About Alpha Factory

Alpha Factory discovers and tests investment strategies across shares, options and crypto. Each independent strategy is a sleeve.

Research draws on prices and volumes, earnings calls and reports, company fundamentals and estimates, news, options data, and recorded observations.

Research agents propose new sleeves each week.

Execution-ready ideas enter Sandbox with a $10,000 paper research budget. Weekly reviews may advance sleeves to Live trading on a small probation size (through Staging while their data is cleared for live use), hold them for more evidence, or mark them Rejected.

┌─ Strategy promotion flow

Ideas88 sleevesResearch & readiness
Sandbox303 sleeves$10,000 paper research
Staging2 sleevesFurther paper validation
Live4 sleevesReal cash · evidence gates

Weekly review → promote, hold, or Rejected (171 sleeves)

Promotion depends on readiness, data and evidence. Live trading uses real cash. Paper results are not live results, and no strategy is guaranteed to succeed.

How evidence builds up

Each week of real results narrows the range a strategy's true average could plausibly be. A strategy is only trusted once that whole range sits above zero, so a lucky streak is not mistaken for skill.

How a strategy goes from idea to real money

1 Ideas 88 sleeves

Every week we look for new ways the markets might be beaten. Each idea is written down as a fixed set of rules before it is tested, so nobody can quietly change the rules later to make the results look better.

Ideas that cannot be tested properly yet wait here.

Weekly research. Two research programmes run every Sunday: established methods (factor, calendar, pairs, macro and event studies) and a frontier "Skunk Works" programme. Frontier ideas face a critic and a screen against frozen, hashed public history before anything is built.
Frozen contract. Each sleeve is registered with an append-only contract: its economic mechanism, its baseline, what result would falsify it, its data sources, cost and risk limits, and its research provenance. Contracts are versioned, never edited; failed and retired versions are kept and counted.
Research observers. An idea without an execution route stays in Ideas. It may record forecasts and exclusions, but never simulated fills, and its records are not promotion evidence.
Data. Ideas and Sandbox may use accessible public data and authorized feeds while unresolved terms are recorded. Explicit restrictions, access controls and point-in-time correctness always apply.

2 Sandbox 303 sleeves

Once an idea's rules, information and trading connections all work, it starts trading with practice money at a real broker. Every idea gets the same practice budget.

We check every day that each idea is getting fresh information and actually trading. Ideas that lose too much, go quiet for months, or reach the end of their planned test are retired automatically.

Admission. A sleeve enters Sandbox when its frozen rule, point-in-time inputs and paper execution route are verified. No profitable backtest is required: Sandbox is for finding a signal.
Budget. $10,000 starting paper NAV and a $9,000 deployment cap per sleeve, sized as cap ÷ maximum positions. New funding pauses while the paper account's estimated buying-power use is at or above 75%; the research target is 50% use, met by adding always-invested sleeves rather than larger bets.
Execution. Each Sandbox sleeve trades only in the Sandbox paper account. Decisions and order intents are written before orders are sent; positions, partial fills and exits come from the broker's own records. No fill is ever simulated, and no decision is copied to another account.
Daily health. A sleeve shows green only when every input is connected and current and it is trading: no entry and nothing held for 30 days, or capital use below target, turns it amber. Paper fills carry no real costs, so results are also shown net of modelled fees, spread and slippage.
Daily retirement. A Sandbox sleeve is retired when its results are deteriorating, it breaches its declared loss limit, it has been funded 180 days with no outcome in 120, or its own protocol has ended. If more than half of Sandbox goes quiet at once, that is treated as an outage and nothing is retired for dormancy. Capital is released once the sleeve is flat.

3 Staging 2 sleeves

Some promising ideas rely on information we are not yet allowed to use with real money. They wait here, still trading with practice money, while we check the rights to that information and what it costs.

As soon as that check is done, they can move to real money. Ideas that start doing badly go back to Sandbox.

Optional. Only a sleeve that passes the fast-track screen (below) but is not yet cleared for live use comes here. It competes for one of 3 places, strongest evidence first, and a sleeve whose results duplicate one already selected (same family, or weekly correlation of 0.8 or more) is held back.
Diligence. Before Staging, every required source needs documented usage rights, retention, derived-data rights, price and review evidence. Live keeps a stricter governance gate: live-use licensing, pricing and ingestion approval.
No minimum stay. Once its data is cleared for live use and it still passes the screen, the weekly review moves it to Live.
Way back. A Staging sleeve whose results deteriorate or breach a loss limit returns to Sandbox once it is flat; it is never liquidated by the review.

4 Live 4 sleeves

An idea that has made money in practice for about two months, more than luck alone would usually explain, starts trading real money.

Two months cannot prove an idea works, and many that pass this test will turn out to be luck. So each new idea starts with a tenth of a normal amount, and new ideas together can never risk more than a fifth of what the fund is prepared to lose.

Every night we look at its real-money results. If they keep holding up, the amount grows in steps to a full share over several months. If it starts badly, it goes back to practice straight away.

Several automatic safety switches can stop any trading at any time.

The screen. Each Sunday, a Sandbox or Staging sleeve qualifies with at least 56 days of evidence (about 40 trading days), 10 completed trades and 5 independent weekly blocks, and a one-sided t-statistic of at least 1.0 on after-cost, benchmark-relative results. There is no multiplicity penalty. Costs include recorded fees plus a stress of at least 20 basis points per round trip.
Also required. A stressed net return above zero; a positive result without the single best trade; drawdown and worst trade inside declared limits; recent results not deteriorating; and use of at least 50% of its capital. It also needs a Live route, live data governance for every required source, and no open inventory in its paper account.
What the screen means. In simulation, a strategy with no edge passes about 16% of the time, and one with a yearly Sharpe ratio of 2 about 42%. It tilts the odds; it is not proof. Size and fast exits carry the risk control.
Probation size. A new Live sleeve gets 10% of an equal share of the fund-wide severe-loss budget. All probation sleeves together hold at most 20% of that budget, scaled down pro rata when over, and never borrow unused shares. The other Live sleeves split the rest equally.
Stepping up. Each night, on the sleeve's own completed Live trades only: to 25% after 60 weekdays with t ≥ 1.5, to 50% after 120 weekdays with t ≥ 2.0, and to a full share once the anytime-valid lower bound is above zero with at least 12 trades (error rate 5% split across every probation entry).
Exit. Back to Sandbox after 20 weekdays and 3 trades with t ≤ -1.0, or at once when realized Live losses reach half its probation allowance. New Live entries stop immediately; open positions keep their own exits.
Sizing. Stocks are charged a gap shock and defined-risk options their maximum loss against the sleeve's share. A budget that cannot be stated sizes to zero.
Safety gates. Per-position equity, liquidity, sector and buying-power limits, a kill switch and a hard order switch apply to every order. A separate overnight profit-factor check can also demote any Live sleeve on its verified Live fills.

Built by web2 boomer