diff --git a/src/test/kc/complex/polygon/design.txt b/src/test/kc/complex/polygon/design.txt
new file mode 100644
index 000000000..e4ccde896
--- /dev/null
+++ b/src/test/kc/complex/polygon/design.txt
@@ -0,0 +1,45 @@
+
+
+*Line Buffer*
+
+Used for rendering lines into. Organized with linear addressed columns of 8 bits, allowing the line routine to addess the y-coordinate using X/Y-addressing.
+
+*Screen Buffer*
+
+Used for filling & showing to the user. There are 2 screen buffers allowing for double buffering. 
+Memory organization is  be adapted to the specific display. This means the same line routines can be used for charsets, bitmaps, sprites etc. 
+
+*Algorithm*
+
+1. Move points
+2. Clear line buffer 
+3. Render lines into line buffer
+4. Wait for screen buffer being available for rendering (mostly instant)
+5. EOR-fill from line buffer into screen buffer
+6. Mark screen buffer for showing, swap screen buffer
+
+*Line Drawing*
+
+Optimized line drawing routines for the 8 different cases. (xd positive/negative, yd , xd-yd positive/negative)
+
+- Lines are organized in a clockwise fashion ensuring that xd positive means the fill starts and xd negative means the fill ends.
+- xd positive (start fill): steep slope lines render the top-most pixel, flat slope lines render the actual pixel.
+- xd negative (stop fill): steep slope lines render the pixel below bottom-most pixel, flat slope lines render below the actual pixel.
+- unrolled line routines with code for each individual x-pixel position can plot lines quite fast
+- unrolled line routines can be called by writing an RTS at code of the stop x-position and JSR'ing into the code for the start X-position. 
+- The RTS must be fixed again after return.
+
+*EOR filling*
+
+Unrolled EOR-filler from the line buffer. The screen buffer does not need to be cleared because the filler always fills all bytes.
+
+*Effective Canvas Shape*
+
+Both the line canvas and screen canvas can benefit from identifying the effective canvas shape where an object can be rendered.
+For instance a freely 3D-rotated object may only occupy a circular shaped canvas.
+By only clearing & filling this effective canvas these time consuming routines arebecome faster.
+
+It might also be worth dynamically detecting the canvas shape for each rendered frame - to allow even faster clear/fill. 
+This will require an efficient way of identifying min/max y-value in for each column that consumes less time than the time saved by not clearing/filling.  
+
+