TradeClaw

Open trading research for testing ideas, modeling costs, and reproducing every result.

Self-hosted by default

Evidence

  • Track record
  • Studies
  • Prospective ledger
  • Methodology
  • Open data

Lab

  • Candidate feed
  • Screener
  • Backtest

Build

  • Self-host guide
  • Documentation
  • API reference
  • GitHub

© 2026 TradeClaw. MIT licensed.

Terms|Privacy|Trading involves risk. Research candidates are not trade instructions or financial advice.
EvidenceLabBuild

Transparency

Data freshness and provenance

TradeClaw does not promise a universal “real-time” cadence. Upstream markets, providers, caches, configured jobs, and client polling all have separate clocks. Inspect the timestamp and source attached to the specific value you are using.

A page refresh time is not a market-data timestamp. A polling interval is not a delivery SLA. When provenance or required data is unavailable, production surfaces should show an empty or unavailable state instead of generated market values.
SurfaceTiming boundaryWhat to verify
Quotes and pricesProvider and client polling intervals vary by deployment.Each returned quote includes its runtime source. Missing provider data is omitted; hardcoded prices are not substituted.
OHLCV candles and indicatorsBounded by the upstream candle close and the operator schedule.Public candidates require provider-backed candles. OHLCV marked synthetic is excluded from signal generation and MTF scoring.
Signal candidatesGenerated by requests or configured jobs; no fixed publication SLA.Use the candidate timestamp, candle source, data-quality field, and persisted gate decision. A rule score is not a calibrated probability.
4h and 24h outcomesAvailable only after the relevant observation window and a successful verifier run.Outcomes are resolved against provider OHLCV. They are unsized price observations, not broker fills or portfolio returns.
Badges, feeds, and embedsHTTP and CDN caches can outlive the client polling interval.Consumers should display the source timestamp and an unavailable or stale state. Empty data must not become a neutral fixture.
Telegram, email, push, and webhooksBest effort after a configured job runs; no delivery-time guarantee.Entry-like delivery requires a persisted approval plus the current cost-adjusted evidence gate. Provider or gate failure blocks fan-out.
Backtests and paper simulationComputed on demand from the selected data and assumptions.Backtests are modeled. Paper fills require explicit observed prices and modeled slippage; neither is live broker performance.

Operational check

Before relying on a value, verify its provider, observation timestamp, data-quality label, gate scope, and whether the endpoint returned a stale or unavailable condition.

The repository documents intended schedules, but the deployed operator controls credentials and jobs. Production behavior must be established from the current response and deployment logs, not from a marketing cadence.

Review the methodology, proof status, and open data before drawing a performance conclusion.