The README says:

A good rule to follow is that if you always control lights together (like bulbs in a ceiling fixture), then they should be in a Zigbee group. Then, only expose the group (and not individual bulbs) in Home Assistant Dashboards and external systems like Google Home or Apple HomeKit.

Unfortunately, it's often the case that I want to change the color or brightness of single lights, and I have several overlapping zones of lights that I control. This means that, when controlling lights at a granularity lower than the groups that are exposed to Adaptive Lighting, I get annoying behavior:

• Turning on a single light or subgroup of lights causes the whole group to be turned on by AL
• Modifying a single light or subgroup of lights isn't noticed by AL, so the lights get changed back on the next adapt cycle

Pointing AL at each individual light or all the intersections of subgroups of lights destroys performance.

One potential solution to this issue would be to allow specifying to AL a separate set of "watched" lights from the control groups. These watched lights would not be used for control, but would be used to trigger "take over control". If a watched light is changed manually, or if a watched light is set to a different on/off state from others, take over control would be triggered. For example:

adaptive_lighting:
  lights:
    - light.living_room_lights
  watched_lights:
    - light.table_lamp
    - light.sconce_left
    - light.sconce_right
    # etc.

would only send adaptations to light.living_room_lights. If light.table_lamp is on, but light.sconce_left is on, however, light.living_room_lights would not be adapted. Similarly, if light.table_lamp has its color changed manually, light.living_room_lights would stop being adapted.