diff --git a/README.md b/README.md
index 7e0dc6a..b5baca7 100644
--- a/README.md
+++ b/README.md
@@ -143,6 +143,14 @@ systemctl --user status external-brightness-sync.service
 journalctl --user -u external-brightness-sync.service -n 50 --no-pager
 ```
 
+## Design Notes
+
+Implementation decisions and failed attempts are documented in:
+
+```text
+docs/implementation-log.md
+```
+
 ## Uninstall
 
 ```bash
diff --git a/docs/implementation-log.md b/docs/implementation-log.md
new file mode 100644
index 0000000..0cac51e
--- /dev/null
+++ b/docs/implementation-log.md
@@ -0,0 +1,155 @@
+# Implementation Log
+
+This file documents the brightness/OSD design decisions, failed attempts, and current state.
+
+## Current Working Design
+
+The current setup intentionally keeps the brightness path simple:
+
+1. A brightness shortcut immediately changes the internal panel with `brightnessctl`.
+2. The OSD is shown or updated after the internal change.
+3. The external monitor follows through `external-brightness-sync.service`.
+
+The internal panel is the master. The external monitor is never adjusted independently by `wluma`.
+
+## Why Internal Is The Master
+
+Letting `wluma` control both the internal backlight and the external DDC/CI monitor independently caused occasional drift between displays.
+
+The cleaner design is:
+
+- `wluma` controls only `/sys/class/backlight/intel_backlight`
+- `external-brightness-sync.service` reads the internal percentage
+- `ddcutil` sets the external monitor to the same percentage
+
+This keeps one source of truth for automatic brightness.
+
+## Native GNOME OSD Attempt
+
+GNOME Shell exposes a DBus method:
+
+```text
+org.gnome.Shell.ShowOSD
+```
+
+On GNOME Shell 49.6, external calls to this method failed with:
+
+```text
+GDBus.Error:org.freedesktop.DBus.Error.AccessDenied: ShowOSD is not allowed
+```
+
+Because of that, the project uses a small GTK4/libadwaita OSD instead of trying to call GNOME Shell's private OSD API.
+
+## GTK OSD Positioning On Wayland
+
+The first GTK OSD version used a normal GTK window. It worked, but GNOME Wayland placed it like a normal window, so it appeared near the center of the screen.
+
+Under Wayland, normal clients cannot reliably choose arbitrary screen coordinates. A top-edge OSD should use the layer-shell protocol.
+
+The project therefore supports `gtk4-layer-shell`:
+
+```bash
+sudo dnf install -y gtk4-layer-shell
+```
+
+When available, the OSD uses it for proper top-edge positioning.
+
+Without `gtk4-layer-shell`, the OSD still works, but GNOME may place it in the normal window position.
+
+## OSD Flicker Fix
+
+An early OSD version killed the existing OSD process on every new brightness key press.
+
+That caused a chopped feeling:
+
+```text
+key press -> old OSD exits -> new OSD starts
+```
+
+The current OSD instead keeps a running process alive and updates it:
+
+```text
+key press -> write new value -> signal running OSD -> reset hide timer
+```
+
+Files used for that:
+
+```text
+/tmp/brightness-osd.pid
+/tmp/brightness-osd.value
+```
+
+This smooths repeated key presses without delaying the actual brightness change.
+
+## Failed Debounce Attempt
+
+There was an attempt to coalesce multiple fast brightness key presses into one final brightness operation.
+
+The attempted flow was:
+
+```text
+key press -> store target value -> show target in OSD -> wait for quiet period -> run brightnessctl once
+```
+
+The debounce delay was controlled by:
+
+```text
+BRIGHTNESS_APPLY_DEBOUNCE_MS="180"
+```
+
+This was reverted because it made brightness changes feel laggy.
+
+### Why It Felt Bad
+
+Brightness keys are expected to affect the panel immediately.
+
+Debouncing the actual `brightnessctl` call introduced a visible delay:
+
+```text
+press keys -> screen does not change immediately -> final jump happens later
+```
+
+That is unpleasant for display brightness because visual feedback matters more than reducing the number of brightness writes.
+
+The external monitor already has some delay because `ddcutil` is slower than an internal backlight write. Delaying the internal display as well made the whole interaction feel worse.
+
+### Lesson
+
+Do not debounce the real brightness change.
+
+Only smooth the OSD:
+
+```text
+brightnessctl immediately
+OSD update/reuse
+external sync follows
+```
+
+## Reverted Commit
+
+The debounce attempt was committed and then reverted:
+
+```text
+324505b Debounce brightness key bursts
+a84963e Revert "Debounce brightness key bursts"
+```
+
+The reverted implementation added:
+
+```text
+bin/brightness-step
+BRIGHTNESS_APPLY_DEBOUNCE_MS
+```
+
+Those are intentionally not part of the current working design.
+
+## Current Recommendation
+
+For brightness controls:
+
+- keep internal brightness immediate
+- keep OSD process reusable instead of restarting it
+- keep external sync asynchronous
+- avoid batching or delaying real brightness changes
+
+This gives the most responsive feel while preserving internal/external sync.