Most trading ideas fail. We publish which ones did.
A strategy earns a place on Velynt by surviving a process designed to make it fail: the rule is written down first, tested on data it never saw, charged every cost, and judged by a statistical bound rather than a best case.
Pre-registered
Rules are written down before any test runs, so nothing is fitted to the result.
Out-of-sample
Each rule is judged on data it never saw, with a statistical bound, not a best case.
Costs included
Spread, commission, swap and slippage are charged in every number you see.
Rejections published
What was tested and failed is listed on each profile. Most ideas fail.
Dated, from the log
Every figure is computed from the agent's own trade log, with the start date on the chart.
No projections
No expected returns, no targets, no testimonials with returns. Records only.
What this looked like for Trend 14
The adopted rules and the rejected ones, with dates. The full research notes sit behind a free account.
- Pre-registered trend rule2026-10-24
The core rule passed an out-of-sample test on broker data with a positive lower bound on the Sharpe-ratio difference.
- Profit lock: half at +0.5R, rest to breakeven2026-10-29
Tested against eight alternative locks on the same trades; the only one that improved the result without cutting winners short.
- Pullback entry timing2026-11-01
16,529 historical orders: entries improved by 0.3 bp before 2023 and 0.4 bp after, with the gain concentrated in crypto.
- Regime scale2026-10-31
Risk is reduced when trends across the book are weak; better in both test periods, adopted as a risk-only control.
- Learned cluster budgets2026-11-01
Weighting asset classes by recent performance: no improvement out of sample.
- Event de-risk before FOMC and payrolls2026-11-01
Halving the book before scheduled events changed nothing measurable.
- 4-hour breakout sleeve2026-11-01
Loses after costs on every test period; not deployed.
- Short-term index mean reversion, carry, calendar effects, 40+ others2026-10-30
Each pre-registered, tested and rejected; the full list is in the research notes for subscribers.