Early access · research tooling for futures strategies
VeriRun Lab is a research IDE for futures strategies. Describe an idea in plain English, review the Strategy Brief and the generated code, then run research-grade backtests with conservative fills and an A–F scorecard that isn’t afraid to say NO-GO.
Research tooling, not trading advice.
How it works
Write the strategy the way you’d tell a colleague — “fade the opening range breakout on ES, RTH only.” The builder asks clarifying questions instead of guessing.
Every assumption is written down in a Strategy Brief you approve first. Generated code arrives as a reviewable diff — production code, tests and test setup accepted separately.
Explore fast with an idealized preview, then grade with conservative fills: trade-through required, post-only makers, per-contract fees.
Every run lands in your library with its exact code version, parameters and data slice — graded, comparable and fully traceable.
The platform
Start from a plain-English idea. The builder asks clarifying questions, writes a Strategy Brief with every assumption made explicit, and only then generates code — a multi-file proposal you review as a diff.
Clarifying questions before code — nothing applies without your accept.
The code is yours to read and change — Python completions, signatures and live diagnostics, with versioned snapshots you can roll back to at any time.
Every run pins the exact code it ran.
Most backtests flatter you. VeriRun Lab runs an idealized preview to explore quickly, then a conservative pass — trade-through required to fill, post-only maker orders, per-contract fees — before anything gets a grade.
A NO-GO verdict caps the grade — the scorecard can’t be talked out of it.
Runs aren’t throwaway. Each one is saved with its grade, settings and results — searchable, taggable and comparable side by side, so last month’s experiment is still evidence today.
Share a read-only results link when you want a second opinion.
A result you can’t reproduce is a rumor. Every run records the exact inputs that produced it, so any number on screen traces back to its source.
Recorded automatically on every run.
The preview chart draws entries and exits on real session structure, so you can eyeball behaviour before spending a conservative run on it.
Idealized preview and conservative grade stay clearly separated.
Optimization studies, one-approval variation fan-out and a multi-instrument screener widen the search; external agents drive the same honest loop over MCP, and the Claude Code team plugin packages the whole workflow.
Same graded scorecards, whichever door you come in through.
Equity charts overlay buy & hold — on new backtests at your run's own capital with engine costs applied — or a flat-cash floor, so a winning curve has to beat the boring alternatives in plain sight. Older runs show the raw frictionless series and say so.
Excess net and correlation are display stats — they never touch the verdict.
Link a journal tag to a strategy’s reference backtest and the platform keeps score as live trades come in: In line, Drifting, or Diverged — dated verdicts that append, never a backtest quietly rewritten. The full trading journal sits underneath — imports, MAE/MFE excursions and a real-candle chart for every trade.
Outperforming live never counts against you — only real divergence does.
Publishing a run takes audit coverage — a deep audit, robustness checks, a stated hypothesis — and the board ranks by that evidence. Dollar P&L never leaves your account. Alongside it sits a global feedback board where every signed-in user can file bugs and feature requests, vote, and follow status — with a moderation gate keeping it usable.
Ranked by audit evidence, not by return. A modest, robust result outranks a spectacular, fragile one.
Built for research integrity
AI proposes; you decide. Generated code lands as a scoped, reviewable diff — gathered in a drafts inbox for review — and nothing is applied to your workspace without an explicit accept.
Your strategies execute in an isolated sandbox with no network access — runs are reproducible and your machine and data stay out of reach.
AI usage runs against your plan's daily quotas — enforced, with honest refusals when a limit is reached. Market-data purchases are quote-first and human-approved — you type the cost to confirm it. No surprise bills, and a clear view of what each session consumed.
The grading rubric is versioned and shown with every verdict. Reports carry their method and their caveats — the scorecard can’t be talked out of a NO-GO.
VeriRun Lab is in early access. Open the app and take a strategy from sentence to scorecard.