A systematic trading firm
Canli Capital
AlphaForge is the firm's first algorithm: a market-neutral, two-sleeve cross-asset book, US equity momentum and crypto funding carry, in simulation and live paper. No real capital is committed until it earns the right.
- Two near-uncorrelated sleeves, run market-neutral at equal risk.
- Tested on honest, leak-proof data going back to 2020.
- Running live, in paper. No real capital is at risk.
- Every result, and every failure, is on the record before a dollar is deployed.
01 / Thesis
The edge was never the strategy. It was the discipline around it. Point-in-time data, bars stamped to the moment they were knowable. A refusal to fool ourselves. We built that discipline first, and we will not commit real capital until the engine has earned it.
// Most platforms show you a chart and ask for your trust. We show you the method, then ask you to watch it prove itself in paper.
02 / The flagship
Algorithm 01
AlphaForge
A market-neutral, cross-asset book across US equity momentum and crypto funding carry. The first algorithm in the house.
AlphaForge scores its cross-asset universe, US equities and crypto perpetuals, across a library of 50 factors registered before any backtest, drawn from published anomalies and market microstructure, keeps only the signals that survive honest testing, and assembles a risk-managed portfolio that trades on the next open, never the bar it decided on. It is a research process that happens to trade, and today it trades only on paper.
-
01 /
Honest universe.
Built from historical listing and delisting records, including the instruments that died. The backtest cannot quietly drop its losers, because the data refuses to let it.
-
02 /
Many signals, one decision.
Every registered factor is measured for real predictive content, then blended. Two standalone sleeves do the work: US equity momentum, dollar-neutral 12-1, carries 54.4% of the risk; crypto funding carry, the funding paid between longs and shorts on Binance perpetuals, carries 45.6%. They are near-uncorrelated, and that decorrelation is the edge.
-
03 /
A portfolio, not a pile of trades.
Positions are sized against an estimated covariance and a target volatility, so the book holds a steady level of risk rather than a steady number of bets.
-
04 /
Costs that tell the truth.
Spreads, fees, and funding are charged in the backtest exactly as they would be charged live. One cost authority prices every fill, in research and in paper trading alike.
-
05 /
Next open, never this one.
Every decision is made at the bar's close and filled at the following open. There is no looking ahead, by construction, not by promise.
The book, by algorithm
Honest forward Sharpe 0.7 to 1.0 in-sample 1.46
The house runs three algorithms in parallel: two standalone sleeves, AlphaForge and AlphaMax, and their equal-risk combination, ALPHAC, which is the flagship the firm reports on. In-sample validation reads 1.46, but after correcting for every strategy we tried it deflates to a central forward of about 0.7, with an honest range of 0.7 to 1.0. That earns a grade of C plus, and a deflated Sharpe of 0.44 against our 0.95 gate, so deployment waits on the live record. That record began on 2026-06-21 and is days young, exactly as it should be.
// The two sleeves are near-uncorrelated, about minus 0.02, and that decorrelation is the edge ALPHAC is built on. The research curves are simulations. The live curves are days young and shown as they accrue. No return is claimed until it is earned in the open.
What this is
A research platform you watch prove itself.
Canli Capital is a quantitative research platform, not a signal you buy. You get a seat to watch AlphaForge be tested honestly, and the tools to test it yourself.
- You own no capital with us. We manage no money for you.
- You stay in control. We supply research, not copy-trading.
- Nothing here is an offer or a recommendation.
What early access includes today
- Watch live paper trading. The same scores, positions, and equity the engine acts on, updating as it runs. No real money in play.
- Research reports and the open build log. Factor attribution, drawdown and risk, full tearsheets, and what cleared the bar and what did not.
- Run your own simulations. Point the engine at a question and see how it holds up, on the same honest data and costs.
03 / Systems
The system, end to end.
Five stages. One unbroken chain of custody from raw market data to a filled order. Each stage is a discipline, and each one can be inspected.
-
01 / Data
Point-in-time, or it does not count.
Bars are stamped to the millisecond they could be known, and one reader enforces that on every query, so a leak is structurally impossible.
3.5M+ hourly bars / since 2020 / zero look-ahead
-
02 / Signals
Edges that survive their own scrutiny.
Each factor is measured for genuine predictive content before it is allowed to vote. The survivors are blended into a single score per instrument. The rest are shrunk toward nothing.
50-factor library / information-coefficient tested
-
03 / Portfolio
Sized by risk, not conviction.
Scores become positions through an estimated covariance and a volatility target. Correlated bets are recognized as one bet. The book is built to hold a chosen level of risk through calm and through stress, rather than to look impressive on a quiet day.
Volatility-targeted / covariance-aware sizing
-
04 / Risk
The brakes are part of the engine.
Exposure limits, drawdown awareness, and a regime view sit inside the decision, not bolted on after. The system is designed to do less when the market gives it less to work with, and to fail toward safety when something is wrong.
Regime-aware / drawdown-constrained
-
05 / Execution
Filled at the open, costed in full.
Orders are placed for the next bar and priced against a realistic model of spread, fees, and funding. The same cost authority that judges the backtest judges the paper fill. Research and paper trading are not two systems that resemble each other. They are one system, run twice.
Next-open fills / crash-safe, resumable
04 / The discipline
A system that will not lie to you.
The hard part of quantitative trading is not finding something that looks good in a backtest. It is building a process honest enough that a good backtest means something, and disciplined enough to wait for proof before risking a dollar. This is where most of the work went.
-
Walk-forward, always.
Every result is produced the way the system would actually have to live: trained on the past, tested on the future it had not seen, then rolled forward and repeated. There is one set of rules for research and one for reality, and they are the same set.
-
Statistics that have been made to confess.
When you test enough ideas, something will look brilliant by luck. We count the trials and discount the result accordingly, so a number that survives has survived the most likely way it could have been a coincidence. We would rather report a smaller honest figure than a larger flattering one.
-
Built to survive its worst day.
State is written down so a process killed mid-decision resumes exactly where it stopped, with no double-counted fill and no lost position. The system is engineered for the moment things go wrong, because that is the moment that decides whether an edge survives contact with a real market.
A claim is worth nothing until the evidence is on the record.
05 / Performance
What is true today.
No marketing numbers live on this page. The figures below are properties of the engine and its data, the kind of facts you can hold us to. Performance is not among them: AlphaForge runs in simulation and live paper trading, with no real money at risk, and we will not publish a return until one has been earned in the open.
- 0 Look-ahead, by construction
- 1 Cost authority, research and paper
- 2020 Crypto history starts
06 / Progress
One method. Many markets.
AlphaForge is the first algorithm, not the only one. The engine beneath it was built multi-asset from the first line of code: one data contract, one cost authority, one validation library, one execution loop. 12 phases are built and tested. Crypto perps alone did not clear our bar; the deployed book pairs that sleeve with a US equity momentum sleeve that does (net Sharpe 0.91), near-uncorrelated, as ballast. The road forward runs two ways: more lowly-correlated algorithms, and wider access to them, read the signals, follow the paper book, then run it yourself.
// Each new algorithm inherits the same standard of proof. Nothing skips the discipline, and nothing trades real money before it earns the right.
07 / Waitlist
Watch it prove the edge, in paper.
Early access is a seat to watch AlphaForge live on paper, nothing real at risk.
// Access opens as each stage is validated, not before. There is no waiting on a sales call, only on the engine.