Didn't find your answer? Email us — a human replies, usually within a day.
Does Tickbloom replace QuantConnect or Lean?+
No, and it isn't trying to. Those are execution and research platforms. Tickbloom sits upstream of them: it validates and documents the data and the code that feeds them. Most users run Tickbloom in the same notebook or CI job that later calls their existing backtester.
Which data providers are supported?+
Databento and local CSV/Parquet at launch, normalized to one schema. Polygon.io and IQFeed are next. The connector layer is modular: you can register a custom loader in about thirty lines by implementing the Loader protocol, and a community connector is a pull request — no fork required. The integrity suite works on any source that returns the normalized frame.
Is my strategy code or my order history sent to your servers?+
No. Checks execute client-side, in your own process. We record only audit metadata for the trail: Tickbloom version, check identifiers, thresholds, verdicts, and a hash of the input. Your code, your data, and your fills never leave your machine. The on-prem Fund tier removes even the metadata call.
How does the look-ahead scanner actually work?+
Two passes. A static pass walks your code's syntax tree looking for known leakage patterns — negative shifts, rolling windows centred on the current bar, scalers fit on the full series, resampling that pulls a future close into a past bar. A runtime pass proves it by truncation: it recomputes your features with every row after time t deleted, and any value that changes demonstrably depended on the future. A mismatch is proof of a leak. A clean result means no leak was observable at the cut points tested — not that the code is provably causal — which is why both passes ship rather than one replacing the other.
How is the Integrity Score calculated?+
It's a declared deduction model, not a learned one: score = 100 - sum(penalties), where each check has a documented maximum weight and a tolerance. Rate-based checks use exponential saturation, so the first defects cost the most and no single defect maxes out a check. Every weight and tolerance is recorded in the report manifest and can be overridden in tickbloom.toml — so you can say "we used these thresholds" rather than "the vendor decided".
What exactly is in the trade-documentation bundle?+
A timestamped order log, a fill breakdown with slippage attributed against a stated reference price, the equity curve with drawdown markers, the Integrity Score of the underlying data, and a manifest recording the versions and thresholds used. It exports to HTML, PDF, CSV, and JSON. It's designed to answer "why did this trade happen and can you prove the data behind it" — it is documentation, not a certification or a guarantee of approval by any particular firm.
Is the AI agent required?+
No. It's off by default and every tier works fully without it. When enabled it operates in dry-run: it proposes a diff and a regression result, and nothing is written until you approve. If you'd rather it never see your code, leave it disabled — the audit and documentation layers are entirely deterministic.
What happens if my score is low?+
You get a findings list ordered by how much each issue is likely to distort your results, with the specific rows, timestamps, or code lines involved. A low score is not a blocker — it's an itemized description of what you'd be trusting if you traded on this data anyway.