A Paper Engine Needs Its Own Evidence Label
By DX Research Group · · DXRG findings
The historical fleet mixed mostly paper execution with a smaller real-capital book. Margin, funding and fill assumptions change what its results establish.
Our historical perpetuals fleet used live prices, but most fills came from a paper engine. That engine filled at live mark with zero slippage and zero funding. Its 0.4% maintenance-margin placeholder liquidated roughly twice as late as real venue margining would. A live price feed alone therefore cannot establish live execution realism.
The continuous record covers June 8 to August 15, 2026 and reports 14,596 fills alongside a small real-capital book of 5,035 fills. The paper flags real-money overlap where relevant and pools its two systems at the metric level rather than joining every row into one homogeneous trading population. We retain those distinctions when interpreting the record.
Four disclosures that make comparison possible
An execution comparison should state the price source, fill rule, carrying costs and liquidation mechanism. These are four separate columns because changing any one can change the same strategy’s observed economics.
The price source tells us when and where a quote was observed. The fill rule tells us how a hypothetical order becomes a trade. Carrying costs include the assumptions about funding and other costs incurred while holding. Liquidation mechanics determine the loss path when collateral becomes insufficient. A simulation can use fresh market data while remaining optimistic on the other three dimensions.
For an illustrative long position, a favorable live mark says nothing by itself about whether an intended size was immediately available at that price. A later funding payment can change the outcome even if entry and exit prices match. A stricter margin rule can end the position before the paper path reaches its eventual recovery. These examples describe mechanisms, without asserting their frequency in the published fleet.
Why the real-capital subset matters
The 5,035 real-money fills provide a distinct evidence population. They cannot silently validate every paper-engine result, because analyses may cover different overlap, assets and periods. We would report each finding’s paper and real-capital membership explicitly and require the same metric definition across both.
A useful transfer test would replay known real orders using the paper engine and compare venue receipts against simulated fills, fees and next portfolio state. The relevant output is an error distribution with timestamps and order size, rather than a single claim that the simulator is realistic. This proposed test needs actual matched records before producing a numerical conclusion.
Our first paper companion describes a separate 21-day real-ETH event on Base. Its 12-token bounded market and 2.3% swap fee supplied a different execution environment. Real capital strengthens the evidence class for that run, while its market structure still limits transfer to perpetuals.
Evaluating an evolving platform
DXAP publicly describes execution on a user’s Hyperliquid account after external policy checks. That is a current product description, separate from historical paper-engine returns. A reader evaluating agentic trading should seek venue-bound orders and reconciled fills for live claims, and declared assumptions for replay claims. We can learn from both populations by keeping their labels visible all the way from the experiment to the article headline.