diff --git a/aephysics/dynamics/module.ae b/aephysics/dynamics/module.ae
index 7df1328..01004b8 100644
--- a/aephysics/dynamics/module.ae
+++ b/aephysics/dynamics/module.ae
@@ -395,24 +395,27 @@ struct BodyState {
 }
 
 // What integration and collision need of a body.
+// What every constraint prepare reads first (the pose, the masses), then
+// the collide pass's extents, the integration's loads, and the time of
+// impact's start pose.
 struct BodySim {
     transform: Transform           // of the body's origin
     center: Vec3                   // of mass, in world space
-    rotation0: Quat                // for the time of impact
-    center0: Vec3
     local_center: Vec3
-    force: Vec3
-    torque: Vec3
     inv_mass: float
-    inv_inertia_local: Matrix3
     inv_inertia_world: Matrix3
-    min_extent: float
+    body_id: int
+    flags: int
     max_extent: Vec3
+    min_extent: float
+    force: Vec3
+    torque: Vec3
     linear_damping: float
     angular_damping: float
     gravity_scale: float
-    body_id: int
-    flags: int
+    inv_inertia_local: Matrix3
+    rotation0: Quat                // for the time of impact
+    center0: Vec3
 }
 
 // A shape in the world: its geometry (the shape module's Shape, with
@@ -833,13 +836,26 @@ struct RevoluteJoint {
     enable_limit: bool
 }
 
+// The point constraint's fields first (every solve reads them), then the
+// impulses and axes the warm start reads, the settings after.
 struct SphericalJoint {
+    enable_spring: bool
+    enable_motor: bool
+    enable_cone_limit: bool
+    enable_twist_limit: bool
+    index_a: int
+    index_b: int
     linear_impulse: Vec3
+    frame_a: Transform
+    frame_b: Transform
+    delta_center: Vec3
     spring_impulse: Vec3
     motor_impulse: Vec3
     lower_twist_impulse: float
     upper_twist_impulse: float
     swing_impulse: float
+    swing_axis: Vec3
+    twist_jacobian: Vec3
     hertz: float
     damping_ratio: float
     max_motor_torque: float
@@ -849,21 +865,10 @@ struct SphericalJoint {
     cone_angle: float
     cone_axis: Vec3                // unit, in frame A
     target_rotation: Quat
-    index_a: int
-    index_b: int
-    frame_a: Transform
-    frame_b: Transform
-    delta_center: Vec3
-    swing_axis: Vec3
-    twist_jacobian: Vec3
     rotation_mass: Matrix3
     swing_mass: float
     twist_mass: float
     spring_softness: Softness
-    enable_spring: bool
-    enable_motor: bool
-    enable_cone_limit: bool
-    enable_twist_limit: bool
 }
 
 struct WeldJoint {
@@ -927,24 +932,27 @@ struct WheelJoint {
 // (filled by the solver's prepare), the constraint's softness, the
 // thresholds, and the kind's data in the block at the end (the
 // reference's union: read through distance_joint_of and its siblings).
+// What the warm start and the solve read comes first and the kind's data
+// right after it, so a joint's pass touches as few lines as it can; the
+// frames and settings only the prepare and the API read go last.
 struct JointSim {
-    joint_id: int
     kind: int
-    body_id_a: int
-    body_id_b: int
-    local_frame_a: Transform
-    local_frame_b: Transform
+    fixed_rotation: bool
     inv_mass_a: float
     inv_mass_b: float
     inv_inertia_a: Matrix3
     inv_inertia_b: Matrix3
-    constraint_hertz: float
-    constraint_damping_ratio: float
     constraint_softness: Softness
-    force_threshold: float
+    force_threshold: float         // the solve's joint event test
     torque_threshold: float
-    fixed_rotation: bool
     data: long[68]                 // JOINT_DATA_BYTES
+    joint_id: int
+    body_id_a: int
+    body_id_b: int
+    local_frame_a: Transform
+    local_frame_b: Transform
+    constraint_hertz: float
+    constraint_damping_ratio: float
 }
 
 struct JointPair {
diff --git a/aephysics/solver/module.ae b/aephysics/solver/module.ae
index 8b849e7..a98f500 100644
--- a/aephysics/solver/module.ae
+++ b/aephysics/solver/module.ae
@@ -97,13 +97,18 @@ stage_clock() -> float[] {
     return totals
 }
 
-// The averages, in the reference's own order and words.
+// The averages since the last report, in the reference's own order and
+// words; the clocks then start over, so a benchmark's next scene reports
+// its own steps rather than an average over every scene so far.
 stage_report() {
     if stages_timed() == false || g_stage_steps == 0 || g_stage_ms == null { return }
     clock = stage_clock()
     n = g_stage_steps as float
     println("aephysics stages ms/step: prepare_joints ${clock[STAGE_PREPARE_JOINTS] / n} prepare_wide ${clock[STAGE_PREPARE_WIDE] / n} prepare_contacts ${clock[STAGE_PREPARE_CONTACTS] / n} integrate_v ${clock[STAGE_INTEGRATE_VELOCITIES] / n} warm ${clock[STAGE_WARM_START] / n} solve ${clock[STAGE_SOLVE] / n} integrate_p ${clock[STAGE_INTEGRATE_POSITIONS] / n} relax ${clock[STAGE_RELAX] / n} restitution ${clock[STAGE_RESTITUTION] / n} store ${clock[STAGE_STORE_WIDE] / n}")
     basics.free_bytes(clock as ptr, STAGE_COUNT * 8)
+    basics.free_bytes(g_stage_ms, native.MAX_WORKERS * STAGE_COUNT * 8)
+    g_stage_ms = null
+    g_stage_steps = 0
 }
 
 const NULL_INDEX = 0 - 1
diff --git a/bench/RESULTS.md b/bench/RESULTS.md
index 28a1749..af600d8 100644
--- a/bench/RESULTS.md
+++ b/bench/RESULTS.md
@@ -626,7 +626,50 @@ the eight-wide (AVX2) contact stages:
 | store | 0.48 | 0.38 |
 
 Matching that needs eight float lanes. Aether's `std.lanes` has `f32x4`
-and no `f32x8`.
+and no `f32x8` (aether-lang-dev/aether#2428).
+
+### The joint grid, stage by stage (2026-10-05)
+
+The joint grid has no contacts, so the eight-wide lanes don't explain its
+gap. It is the scalar spherical joint, the same code as the reference's.
+Each scene run alone (`AEPHYSICS_BENCH_SCENE=grid`), one thread:
+
+| stage | aephysics before | aephysics after | Box3D `e77352c` |
+|---|---|---|---|
+| prepare joints | 1.21-1.42 ms | 1.27-1.49 | 0.61 |
+| integrate velocities | 1.25 | 1.25 | 1.15 |
+| warm start | 2.09-2.95 | 2.04-2.39 | 1.31 |
+| solve | 4.36-4.98 | 3.90-4.14 | 3.33 |
+| relax | 4.21-4.88 | 3.88-4.14 | 3.22 |
+| step | 14.4-17.0 | 13.6-14.7 | 10.4-10.6 |
+
+"After" is #107's layout. The solve and relax were reading about eleven
+cache lines per joint, because what they need was spread through the
+records. `JointSim` kept the local frames (read only by the prepare)
+between its header and the masses, and `SphericalJoint` kept the enable
+flags, tested first, 536 bytes in. `JointSim` now puts the solve's fields
+first: kind, masses, inertias, softness and the joint event thresholds.
+The kind's data comes next, and the frames and settings go last.
+`SphericalJoint` puts its flags, body indices, impulse, frames and centre
+delta first, then what the warm start reads. The arithmetic is unchanged
+and the height sum is the same.
+
+`BodySim` now starts with what every constraint prepare reads: the
+transform, centres, inverse mass and world inertia, in three lines
+instead of five. That took the joint prepare from 1.43-1.49 to 1.26-1.37 ms
+and the many pyramids' wide prepare from 3.53-3.60 to 3.38-3.49. The
+large pyramid was about even.
+
+What's left of the gap is the records' size in double precision. Our
+`JointSim` is about 900 bytes, against the reference's float union. A
+`BodyState` is 112 bytes against 64.
+
+Before #107, `AEPHYSICS_STAGES` never reset its clocks between scenes. A
+benchmark's second and third scenes reported averages over every step
+since the start, so the joint grid showed a wide prepare and a store
+although it has no contacts. Stage figures for the many pyramids and the
+joint grid from before then (in this file and on the issues) carry the
+earlier scenes' stages. The large pyramid, which runs first, was right.
 
 ### Stage by stage