Trading-hour transitions need explicit agent fixtures

Trading-hour transitions change what an agent can infer and what its intended action means. We propose a synthetic calendar suite that tests opening, closing and interrupted sessions with explicit market-state labels. This is an unrun cross-environment research fixture, rather than a statement about DXAP adding a venue or instrument.

The continuous record companion describes a historical perpetuals fleet including synthetic instruments. Its scope motivates calendar-aware evaluation while providing no universal rule for their trading sessions. The controls paper companion supplies the state-to-policy-to-execution chain into which a session status can be inserted.

The last quote before a closed interval

Consider a synthetic instrument whose declared session runs from 09:30 to 16:00 local exchange time. At 15:59:58 its latest tradable quote is $100. At 16:00:02 the feed still displays $100, but the session status is closed and the next permitted regular-session execution is tomorrow. An agent that treats the display as a current executable opportunity confuses price availability with trading eligibility.

An illustrative order expires at 16:00:00. A decision started at 15:59:55 finishes at 16:00:03. The fixture expects the runtime to recheck session status and expiry at the action boundary. A successful plan drafted before the close may become ineligible by submission time. Define an explicit state for this transition rather than silently substituting the next opening price.

Add a daylight-saving boundary using the synthetic calendar's declared timezone. If the local open stays 09:30 while the UTC offset changes from minus five to minus four hours, the UTC open shifts from 14:30 to 13:30. A fixed UTC schedule would be an hour late. This arithmetic concerns the fixture's chosen rule; actual venue calendars require their own authoritative specifications.

Price response and eligibility are separate tests

The suite should include a scheduled halt, an unexpected interruption and an auction-only interval. Provide the expected interpretation of resting orders and new orders for each synthetic venue. An agent may correctly predict an opening gap while proposing an action that the declared session rules reject. Score both tasks separately.

We would freeze calendar version, timezone database version and market-state snapshots in the manifest. Carry holidays and early closes as explicit cases with expected outcomes. Keep unknown session status visible, so a missing calendar entry cannot become an invented open session. A no-trade decision should identify which eligibility condition remains unresolved.

The output would be a transition coverage table showing which decisions remain valid across each boundary. Its value is operational: it reveals whether session knowledge survives long inference, stale quote display and state refresh. A later price-forecast study could investigate session-specific patterns, but the initial fixture asks whether the harness respects the calendar and execution state it was given.