Paper is not live: how we grade a strategy's evidence
Backtests, paper trading, testnet and mainnet answer different questions. Here is how Algonius keeps them apart, and why a good run never promotes a strategy.
The quickest way to make a trading strategy look good is to blur where its numbers came from: present a backtest curve as live results, or a paper return as a track record. We keep five grades of evidence apart, and a strategy moves forward only one grade at a time.
Five stages, one direction
- Research. Backtests and stress tests. Never shown as investable.
- Paper. Live market data, virtual funds, an independent ledger.
- Testnet. The full vault cycle on testnet: deposit, trade, redeem, reconcile.
- Limited mainnet. Our own capital first, then a small whitelist under strict capacity limits.
- Live. Open to eligible users, with results computed from on-chain fills.
No stage can be skipped. Paper cannot jump to live, and limited mainnet is not optional. Every promotion needs an independent review of the evidence at the current stage. Short-term returns never trigger one.
What each stage can and cannot tell you
- A backtest shows how the rules would have behaved on history the author could already see. It is good at exposing broken ideas, and poor at proving good ones.
- Paper trading runs the rules forward on live data, but with virtual money. Nothing actually fills, so the results rest on assumptions about prices, fees and slippage.
- Testnet proves the whole cycle with real contracts and real signatures: deposit, trade, redemption and reconciliation. But testnet markets are not real markets.
- Only mainnet fills, positions, fees and settlement from the venue can support a statement about real performance.
The rules we hold ourselves to
- Every number carries its grade. The Hub, the API, our logs and our reports label backtest, paper, testnet and mainnet separately, and never merge them.
- Mainnet results come only from verifiable venue data: fills, positions, fees and settlement. They never come from a static file, a code commit or a manual upload.
- Every figure can be recomputed from raw events. That covers net asset value, returns, drawdown and fees. Each figure carries its source, methodology version, period, checkpoint and freshness.
- Stale or unreconciled data stops new positions, and is never shown as healthy or real-time.
- History is permanent. Renaming, upgrading, pausing or closing a strategy never deletes or rewrites its record.
What this looks like today
Our first official strategy, Trend Rider · Majors, is in paper trading. Its current paper sample replays frozen public candles to validate the infrastructure. It has no real samples behind it and supports no profitability claim. So the Hub does not show its returns. It states what evidence is still required. When there is evidence worth reading, its grade will be on its face.