01
US equity momentum
AlphaMax
Picks US stocks by momentum, how strongly each has risen against the rest, from a list that keeps companies that later failed.
- Execution
- Alpaca paper
- Target weight
- 25%
Observed broker$1M equity · 174 positions
Open evidence for quantitative finance
A good backtest
can be luck.
Canli Capital is an open research lab for quantitative finance that tests whether trading backtests are real or luck. Use its free validators in your browser, over an API or through an MCP client. Inspect SEC-traced company data and its own public paper-trading record, with failures and corrections kept visible.
Research, not a fund. Paper execution, not funded returns.
TradingPaper only
Broker3 Alpaca paper accounts
Paper recordSince 24 Sept 2026
Strategies3 running · goal 14
The vision
Finance that
can be checked.
A result should arrive with the work behind it.
Canli Capital is building open infrastructure for financial research: inspectable inputs, reproducible tests and a record that keeps its mistakes. For the people asking hard questions, and the agents working alongside them.
Validation, research, company data and paper execution belong to the same story. Each makes the next claim easier to examine.
How we approach researchThe problem / selection pressure
Look beyond
the winning chart.
The result changes when you count the search.
Try enough strategies and an impressive chart can appear by chance. A reported winner says little without the unsuccessful trials, the costs, and the data available at each decision.
Our method keeps that history visible. Test the mechanism, account for selection, then observe what happens forward.
Explore backtest overfittingEvidence Core / Public schema
The route from a frozen idea to a signed record.
Every shape has a job. Identities freeze before returns are opened. Failed trials remain in the union. Strategies stay separate. Broker observations enter from the edge. Signed records join the spine.
- 01Idea fieldCandidate mechanisms enter without a result.
- 02Frozen identityParameters, data and gates become addressable.
- 03Trial unionKilled work recedes, but never disappears.
- 04Broker railsEach strategy trades in its own account.
- 05Signed claimDisclosed payloads join the append-only spine.
Observed public stateIdea field
Trial identities228
Killed or screened46
Active strategies3
Signed entries1,379
Open research book
The failed trials stay visible.
The goal is not to look right in hindsight. It is to preserve what was tested, what failed, what changed, and which claims remain unearned.
114published documents
46rejected or screen-killed
Hypothesis identities, family decisions, and complete trial packets are separate accounting units. The trial ledger shows the denominator.
Read the research libraryPortable reproduction
Crypto Carry, without the private lake
A bounded package for reproducing the selected carry result from public inputs.
Read paper Data integrityCorporate-action basis reconstruction
Why split and dividend semantics can move an apparently stable equity result.
Read paper Research standardWhat would count as forward Sharpe evidence?
The observation, provenance, and maturity requirements fixed before the result arrives.
Read paperThe paper-trading record
The record.
Losses included.
Paper-trading results and positions, as the broker reported them, updated every hour. Switch between the whole portfolio and each strategy. Losing days stay in the chart.
Open the live dashboardOur strategies right now
3 strategies.
One open engine.
Each strategy gets an equal share. How much they move together is measured on past research data; whether that holds in live trading is not known yet.
Average correlation between strategies +0.0345 · on past research data, not the live record
02
Managed futures
AlphaTrend
Follows price trends in stock, bond, commodity and currency markets.
- Execution
- Alpaca paper
- Target weight
- 25%
Observed broker$1M equity · 15 positions
03
Macro surprise
AlphaVintage
Holds small US companies against large ones (IWM against SPY), positioned by how much inflation (CPI) surprised, using only data known at the time.
- Execution
- Alpaca paper
- Target weight
- 25%
Observed broker$1M equity · 2 positions
04
Crypto funding carry
AlphaForge
Collects the funding payments on crypto perpetual futures, priced from live exchange order books.
- Execution
- Local paper
- Target weight
- 25%
Observed localLive books, simulated fills
Dated evidence dashboard
What was observed, what was simulated, and what is still only a goal.
Performance targets never inherit the styling of earned results. Every figure below is bound by the same honesty pledge published in full.
Current forward recordObserved / paper
10daily return observations
Inspect maturity evidenceRecord freshness and execution basis
The operating facts behind the curve.
Latest signed record08 Oct 2026, 08:26 UTCseq 1,378 / 8b51d8cea022f5a7…Verify this record
14 paper-trading days, 10 return observations.
Open maturity record+0.36% to +31.20% net across dedicated broker accounts. Not a composite exposure.
Open broker evidenceThe public record does not yet expose a governed composite turnover series.
Read the measurement ruleEquity fills lack a decision-price reference. The measured crypto component predates the current forward window.
Open cost evidenceStrategyExecutionLatest observationState
AlphaMaxAlpaca paper174 positions / 08 Oct 2026, 02:40 UTCPass
AlphaTrendAlpaca paper15 positions / 08 Oct 2026, 02:40 UTCPass
AlphaVintageAlpaca paper2 positions / 08 Oct 2026, 02:40 UTCPass
AlphaForgeLocal paperLive exchange books, simulated fillsLocal only
Broker check passed · goals not met yet
How the goals are judgedBroker reconciliation passes. The Sharpe ratio goal (above 2.00) is not established yet; the paper record has 10 daily returns.
Evidence grammar
Three labels. No ambiguity.
Broker & paper record
Marks, positions, orders, and drawdown already recorded by the current paper system.
Research evidence
Walk-forward, stress, and capacity results under declared data and cost assumptions.
14-strategy goal
A governed breadth objective. It is not represented as a current portfolio achievement.
One decision / end to end
Follow a rebalance through the entire system.
Each layer answers a different question: what the model knew, what it chose, what the broker accepted, and what was published.
- 01 / InputWhat the model could know
Point-in-time data and survivorship contracts.
Data contracts - 02 / DecisionWhat was fixed first
Reserved identity, parameters, and admission rule.
Trial ledger - 03 / RiskWhat the portfolio allowed
Weight, exposure, drawdown, and kill state.
Risk evidence - 04 / BrokerWhat Alpaca accepted
Positions, orders, account mark, and reconciliation.
Broker evidence - 05 / RecordWhat cannot be silently replaced
Signed, append-only public payloads.
Verify chain
System films / Artifact rendered
Watch the system prove what state it is in.
Each silent loop is rendered from a sanitized public artifact, and carries its timestamp, basis, source and limitation in the frame.
01 / Research state
Open trial accounting
The engine is testing
Hypothesis identities freeze before executions enter the ledger. The preregistered budget remains visible.
- Artifact time
- Playback
- 12 seconds / silent
Read state transitions
- Candidate mechanisms enter without a result.
- The hypothesis identity freezes.
- Immutable executions accumulate against a visible budget.
02 / Execution state
Open broker evidence
The broker is observed
Each strategy keeps its own paper account. Aggregate reconciliation is shown without account identifiers, credentials, or holdings.
- Artifact time
- Capital basis
- Paper only
Read state transitions
- Three paper strategies run in separate accounts.
- Aggregate positions and orders are observed.
- The published reconciliation resolves pass or open.
03 / Publication state
Verify the public record
The record is sealed
Published states bind into an append-only sequence. The current head appears with its timestamp and short hash prefix.
- Artifact time
- Verification
- Signed chain
Read state transitions
- Published states join the sequence.
- Each node binds to the previous chain hash.
- The signed head resolves with a timestamp.
Observed state is not the same as validated performance. These films make the research, paper-broker, and signed-record boundaries easier to inspect. They do not imply funded execution or an earned Sharpe result.
Start here / free, no account
Check a backtest
before you trust it.
Ask Claude, Cursor or any AI assistant whether your backtest is real, or call the free API from your own code. You get three answers: whether the Sharpe ratio survives the number of versions you tried, how likely the backtest is to be overfit, and how long a live record must run to prove it.
Every answer says what it does not prove. API answers come with a signed receipt anyone can check.
Pick one way to start.
In Claude Code, on your own machine:
claude mcp add canli -- npx -y canli-validation-mcpIn any AI assistant, with nothing to install:
https://canlicapital.com/mcpFrom your own code, with a free key:
curl https://canlicapital.com/api/v1/status- Backtest a rule over a whole grid of settings, counted honestly
- Stress it against crashes and stuck positions
- Check it against a real broker's limits and costs
Answers describe your numbers. They never predict returns.
What you getOpen now / no paid plan
The work stays open.
Research, execution, and improvement belong in the same record. Inspect the decisions, reproduce the calculations, and see the corrections. Our free research tools are open to everyone.
The live paper record
Equity, positions, decisions and broker reconciliation, republished hourly. Paper throughout, never funded.
11 published equity marks · 3 broker accounts reconciled
Open the live recordThe research, and the retractions
Every hypothesis tested, the ones that failed, and the figures since withdrawn.
114 documents · 46 published kills · 228 trial identities (the retired legacy set)
Read the correctionsRun your own numbers
Free key, no account. Send your own returns through the same validation arithmetic the record is held to, and get back a receipt you can cite. Behind it sit three open repositories: the engine, the point-in-time data layer and the backtester.
1,220 tests passing · mypy --strict across 83 files
Validate your numbersResearch, not a fund
No managed capital, copy-trading, promised return, or investment advice.
0 external reviews · 0 independent replications so far
Read boundariesCompany financials
Numbers from SEC filings.
Each traced to its source.
Look up a US public company's reported financials, with every figure linked to the filing it came from. An AI assistant can also ask for a value as it stood on a past date, with later restatements flagged.
Authorship & accountability
A named builder.
An open record.
Built by Arhan Canli in Dubai. Research, code, corrections and operating evidence carry a name. This is a research project, not a fund or investment adviser.
Arhan Canli / Dubai
Quantitative researcher and builder of ALPHAC.
Building since July 2024 · public record from 24 Sept 2026
Founder recordProspective admission v7
New strategies are admitted only under the rules in force.
3 admitted of a 14-strategy goal
Read methodologyPaper record, real broker plumbing
Alpaca paper fills and local simulated crypto fills are never presented as funded execution.
3 dedicated accounts · 0 funded
Broker evidenceAppend, explain, preserve
A correction keeps the superseded claim and the evidence that exposed it.
1,379 signed entries, append-only
Correction recordThe people behind the work
Make the evidence
stronger.
Open work improves when more people can check it.
Reproduce a result, fix a tool, improve a source or make the explanation clearer. Contributions stay in the public history, with the work they changed.
Become a contributorMore room to research
A higher hosted API quota for contributors.
Help shape what comes next
Early access to new MCP tools for contributors.
The contributor reward program is in development. Quota amounts, eligibility and activation instructions will be published when the API integration is ready.
What comes next
The next claim
must earn its place.
More useful evidence. More independent scrutiny.
We are working toward stronger financial validation, independent reproduction, economically distinct qualified strategies and financial reasoning datasets reviewed by rights-cleared experts.
Those are continuing objectives. The published record, research status and corrections show what has been established and what remains open.
Follow the workHow it compares
Canli Capital, QuantConnect and QuantStats
They answer different questions. QuantConnect runs your strategy, QuantStats reports on its returns, and Canli Capital checks whether the result could be luck. Many people use more than one.
| Question | Canli Capital | QuantConnect | QuantStats |
|---|---|---|---|
| What it is | Free validators for backtest results, with SEC-traced company data and a public paper-trading record | A cloud platform and open-source engine (LEAN) that runs strategy code for backtests and live trading | An open-source Python library that turns a returns series into performance reports |
| Runs your strategy code | No. You bring the returns your backtest produced | Yes | No. You bring a returns series |
| Corrects the Sharpe ratio for how many variants you tried (deflated Sharpe ratio) | Yes | No. It reports a probabilistic Sharpe ratio: the chance the Sharpe ratio is above 1 | No. It offers a probabilistic Sharpe ratio |
| Probability that the backtest is overfit (CSCV) | Yes | No. Teams can instead require backtests to end months before today | No |
| Minimum track record length | Yes | No | No |
| Use it from an AI assistant (MCP) | Yes. A hosted URL with nothing to install, or an npm package | Yes. An official MCP server for its cloud API | No official server |
| Results anyone can recheck later | Yes. API results carry an Ed25519-signed receipt you can verify offline | No | No |
| Price | Free. MIT-licensed code, CC BY 4.0 data | Free plan and paid plans | Free. Apache 2.0 license |
Checked on against QuantConnect backtest results, QuantConnect pricing, LEAN statistics source, QuantConnect MCP server, QuantStats statistics source. Each project changes over time; its own pages are the authority.
What is Canli Capital?
Canli Capital is an open research lab for quantitative finance that tests whether trading backtests are real or luck. It publishes free validators, company financials traced to SEC filings, open datasets and its own paper-trading record, failures included. See what stays open.
How do I check whether my backtest is overfit?
Send the returns your backtest produced to the validators. The deflated Sharpe ratio corrects for how many variants you tried, the probability of backtest overfitting compares those variants with each other, and the minimum track record length says how much history the Sharpe ratio needs. Read the step-by-step guide.
How do I use it from Claude, Cursor or another AI assistant?
Add the hosted MCP server at https://canlicapital.com/mcp, which needs no install and no sign-up, or run npx -y canli-validation-mcp on your own machine. Your assistant can then call each validator by name. Follow the setup guide.
Is it free?
Yes. The validators, the API key and the MCP servers are free, with fair-use limits on the hosted API. The code is MIT-licensed, the published data is CC BY 4.0, and no paid plan is offered. Get a free API key.
Does my data leave my computer?
Only if you use the hosted API. Run the npm server with CANLI_LOCAL=1 and every calculation happens on your machine, with nothing about your returns sent anywhere. See the local mode setup.
How is this different from QuantConnect or QuantStats?
QuantConnect runs your strategy code and QuantStats reports on a returns series. Canli Capital checks whether the result survives the number of variants you tried and how likely the backtest is to be overfit, and signs each API result so anyone can recheck it. See the side-by-side comparison.
Where do the company financials come from?
From the companies' own filings with the US Securities and Exchange Commission. Each figure names the filing it came from, with its accession number, so you can check it at the source. Browse the company pages.
Are these strategies trading real capital?
The published strategy record is paper execution, not funded trading. Three strategies use Alpaca paper accounts; AlphaForge uses simulated fills against live exchange prices. Paper results do not establish executable future returns. Inspect the record and its basis.
What does “glassbox” mean here?
You can inspect the research, strategy rules, paper observations, failed work and corrections. The public record is self-published; transparency is not the same as an independent audit. See how to verify the evidence.
What can I do with an API key?
Submit your own research inputs to the validation tools and inspect their calculation receipts. The developer page includes key generation and Python, JavaScript and curl examples. A validation result is not an endorsement of a strategy. Start with the developer quickstart.
Can I invest or copy the trades?
Canli Capital is a research project, not a fund or investment adviser. This site does not offer managed capital or copy-trading. Read the project's boundaries.
Research access
Start with the evidence.
Everything here is free and needs no account. Leave an email only if you want release notes. No paid plan is offered.