Research for independent systematic traders
Describe a strategy in plain English. Evrant builds it, backtests it and then tries to break it. Every number comes from a deterministic engine you can rerun.
A good equity curve is easy to make. These are the four ways it usually misleads you.
The best of 2,000 trials looks brilliant because it was picked for that. Evrant counts every trial.
A signal reads a bar it could not have seen yet. Every value in Evrant carries an available_at time.
Tune, check, tune again, and the test set becomes training data. Evrant counts every look.
A library update or a new roll date breaks last month’s result. Evrant freezes the inputs.
An AI agent is good at reading, planning and writing code. It is never a source of numbers.
Development Sharpe is 1.38{run:9f3a2c/metrics.sharpe} across 1,284{run:9f3a2c/metrics.trades} trades. After 11:00 expectancy falls to 0.02 R{run:9f3a2c/by_hour.after_11}, which weakens the thesis for afternoon entries.
| Value | Artefact | Dataset |
|---|---|---|
| metrics.sharpe | run 9f3a2c | bars-1m v12 |
| metrics.trades | run 9f3a2c | bars-1m v12 |
| by_hour.after_11 | decomposition 4be1d0 | bars-1m v12 |
Every figure the agent quotes cites an artefact the engine produced. Anything it can't cite is flagged.
Follow one opening-range breakout on an example market from a plain-English idea to a candidate you can defend.
Write the strategy in plain English. Evrant keeps your words verbatim.
Rules, parameters and ranges written down as a version you can read and question.
C# compiled in a sandbox and repaired within a five-attempt limit.
An automatic backtest on development data. You check every trade against what you meant.
Ask why it works, when it fails and which variables matter.
Search the parameter space and keep the stable plateau, not the peak.
Lock one configuration and the pass criteria before the run.
Out-of-sample, walk-forward, Monte Carlo and stress tests.
Each run on validation data adds to the exposure count.
A frozen configuration with its complete evidence dossier.
Sealed data opens only by your explicit, recorded action.
Record the decision against the full evidence.
Move in any direction. The hard rules only protect the evidence.
See all nine chaptersEvrant tries to break a strategy before you rely on it, and every number traces back to its inputs.
Only you can open the sealed range, once, on a frozen configuration.
History is split by date with a one-week embargo. The agent can never request the sealed range.
3d2d bb4e 2ab9 4f4b a58e d7d33d2d bb4e 2ab9 4f4b a58e d7d3Every run is a pure function of frozen inputs. Rerun it next year and the hash matches.
Each result links back through the runs and findings to the thesis that motivated it.
The optimiser flags isolated peaks. The top trial is never promoted automatically.
Early access
Join free for an early-access invitation. No account or payment required. We’ll confirm market coverage before you choose a plan.
You’re on the list. We’ll email you when your invitation is ready.
For your early-access invitation only. Request removal anytime at [email protected]. See our privacy notice.