I recommend setting an explicit Kanban style policy concerning when to take the time to update the Change Canvas. Options include whenever an experiment is finished, or on a set cadence, perhaps biweekly or monthly.
I often run a session with change participants to annotate the canvas for correctness, again using green, yellow, and red markers for portions of the change model that are believed to be correct, partially correct, and completely incorrect. Participants may also annotate the canvas with a statement that describes "why" a certain assumption is not turning out to be true.
The Change Canvas can then be updated to reflect the latest understanding and newest learning. This can include looking at the existing Improvement Experiment backlog and re-factoring it to consider the new information added to the Change Canvas.
Performing a Change PivotWhen a large number of experiments do not meet expectations, then it is time to consider a change pivot. A change pivot involves wholesale modifications to the Change Canvas, and underlying change model. When executing a change pivot, one key aspect of the change model is altered, while keeping another aspect intact. Examples of a change pivot include:
1. Choosing a different set of change participants
2. Selecting a different set of methods, tools or techniques to adopt
3. Switching up actions and tactics, perhaps going from light touch to high touch or vice versa
4. Scaling back benefits to better reflect time commitments that your change participants can make
For more check out the Lean Change Method: Managing Agile Transformation with Kanban, Kotter, and Lean Startup Thinking .