Phi dependence is completely coded. Changes were required in the packages irfs/handoff_response, likelihood Likelihood, map_tools. All parts were separately tested, here we perform an overall check before the tools are considered ready for science analysis.
A new "P6_V5_DIFFUSE" irf was generated, with phi dependence in the aeff. The pruned goodEvents dataset used was the same used to generate "P6_V3_DIFFUSE", so the 2 irf sets should be directly comparable.
First, pyIrfLoader was used to compare the olf aeff with the new, averaged over all phi. This should make the 2 results identical. Let's see:
Aeff comparison 
Again, second plot is the aeff averaged over phi, the average is calculated by the script I used to make the plots. The P6V3 plot is the usual one.
Same for acceptance: