One of the most interesting economic 'artefacts' featured on the campus of my alma mater is the Philips' machine or Monetary National Income Analogue Computer (MONIAC) pictured below.
MONIAC was a hydraulic computer, with a rudimentary macroeconomic model, build by none other than William Philipps as a part of his dissertation. MONIAC could be used to visualize and even predict several macroeconomic variables depending on its calibration (such as interest rates or prices resp. inflation). The machine was originally intended just for classroom use (Bissell, 2007), but eventually several research/policy institutes around the world decided to commission the machine as well (for example New Zealand Institute of Economic Research bought one). This makes me wonder if the machine was ever accurate enough to be useful outside classroom (e.g. research/policy analysis/forecasting). Since it is based on rudimentary macroeconomic model I would expect relatively large forecasting errors.
I tried to do research on the MONIAC's accuracy but I am finding conflicting reports. According to Wikipedia:
To their surprise, Phillips and his associate Walter Newlyn found that MONIAC could be calibrated to an accuracy of 2%.
but no reference to the work where Phillips and Newlyn made the claim is provided.
I dug bit deeper and managed to find paper (Bissell, 2007) that claims the MONIAC's accuracy was reportedly only $\pm 4\%$ but again the claim is not directly supported by citation. In the next paragraph the paper cites Newlyn (1950) so I thought it might have the answer (since Wikipedia mentions Newlyn as well) but it turned out it does not mention any number for the accuracy of the machine.
This leads me to my question, is there any primary source in the economic literature for the accuracy of the MONIAC? If so what level of accuracy does it report?