In first-principle planning, one gathers evidence about the world, makes a characterization of the problem, and attempts to generate a plan--a sequence of operators that will, if executed, lead to a desired state. Planners construct a plan by simulating the results of applying each operator and choosing the next operator based on the predicted world state. When the plan is passed to the executive, this state information has been removed. The executive can have a difficult time determining why plan failure has occurred, or even that plan failure has occurred. In order to enable agents to recognize and recover from plan failure, we must find a way to put some information back into plans. I argue for an extended plan representation, EPs, or Execution Paths, that include the expected results of the operators in the plan. As implemented, in a case-based system, the results make error detection and recovery with EPs a fast process of comparison between the system's execution trace and its EPs. Bertha is a hand-eye system that builds walls from Duplo bricks. It has a library of EPs from which it selects a single EP to execute. It executes the actions from its active EP, one at a time and collects actions and results in an execution trace. When there is a mismatch between the results of the last action and the results in the EP, the system rematches the trace to the EPs in the library and selects a new EP that better matches the execution trace. Bertha is an elegant solution to the problems of recognizing plan failure and recovering from that failure. It accomplishes this through an extended plan representation whose results allow it to recognize failure and also to select a better plan. We support this claim with extensive testing in the wall-building domain.