3. Testing

3.1. Verification

Verification tests do not involve any real-world data. The main aim is to make sure that the pipeline is unbiased. These runs involve treating some subset of the simulations as the data, and then averaging the results to residuals with respect to the input. In practice, we expect to find percent level residuals that constiute the MC bias. We provide the following:

  • Baseline MC with 1d sims

  • Baseline MC with tiled sims

  • Baseline MC with time-domain sim

  • Extragalactic foreground test

  • Galactic foreground test

  • Varied cosmology runs

The last run is crucial in re-inforcing our confidence in the percent level MC bias, when percent level precision is called for in the data. We expect the MC bias to primarily be dictated by things like the mask which our simulations accurately capture. The varied cosmology run looks for a cosmology dependence in the MC bias, which should be ideally be negligible.

3.2. Null tests

We run a series of tests which should result in bandpowers we expect to be consistent with zero. These come in two classes, curl tests and data split tests. The latter involves taking two splittings of the data, differencing them and running them through the lensing pipeline. In the absence of systematics, we expect the CMB and foregrounds to cancel. If the debiasing steps are working properly, the resulting bandpowers should be consistent with zero. We provide:

  • Curl tests

  • 90 GHz - 150 GHz null

  • Individual array split differences

  • Co-added split difference

  • Day vs. night difference

  • Difference from Planck nulls

3.3. Bandpower consistency tests

We run the lensing pipeline on various splittings of the data. We then examine the difference of bandpowers, which should be consistent with null. We provide the following:

  • Individual array consistency

  • Individual array consistency (tiled sims)

  • Polarization combination consistency

  • Isotropy tests

  • Minimum multipole and maximum multipole variations

  • Night-only

  • Cross-only split-based estimator

  • Aggressive dust masks

  • 353 GHz subtraction

  • Samples from beam error

  • Samples from calibration error