Quant research that
shows its work.
Most trading ideas that look good in a backtest are luck. We build AI-assisted software that tests ideas against data it has never seen, and records every result, including the failures.
Many random paths look good on past data. Once hidden data starts, nearly all stop working. Random numbers, not a real result.
A backtest can prove almost anything.
Try enough ideas and some will look brilliant by chance. Strategy generators rarely tell you how many ideas were tried, or what happened when the winner met fresh data. We built the opposite.
Write the rule down first
Before a test runs, its rule and pass mark are committed to version control. Nothing is changed after the result is known.
Adjust for luck
Every test is counted. A strategy must beat what pure luck would produce across all the tests we have run, not just its own.
One look, no retries
A slice of history is locked away. A strategy that survives the gates gets a single, logged look at it. A failure stays a failure.
Find it. Prove it. Run it. Watch it.
KynovaX Research finds and tests strategies. KynovaX Portfolio shows how a set of strategies behaves once it is running, measured against the backtest it came from.
Know whether an edge is real.
One funnel takes an idea to a demo account and removes everything that does not hold up on the way.
A pass on one market proves nothing about another.
Each idea is run on 30+ markets and a grid of settings. Every market picks its own best setting on the first 75% of its history, then has to prove it on the last 25%, which was never used to choose.
- Beats random entries on the final quarter
- Positive after trading costs
- Failure reasons in plain words, per market
Illustration. Each dot is one setting on one market. Many shine in development and fade on the final quarter.
Fixed pass marks, set before the results.
Every strategy faces the same five pass marks, written down before any result. It has to reach the mark on all five, and stricter checks are shown beside them for information.
- Average and luck-adjusted Sharpe
- Stable across nearby settings and periods
- Beats random timing
Illustration. The dashed line is the pass mark on each gate; a strategy has to reach it on all five.
Know when your edge starts fading.
Live monitoring for portfolios of strategies built in StrategyQuant X and run on MetaTrader 5. Every strategy is checked against its own backtest.
- Connect and uploadConnect your MetaTrader 5 terminals and upload each strategy's backtest into the strategy library.
- Match live to backtestLive trades are matched to each strategy by magic number and compared with what its own backtest did over the same number of trades.
- Read the verdictsHealth checks, a portfolio verdict and a drawdown budget say whether live trading looks like the backtest, with alerts to Telegram.
Nine checks, each fine, look or act.
Terminals online, setup checks, result and drawdown against the backtest, trade pace, entry quality, open risk and execution cost, on one screen.
- 4 fineTerminals · Trade pace · Open risk · Execution cost
- 2 lookSetup checks · Entry quality
- 3 actResult · Drawdown · Adverse-month profit factor
One portfolio on one day, redrawn from the dashboard. It shows red when live trading is off its backtest.
See live inside the range it should stay in.
The live running total in R is drawn over the backtest's 50% and 90% bands and turned into a verdict from 0 to 100, where 50 means as expected.
A budget set from the backtest, with alerts.
The budget comes from the backtest's worst case. As the live drawdown uses it up, Telegram alerts fire at 50, 75 and 100%.
Every strategy against its own backtest.
Average trade, win rate and profit factor, live versus backtest. A short live record is judged against what the backtest does over the same number of trades, so normal bad luck is not flagged red.
Two strategies from one portfolio, redrawn from the dashboard. Names hidden.
Snapshots are from our own dashboards and accounts, early October 2026. Account numbers, balances, money amounts and strategy names are hidden. They include losing periods; the products are built to show them.
Updates where you already are.
Each product has its own private Telegram bot, so nothing needs a dashboard open.
Daily results and alerts.
Reports on your live portfolios: a daily summary every morning, and an alert the moment a strategy or a broker connection needs a look.
- Previous broker day, per portfolio
- Strategy falls to the 5th percentile of its own backtest
- Broker connection dropped and restored
Every pass, as it happens.
Reports on research and the demo account: strategies that clear every research test, and new strategies added to the demo with their research Sharpe.
- New survivors with Sharpe and alpha t-stat
- New demo strategies, sized by risk budget
- Verdicts on new ideas
One funnel from idea to demo account.
Each stage removes ideas that do not hold up. Live trading is always a human decision, and nothing here trades client money.
From papers, books, our own hypotheses and improvements of earlier results.
Each idea is run on 30+ markets and up to 125 settings, against random entries and shuffled data.
Five gates, then a check for real skill versus riding the market.
The locked slice of history, opened once per strategy.
A broker demo account, risking at most 0.1% per trade with a stop on every trade.
Needs 60 days and 30 trades that match research, then a human decision.
Where ideas die
- 6,960strategy tests, one dot each
- 44passed all five gates
- 4got one look at hidden data
- 1passed hidden data
To scale: 6,960 dots in all. The few that light up are the point. As of 8 October 2026.
What a strategy has to prove.
Every strategy and every market goes through the same five fixed tests. The pass marks were set before the results were known and are never loosened for a single idea.
Average Sharpe at least 0.30
Sharpe measures return per unit of risk. It is averaged over all nearby settings, so one lucky setting cannot carry the idea.
Luck-adjusted Sharpe at least 0.50
Adjusted for how many tests the whole programme has run. Try enough ideas and something looks good by accident; this gate removes that.
At least 70% of settings profitable
A real edge is a plateau, not a spike. Neighbouring settings must make money too.
At least 60% of 2-year blocks positive
It has to work across periods, not only in one good stretch of history.
Beats random timing (p at most 0.05)
Random entry dates with the same exposure must rarely do as well. If they do, the timing is not the reason it works.
Three more checks are also required: a bootstrap test that resamples the trades, positive results in three equal periods, and still profitable with trading costs doubled. A passing strategy is then screened for real skill against its own market, and only then gets one look at the hidden data.
A rule that looked great, then did not last.
A published paper described a bitcoin rule: buy at the US close when bitcoin is at a 10-day high, sell the next morning. We wrote the test down and ran it with real trading costs.
In research, the rule beat holding bitcoin and beat random timing (p = 0.004). On hidden data it earned nothing over 89 trades and its timing was no better than random (p = 0.54).
We did not adjust the rule and test again. A version with a stop-loss would only re-test a failed idea on data it had already seen. The result went into the log as a failure.
The demo account has hard limits.
Strategies that pass trade automatically on a broker demo account, so we can watch how they behave going forward. The limits are in code, not in a policy document.
At most 0.1% at risk
Position size is set so that hitting the stop loses no more than 0.1% of the account. A trade that cannot meet this is not taken.
A stop at the broker
Each strategy has the stop it was tested with, placed at the broker, never moved closer.
3% in total, 1% per market
All open stops together stay within 3% of the account and within 1% for any single market.
Short-term trades go flat
Strategies on 30-minute and 1-hour bars close before the weekend. Slower strategies may hold.
No more than 30 strategies
The demo is an incubator with a cap, and each idea gets one place per market.
A human decision
A strategy needs 60 forward days and 30 trades that match its research before it is even considered. Nobody and nothing else decides.
What it takes to get to the demo account.
The one strategy so far that cleared every stage: a bitcoin trend breakout on 4-hour bars. Each step below was a place it could have failed.
- Idea
A well-known breakout rule with a trend filter
A standard price-breakout pattern, written down with its parameters before any test. We publish results, not rules.
- Research
Studied on every market on its own
Run on 30+ markets with a grid of settings, against random entries. Each market is judged on its own, so a pass on one market proves nothing about another.
- Gates
Five gates and three extra checks
Average Sharpe 1.59 across nearby settings, luck-adjusted, stable across periods, and better than random timing.
- Skill check
More than just owning bitcoin
Screened against buy-and-hold on bitcoin itself, with trading costs doubled.
- Hidden data
One look at data it had never seen
Nov 2023 to Jun 2026: Sharpe 0.97 after costs over 79 trades, timing better than random (p = 0.055). It passed. The same step failed the bitcoin overnight rule above.
- Demo
Paper-trading, with limits
Now trades on a demo account at 0.1% risk per trade. The forward record is far too short to judge, and live trading is a separate human decision.
Research and hidden-data results, not live results and not investment advice.
Most ideas fail. We publish that.
A few recent results, in plain words. The full research log is here.
Bitcoin trend breakout
Survived five gates, the skill check and one look at hidden data (2023 to 2026). Now paper-trading. The forward record is too short to judge.
Bitcoin overnight rule
Strong in research. On hidden data it earned nothing, and did no better than random timing. We did not tweak and retry.
Momentum filters on breakouts
None of nine popular filters beat randomly removing the same share of trades. The closest was a quiet-market filter.
Stop guessing. Start knowing.
KynovaX Research and KynovaX Portfolio are opening to a small group first. Join the waitlist and we will write when your place is ready.
Your email app opens with the request filled in; send it to join. We use your address only to contact you about early access.
Straight answers.
Do you manage money or sell trading signals?
No. Kynova Ventures does not manage client money, run a fund, sell signals or give investment advice. We build research software and publish what it finds.
Is this "AI trading"?
No. The strategies are fixed rules, tested by deterministic code so every result can be reproduced. AI tools support the research process, but they never make trading decisions and they cannot place trades.
Why publish failures?
Because the failures are the evidence. A track record of only winners hides how many ideas were tried. Counting every attempt is what makes a pass mean something.
What is "hidden data"?
A slice of history that is locked away during research. A strategy that survives the gates is allowed one look at it, logged in advance. If it fails, it stays failed.
What is a Sharpe ratio?
Return per unit of risk. Around 0 means no edge; above 1 is good. Because it is adjusted for costs and for how many ideas we tried, a high number here is harder to earn than it looks.
Which markets do you test?
More than 30 daily markets: stock indices, gold and silver, oil and natural gas, major currency pairs, bitcoin and ether. Intraday tests use the markets with intraday history.
Can I get your strategies?
We publish results, not rules, and we do not offer strategies for sale or copying. Anyone with research to share, or data to check our work against, is welcome to write.
A research company, not a signal service.
Kynova Ventures was founded by a former banker who left banking to build in AI and quantitative research. Trading is something she has always been passionate about, and the company exists to put modern AI to work on trading research: testing more ideas, more carefully, than one person could by hand.
We treat a trading idea the way a scientist treats a hypothesis.
Write the test down first
The rule and the pass mark come before the result.
Count every attempt
A pass has to beat what luck produces across all tests.
Keep the failures
Losers stay in the log, and a locked slice of data gets one look.
Talk to us.
Researchers, data providers and anyone who wants to check our work are welcome to write.