KEB Capital
A trading lab

Every edge starts as a test.

KEB Capital is a trading lab. We research, build, and stress-test trading algorithms, and nothing leaves the lab until the data says it's ready.

Experiment 0412Running
US30New York session breakout
In-sampleOut-of-sample
Real spreadsSlippage modeledCommissions inHard stops on

Experiment log

  • US30New York session breakoutStress test
  • NAS100Controlled growthForward test
  • SPX500Index confluenceResearch
  • XAUUSDTrend captureBacktest
  • EURUSDRange reversionRetired

Out-of-sample is data an idea never saw while it was being built. It's where most ideas fail, and why every one gets tested there.

Researching across

  • NAS100Nasdaq 100
  • US30Dow 30
  • SPX500S&P 500
  • XAUUSDGold
Why a lab

Every idea runs through the same lab.

Most trading ideas look great right up until they meet real market conditions. So every idea, whether it starts from market structure, a session pattern, or a hunch, goes through the same engine, with the same costs, rules, and standard of proof.

Market dataYears of history
IdeasPatterns worth testing
Risk rulesLimits come first
The LabTests all of it
BacktestsReal costs included
Forward testsLive market data
AlgosOnly what survives
The process

From idea to algo.

Most ideas never leave the lab. That's the point. Here's the path the few that do have to take.

1

Research

Start with a market behavior worth testing, and write down exactly what would prove it wrong.

Market structureSessionsRegimes
2

Stress test

Run it through years of history with real spreads, slippage, and commissions. Then try to break it.

Real costsWalk-forwardMonte Carlo
3

Forward test

Survivors trade live market data on paper and demo accounts. Only what holds up graduates.

PaperDemoProof pack
Inside the lab

Built like a research desk, not a signal group.

Every run is recorded, every algo carries hard limits, and an AI research team keeps the experiments moving day and night.

RunMarketTestResult
#0412US30Walk-forwardLogged
#0411NAS100Paper, day 9Logged
#0410XAUUSDStress testFailed
#0409SPX500Data checkPaused
Experiment log

Every experiment, on the record

Each run is logged with its settings, its data, and its result. Wins and losses alike, so nothing gets hand-picked.

Daily loss limitStops new trades for the day
Max drawdown guardFlattens before the limit
News blackoutSits out high-impact events
Kill switchOne switch, everything flat
Risk rules

Limits come before profits

Daily loss limits, drawdown guards, and kill switches are built into every algo from day one, not added later.

Lab agentsOvernight
  1. 02:14Queued 24 walk-forward runs on US30.
  2. 03:40Flagged a symbol mismatch on SPX500 and paused scoring.
  3. 05:12Logged day 9 of the NAS100 paper test.
  4. 06:00Morning report is ready for review.
AI research team

AI agents keep the lab running around the clock

AI agents queue experiments, read the results, and flag problems overnight. A person signs off before anything moves forward.

Our principles

Most trading algos are sold. Ours have to survive.

There's no shortage of bots with perfect backtests. The lab exists to find out what actually holds up.

How most algos get sold

A perfect backtestTuned until the past looks flawless, right up until it meets the live market.
Screenshots, not dataHand-picked wins with nothing behind them.
One good monthA hot streak, sold as an edge.
Set and forgetNo plan for the day the market changes.

Inside KEB Capital

Tested on data it has never seenWalk-forward and out-of-sample runs, not just the years it was built on.
Every result on the recordWins, losses, and the data behind both.
Forward tested firstLive market conditions before any real money is involved.
Retired when the edge fadesAn algo that stops working gets pulled, not promoted.

No guaranteed returns. No hype. Just research, proof, and algos that earn their way out of the lab.

Research in progress.

The next generation of KEB algos is being built and tested right now. When something earns its way out of the lab, it shows up here first.