Can one trading intent survive a change of venue?

A trading intent should preserve its economic constraints when an execution adapter changes. We propose testing that claim with two synthetic venues before asking whether any particular venue integration is useful. This is an unrun research design, with no announcement about DXAP venue support or a future integration schedule.

The continuous record companion describes historical agents operating in two materially different environments: bounded token swaps on Base and a perpetuals research fleet. Their shared harness lineage supplies a reason to study translation, while their different execution assumptions prevent treating transfer as established. A position expressed as a dollar exposure has different consequences when the instrument, margin model or settlement currency changes.

Two acceptable payloads, two different exposures

Consider an illustrative intent: add at most $1,000 of long exposure, permit at most $4 of immediate execution cost, and preserve an existing protective order. Synthetic venue A accepts quantities in asset units with a 0.01-unit step. At a $100 reference price, ten units represent $1,000. Synthetic venue B accepts contracts with a fixed $50 notional face value per contract and whole-contract quantities, so twenty contracts represent the same intended exposure.

Now move the reference price to $103 before submission. Sending ten units to A produces $1,030 exposure. Sending twenty fixed-dollar contracts to B remains $1,000. A serializer can accept both payloads while only one preserves the original cap. The fixture should expect A to recompute a quantity of 9.70 units, or $999.10 at the new price, subject to the declared price bound. It should also require a fresh check of protection semantics.

The example deliberately chooses simple multipliers. A broader fixture would cover settlement conversion, account position modes and different meanings of closing an exposure. Each case needs an expected economic state after acceptance rather than an expected string alone. Where the venue rules cannot represent the intent, the expected result is a specific rejection with the unrepresentable constraint identified.

The comparison we would run

Freeze the mandate, account state and market observation. Feed each adapter an identical typed intent and score exposure preservation, cost-cap compliance and protection continuity separately. Randomize adapter order to avoid learning from the first answer. Retain rejected translations in the denominator; otherwise a narrow adapter can appear perfect by accepting only easy cases.

The controls paper companion grounds this design in the recorded path from mandate to typed action, policy checks and reconciliation. Its bounded construction evidence motivates checking that path; it does not measure these synthetic adapters. A useful first output would be a matrix showing which economic constraints survive each translation and which require explicit owner revision. Cross-venue forecasting quality would remain a later, separate question.