diff --git a/src/constants.odin b/src/constants.odin
index 1e59da8..2195745 100644
--- a/src/constants.odin
+++ b/src/constants.odin
@@ -8,19 +8,24 @@ package shoal
 //	POST_UPDATE  /   (signal_system clears the buses here)
 //	PRE_DRAW     \
 //	DRAW          | every frame, the raylib drawing part
-//	POST_DRAW    /
+//	LATE_DRAW    /   (LATE_DRAW runs after every DRAW system, for overlays)
+//	POST_DRAW    once per frame, after all drawing
 //	DESTROY      once, after the loop exits and the window closes
 //
 // DefaultPlugin opens the raylib frame at PRE_DRAW (BeginDrawing + clear) and
 // closes it at POST_DRAW (EndDrawing), so DRAW is where your draw calls go.
+// renderers that flush accumulated draw state (like the ui plugin replaying
+// microui's command list) go on LATE_DRAW, so they run after all game DRAW
+// systems but still inside the raylib frame.
 STARTUP :: 0
 PRE_UPDATE :: 1
 UPDATE :: 2
 POST_UPDATE :: 3
 PRE_DRAW :: 4
 DRAW :: 5
-POST_DRAW :: 6
-DESTROY :: 7
+LATE_DRAW :: 6
+POST_DRAW :: 7
+DESTROY :: 8
 
 // easing ids for the (wip) Tween.ease field. see tween.odin, nothing reads
 // them yet
diff --git a/src/ecs.odin b/src/ecs.odin
index e5aae25..e57f278 100644
--- a/src/ecs.odin
+++ b/src/ecs.odin
@@ -79,7 +79,8 @@ get_world :: proc(app: ^App) -> ^World {
 
 // run every plugin (in registration order), then the STARTUP systems once,
 // then the frame loop: PRE_UPDATE -> UPDATE -> POST_UPDATE -> PRE_DRAW ->
-// DRAW -> POST_DRAW, over and over until rl.WindowShouldClose() says stop,
+// DRAW -> LATE_DRAW -> POST_DRAW, over and over until rl.WindowShouldClose()
+// says stop,
 // and finally the DESTROY systems. world systems get handed the app's own
 // world, the same one run() is holding. there's no second copy
 run :: proc(app: ^App) {
@@ -94,6 +95,7 @@ run :: proc(app: ^App) {
 		for s in get_system(app, POST_UPDATE) {run_system(app, s)}
 		for s in get_system(app, PRE_DRAW) {run_system(app, s)}
 		for s in get_system(app, DRAW) {run_system(app, s)}
+		for s in get_system(app, LATE_DRAW) {run_system(app, s)}
 		for s in get_system(app, POST_DRAW) {run_system(app, s)}
 	}
 
diff --git a/ui/rlmu.odin b/ui/rlmu.odin
index adb075a..71cd408 100644
--- a/ui/rlmu.odin
+++ b/ui/rlmu.odin
@@ -87,7 +87,7 @@ begin :: proc(state: ^RLMU_State = global_state) -> ^mu.Context {
 }
 
 // tell microui the frame is finished, then replay its command list as raylib
-// draw calls (and close any lingering scissor). UIPlugin runs it at POST_DRAW
+// draw calls (and close any lingering scissor). UIPlugin runs it at LATE_DRAW
 end :: proc(state := global_state) {
 	// Tell mui that we're done drawing ui
 	mu.end(&state.ctx)
diff --git a/ui/shoal.odin b/ui/shoal.odin
index 6fdbcd2..d6a0efa 100644
--- a/ui/shoal.odin
+++ b/ui/shoal.odin
@@ -11,16 +11,14 @@ ui_ctx: ^ml.Context
 //
 //	STARTUP     init() bakes the microui atlas into a texture
 //	PRE_DRAW    begin() forwards raylib input, starts the frame, stashes ui_ctx
-//	POST_DRAW   the frame's temp arena is freed, then end() renders the ui
+//	LATE_DRAW   end() renders the ui after all game drawing, still in-frame
+//	POST_DRAW   the frame's temp arena is freed
 //	DESTROY     destroy() unloads the atlas texture
-//
-// (systems within a schedule run in registration order, so POST_DRAW frees
-// the arena first and renders second)
 UIPlugin :: proc(app: ^ecs.App) {
 	ecs.add_system(app, ecs.STARTUP, proc() {init()})
 	ecs.add_system(app, ecs.POST_DRAW, proc() {free_all(context.temp_allocator)})
 	ecs.add_system(app, ecs.PRE_DRAW, proc() {ui_ctx = begin()})
-	ecs.add_system(app, ecs.POST_DRAW, proc() {end()})
+	ecs.add_system(app, ecs.LATE_DRAW, proc() {end()})
 	ecs.add_system(app, ecs.DESTROY, proc() {destroy()})
 }