diff --git a/docs/superpowers/specs/2026-08-05-movementtest-longitudinal-time-step-design.md b/docs/superpowers/specs/2026-08-05-movementtest-longitudinal-time-step-design.md
new file mode 100644
index 0000000..86e5106
--- /dev/null
+++ b/docs/superpowers/specs/2026-08-05-movementtest-longitudinal-time-step-design.md
@@ -0,0 +1,26 @@
+# MovementTest Longitudinal Time-Step Design
+
+## Goal
+
+Reduce the observation-test longitudinal QP size without relaxing the solver accuracy, time budget, vehicle limits, or observe-only behavior.
+
+## Decision
+
+Add a configurable `OutputTimeStepSeconds` to the trajectory-observation MovementTest with a default of `0.10d` seconds.
+
+The EM time horizon remains `6d` seconds. At the parking-test speed limit of `0.20m/s`, a 0.10-second knot interval corresponds to at most 0.02m of travel between knots.
+
+## Scope
+
+- Add, snapshot, and validate `OutputTimeStepSeconds` in `TrajectoryObservationSettings`.
+- Expose the setting in the MovementTest as a public field with a default of `0.10d`.
+- Copy it to the session-local `EmPlannerConfiguration.Scheduling.OutputTimeStepSeconds`.
+- Verify custom settings reach the planning request and non-positive values are rejected.
+
+## Non-goals
+
+- Keep `TimeHorizonSeconds = 6d`.
+- Keep `ObserverPeriodSeconds = 0.05d`.
+- Keep `SolverTimeoutSeconds = 0.50d`, `MaximumOsqpIterations = 12000`, and all `1e-5d` solver tolerances.
+- Do not change production EM defaults, maps, vehicle constraints, or command output behavior.
+