Algorithmic trading · audit-honest · live demo
Quantitative trading, honestly audited.
Algorithms held to the standard of published science. Every hypothesis pre-registered. Every result audited, deflated, and made public — including the failures.
The fleet
Eight specialists, six live, two paused. Each bot trades exactly one strategy. Signals do not mix across bots — a failure mode in one cannot cascade.
| Bot | Symbol | TF | Sharpe | CAGR | Max DD | Trades / yr | Status |
|---|---|---|---|---|---|---|---|
| xag | XAG | 1h | +1.75 | +25.9% | -9.4% | 46 | Live |
| xau | XAU | 1h | +1.23 | +35.6% | -17.0% | 110 | Live |
| gc | XAU | 1d | +1.11 | +9.8% | -8.5% | 8 | Live |
| event | XAU | 1h | +1.00 | +7.2% | -9.5% | 15 | Live |
| ro | XAU | 1h | +0.75 | +4.2% | -8.0% | 3 | Live |
| event5m | XAU | 5m | — | — | — | — | Live |
| bb | XAU | 1h | +0.39 | -3.5% | -8.0% | 39 | Paused |
| btc | BTC | 1h | +0.04 | -1.0% | -20.0% | 97 | Paused |
Sharpe, CAGR, and Max DD from Engine + daily_sharpe backtest (audit framework). Paused bots remain registered but trade nothing live.
Methodology
Independent specialists
Eight bots, eight strategies, eight magic numbers, eight log streams. Each runs in its own process, supervised by an external watchdog. If one bot crashes, the others continue without it.
Pre-registered audits
Every hypothesis is documented in /docs before backtesting starts: the entry rule, the exit rule, the parameter sweep, the accept criteria. If the final result does not match what was promised in writing, the sprint is rejected. Twenty-three sprints tested in 2026. Two shipped.
Five-layer risk
Per-strategy drawdown halt (-10%), account-wide margin watchdog (<150% margin level), daily-loss circuit breaker (-2% to -4% by profile), session kill switch (-15% to -30%), and the live state of each guard persists across restarts. A tripped kill cannot re-arm itself.
Performance
Backtest numbers below come from the audit-honest engine (Engine + daily_sharpe, zero-filled equity curve, no per-trade Sharpe inflation). Live expectations are deflated by the R83 audit's marginal Bonferroni / random-null findings — not by guesswork.
Backtest (8.3 yr)
2018 – 2026 (8.3 yr)Live forward (deflated)
R83-deflated+2-3% CAGR with smaller drawdowns. We are not Renaissance Medallion.
Past performance does not guarantee future results. The ensemble is calibrated for the 2018-2026 regime; gold buy-and-hold contributed most of the headline CAGR. Forward expectation deflated from IS +1.316 to +1.10 per R83 audit findings.
Live status
- Deployed
- 2026-04-30
- Broker
- Pepperstone Demo
- Account
- $1,400
- Profile
- aggressive
- Fleet
- 6 / 8
- Live trades
- 1
NFP 2026-05-08 — Michigan sentiment landmine. Reversal caught the bot at near-peak entry. Fixed by upcoming-event veto (blocks entry if another high-impact event imminent).
Brand family
Vrelum is the trading-bot arm of the Plovum group. The names share Latin and Slavic roots:
Latin/Romance root. Abundance, flow, rain — that which provides.
Croatian root vrelo. Spring, source of water — that which originates.
Vrel- (source) + Plovum's -um suffix. The source instrument — that which extracts.