I was able to work through the dataframe summation issues on my own, arriving at a dataframe with allocations for each security in the limited universe:
                             
Equity(19657 [XLI])   0.123386
Equity(19662 [XLY])   0.119187
Equity(32015 [IHI])   0.223302

Then, I was able to manipulate this dataframe into dictionary format, using the .T.to_dict('list') function:

{Equity(19657 [XLI]): [0.10066003341707783], Equity(19662 [XLY]): [0.15936743400110154], Equity(32015 [IHI]): [0.25590631006228276]}

When I tried to use the dictionary formatted output directly in the Optimize Target Weights function, I got the following error message:

TypeError: TargetWeights() expected a value with dtype 'float64' or 'int64' for argument 'weights', but got 'object' instead.  

Can anyone help me get this into format required by the Optimize Target Weights function:

objective = opt.TargetWeights(weights)

Apparently, the required format for weights is:

{XLI: 0.10066003341707783, XLY: 0.15936743400110154, IHI: 0.25590631006228276}

The nice thing about Python is it isn't very strict about types. The TargetWeights objective will accept either a dict or a series (check out the docs here). The nice thing about using a pandas series is that is simply a column of a pipeline dataframe. So assume that 'weight' is defined as a column in the pipeline. Then, the following will use that weight value to order the associated weights

    # Fetch the pipeline data  
    pipe_data = pipeline_output('my_pipe')

    # Determine weights  
    # In this case the calculations were already done in the pipeline  
    # Just use the 'weight' column  
    weights = pipe_data.weight

    # Run the order_optimal_portfolio method to place orders  
    order_optimal_portfolio(objective=opt.TargetWeights(weights), constraints=[])

One could also turn the 'weights' series into a dict and get the same results but that isn't necessary.

    # The following works the same but isn't necessary  
    weights = pipe_data.weight.to_dict()

See the attached algo. Good luck.

Thanks very much, Dan, for your clean and elegant solution, as well as your thoughtful comments.