The week by week discovery of new software problem reposts (PRs) is compared to an idealized curve defined by a goal of no new PR discovery after launch. The approach was back tested against NCS R1 and Flight 2A data files data and found to yield reasonable predictions up to 18 months before launch.

The following is a summary of the prediction methodology:

• PR's listed in PCMS are grouped in 7 day buckets based on creation date.
• PR's are included against documentation as well as software code (CSCI's) or command telemetry data files (Data PR's).
• Invalid or duplicate PR's that are rejected by a board are removed from the counts.
• A smooth Weibel curve is fit to a combination of
  • a) the observed data
  • b) a target set with 0 PR's each week for the first 26 weeks after the anticipated launch date.
  • c) interpolated between the data and launch date adding the predicted each week to the last data point's cumulative number of PRs.
  • Because of the interpolation, the fit is iterative, Excel's solver Add-in converges in less than 50 iterations
• The three (3) fitting parameters are the total number of PR's, a time constant, and a shape or symmetry factor.

The standards for reporting metrics using the Curve Fit Methodology are as follows:

• Use data sets that are amenable to this sort of analysis
  • Need to be comprehensive over the period analyzed
  • No arbitrary filters
• Clarify the pedigree of the data used
  • Identify the data base from which the data was extracted and any filters used
• Present the parameters used in the calculations to the degree that they are not obvious from the graphics