share of a working trading desk, and the desk pays you. Every step — the capital, the positions, the profit, and each payout — happens on-chain and is verifiable by anyone.

Why it matters

Quant trading is the most profitable corner of finance and the most gated. Access historically required being a hedge fund, or being wealthy enough to invest in one, and paying two-and-twenty for the privilege. The strategies themselves were never scarce — access to them was.

Beta removes the gate. It puts an institutional-style systematic desk on-chain, funds it with its own token instead of an LP raise, and returns the profit to holders automatically. The larger the token trades, the larger the desk becomes — and the payouts scale with it.

The self-funding loop

Beta never took outside capital. Its trading capital comes entirely from the token's own activity, cycled through a fully on-chain loop:

1. Fees accrue. Every trade of the token generates a fee in native ETH, held by the bonding curve / fee escrow.
2. Harvest. A keeper sweeps the fees the moment they're worth moving.
3. Swap. ETH is wrapped and swapped to USDG on a Uniswap-v3 0.01% pool, priced with a slippage floor.
4. Deposit. The USDG is deposited as margin into the fund's Lighter perpetuals account.
5. Size. The strategy sizes each market and adjusts the book to hit its leverage target.
6. Distribute. Realised profit is withdrawn and paid to holders (see Distributions).

The strategy

Beta runs one book on Lighter, assembled from several systematic sleeves that each do a different job. The trend book below is the core and carries most of the risk; two market-neutral overlays strip out market direction; and a deep-trend engine exists so the fund can keep growing past the capacity of its concentrated alt book. All of them net into a single set of target positions — there is no discretion anywhere in the chain.

The trend book · the core

The base is cross-market time-series momentum with a funding-carry tilt, run on the venue's highest-volatility perpetuals. Signals are evaluated on hourly closes — the 5-second poll drives execution and risk checks, not the signal — and every input is measured on the window ending at the previous bar, so there is no lookahead. Each market i gets a score in [−1, 1]:

Trend is an ensemble of three horizons (~1 week / 1 month / 3 months); a market earns full weight only when all three agree in sign, and taking the sign() of each return rather than its magnitude stops one outlier bar from dominating. Carry tilts against crowding — perps paying rich funding (high positive z-score) are tilted short, cheap or negative-funding ones long. α is a small, fixed, a-priori constant, deliberately not fit to the data (set α = 0 for plain trend).

Sizing

Each leg is scaled inverse to its own volatility, then the whole book is scaled to a portfolio volatility target:

Inverse-vol weighting (risk parity) means no single alt can dominate; the overlay then scales the entire portfolio so its realized annualized volatility sits near 30%. Gross exposure targets 3× and is hard-capped at 5×, per-market leverage is capped at 10×. The target book is recomputed against updated realized volatility once a day (every 24 hourly bars) to control turnover; between those, the desk re-sizes to current equity every 5 minutes, so the book tracks AUM as fees and profit flow in.

Funding overlay

The carry leg above is the funding overlay: the funding rate is turned into a rolling z-score, squashed through tanh to bound it, and blended in at α = 0.25. It trims longs paying rich funding and adds to cheap or negative-funding names, netting out carry drag before it erodes the trend return — you don't want to be long a move while overpaying everyone else to hold it.

Market-neutral overlays

Two cross-sectional momentum sleeves run on top of the trend book — one across the crypto perps, one across the equity perps (NVDA, TSLA, AAPL, MSFT, GOOGL, META, AMZN, COIN, PLTR…). Each ranks its basket by momentum, longs the strongest names and shorts the weakest, and is balanced to zero net dollar exposure, so it earns from dispersion between names rather than from market direction. Being dollar-neutral, they add return without adding directional risk, and they lower the desk's net beta. Every name is liquidity-capped: no position may exceed a fixed fraction of that market's average daily volume on Lighter, so the book can never outgrow the venue it trades on.

The deep-trend engine · capacity

The concentrated alt book has a ceiling — those markets are only so deep. The deep engine runs the same trend approach across the deepest books on the venue (SPY, QQQ, BTC, ETH, gold and silver proxies, and the mega-cap equity perps) on daily closes, so the fund keeps compounding past that ceiling.

It is dormant below $150k of AUM and ramps to full size by $1M, and — this is the important part — it is allocated only the leverage headroom left over after the trend book and the neutral overlays have been sized. It can never dilute a sleeve that is already working; it only spends capacity nothing else is using.

Signals under research

Not every idea the desk studies earns capital. This section is what is being researched but is not trading — published because a fund that only shows you its winners is telling you half the story.

Open interest — the total leverage positioned in a market — is the clearest example. Price is one dimension; it can't tell you why a market is moving. The same green candle can be fresh capital committing to a move, or shorts being squeezed out of a dying one. Two signals read the difference:

• OI momentum: rising open interest alongside price confirms new capital entering — continuation.
• OI / price divergence: price rising while open interest falls is short-covering, not conviction — a weak move to fade.

Why it isn't on capital. Neither venue publishes an open-interest history — Lighter and Hyperliquid both expose a live snapshot only — so the signal cannot be tested out-of-sample on the data available. An early in-sample read over roughly three weeks looked strong, and three weeks is far too short to distinguish an edge from luck. So the desk logs it forward instead: an hourly recorder has been writing open interest, basis, funding and book depth for every Lighter market since September 2026, building the history an honest test needs. Measured on the hours the fund has actually been live, the rule is profitable but statistically indistinguishable from noise — its sign flips depending on the lookback you pick, and a few days of data cannot tell an edge from luck. So it stays in research, unfunded. It goes on capital if and when it passes an out-of-sample test — and not before.

Distributions

Every epoch — once a day, at the 4pm ET close — the keeper rebuilds each holder's balance from the token's Transfer history, excludes the addresses that hold supply but are never paid (bonding curve, LP, dead/zero), and pays out 100% of the epoch's new profit pro-rata to the eligible float. “New profit” is the desk's marked profit above its all-time high; see Settlement for exactly how that is measured.

Accounting runs on a persistent, churn-safe ledger. Each epoch allocates only the new profit — pool − pool_prev — across the current holders by balance, and adds it to a per-wallet running total. That cumulative is monotonic and never decreases, so selling can't let the same profit be re-credited to whoever buys next — the failure mode that silently over-distributes a naive per-epoch snapshot. The resulting (address, cumulativeAmount) set is committed as a cumulative Merkle root:

Because the leaf stores lifetime-owed, a claim always transfers cumulativeAmount − alreadyClaimed. A dividend too small to clear the gas floor in one epoch is therefore never lost — it stays in the root and pays out once it's worth the gas.

You can view the full holder set and each holder's live share on the distributions page.

Settlement — how profit is freed without closing the book

Paying holders can't be allowed to disrupt the strategy, so the desk never flattens to distribute. At each settle it marks profit, takes only the surplus above its all-time high, and frees exactly that much margin by trimming winners:

1. Mark total profit as realised + net unrealised PnL across the whole book.
2. Distribute only the amount above the high-water mark. A recovery back toward a prior peak pays nothing (never paid twice); a drawdown pays nothing (never against a loss); and nothing already paid is ever clawed back.
3. Free exactly distributable USDG by trimming winners proportionally — the rest of the book is untouched, so the strategy's stance is preserved.
4. Fast-withdraw the freed USDG from the venue and fund the epoch's Merkle root.

Only surplus margin leaves; the desk keeps trading through every distribution.

The treasury

The treasury is a contract that holds $BETA and can never sell it. It is a holder like any other: it owns tokens, so it is paid its pro-rata share of every distribution, exactly as a wallet is — the dividend keeper refuses to exclude it, precisely so this can't silently break.

What it does with that income is the whole point. Each cycle it spends its USDG buying $BETA on the open market — USDG to WETH on the 0.01% Uniswap-v3 pool, unwrapped to ETH, then ETH to $BETA on the token's own pool — and keeps what it buys. A larger stack earns a larger share of the next distribution, which buys more still.

One way only. No function in the contract releases those tokens — not for the owner, not for the keeper, and it is not a timelock. renounceOwnership is disabled and the rescue path reverts on $BETA. Supply that reaches the treasury leaves the market permanently. Buybacks are rails-limited per cycle because the pool is thin, and each one is rehearsed with an eth_call first so the minimum output is quoted net of the pool's own 3% fee.

Verify it on-chain

None of this requires trusting us. Every claim on this page is a transaction or a contract you can read yourself.

• The desk: the fund's Lighter account is public — live collateral and every open position, straight from the venue.
• The money loop: the fund wallet on the Robinhood-Chain explorer shows the fee harvests, ETH→USDG swaps, and deposits into the Lighter contract.
• The distributions: the Dividends contract holds every posted root and every claim — total paid, per-wallet, all of it.
• The treasury: its $BETA balance is a public token balance, and every buyback is a transaction on the same explorer.

See Addresses for every link.

Risk & controls

• Leverage is bounded. Gross exposure targets ~3x and is hard-capped at 5x. Positions are volatility-sized so risk stays balanced across names.
• Risk is bounded by construction, not by a halt. The desk's protection is its sizing — bounded gross leverage, volatility-targeting and per-name liquidity caps — rather than an automatic stop. A daily-drawdown kill-switch exists in the code and is currently switched off: the desk runs cross-margin and relies on the venue's own liquidation logic instead of halting itself on a drawdown. If that changes, this page changes with it.
• Non-custodial. The fund never holds holder capital. You hold your own token and can exit at any time; the chain is the referee.
• Leverage cuts both ways. This is a real trading strategy on volatile assets. It can and will have losing periods. Distributions reflect realised profit, so they vary with performance. Nothing here is a guaranteed yield, and nothing here is investment advice.

Addresses

Robinhood Chain (chain id 4663) · RPC rpc.mainnet.chain.robinhood.com · explorer robinhoodchain.blockscout.com