Calibration Challenge

Overview

We propose to run a "Calibration Challenge" starting the first week of December 2015.

The goal will be to take a run with zeroed out calibrations and to see how many calibrations can be automatically extracted.

Test runs will be performed in the weeks leading up to this challenge.

Types of Tests

Various types of tests could be run to exercise our calibration procedures.

1. Calibration values are normalized to some neutral value (e.g. 1 or 0), and the calibrations are derived "ab initio".
   • Given the diversity of run conditions among the few production runs in the spring data, it makes sense to do this test with just one run at a time, and not combine runs. Therefore, we can test calibrations that only need ~1 runs' worth of data.
2. Current calibration values are perturbed by some amount, and procedures are run to see how well the current values are recovered.
   • The perturbations would be specific to each calibration. Different functional forms? Gaussian? Square wave?
   • This would give us a way of testing calibrations that require more than one run's worth of data to be done from an uncalibrated state.

Types of Data

References:

• Automated Timing Calibrations
• Spring 2015 Dataset Summary

In principle, simulations could be used as well. Some effort is needed to work on the software to support this.