Three tests, three questions
- Backtest. Would this strategy have worked on the history I have? It is fast and covers many market regimes, but every design choice was made with that history in view.
- Forward test. Does the signal still behave on data the strategy has never seen? The strategy runs on its schedule as new data arrives, and its signals and target positions are recorded.
- Paper run. Does the whole strategy hold up, orders and costs included? A paper run is a forward test that also sends its orders to a fill simulator and keeps an account, so fees, funding, skipped orders and equity are part of the result.
What the same pipeline makes identical
On Keel, a backtest, a paper run and a live run execute the same strategy pipeline. The data loaders, signals, sizing and rebalancing rules are the same objects, so a paper run cannot quietly run a different strategy from the one you backtested. The costs line up too: the backtest a paper run is compared against charges the same fee schedule and applies the same minimum order size, and both use real Hyperliquid funding rates.
What still differs
A paper run is not live trading. Keel's published fill model lists the differences it knows about:
- Fills are modelled from the venue mid with a fixed slippage, so order-book depth and partial fills are not simulated.
- The model does not round order sizes to each asset's lot size.
- The first fill happens when you start the run, part-way through a bar, not at a bar close.
- A held asset with no fresh mid is marked at the last mid Keel recorded for it, and equity is not snapshotted until every mark is fresh again.
- A stopped run freezes with its open positions at their final marks.
- A funding hour whose rate never arrives becomes a disclosed gap record instead of an estimate.
- A fill's mid is the last live observation of its minute. If that minute is later re-sealed from the venue's archive, the record shows the difference, and the fill itself never changes.
How Keel measures the gap
Every paper run is tracked against a backtest of the same strategy over the run's own window. The run's page shows whether it is tracking its backtest and the residual between the two. Because the pipeline and the cost schedule are shared, what remains in that residual comes from the differences listed above, which is the part a backtest cannot show you.
Measured backtest-versus-paper divergence
This section will report measured divergence across Keel's paper runs once enough of them have a track record. Until then it reports nothing: we describe the method above and publish no projected figures.
Try it
Backtest a strategy, then press Paper to run it forward on live Hyperliquid prices.