What is "Cannonisable"? Write an introduction explaining the input data requirements.

[annayqho]

• radial velocity shifted
• common wavelength grid

[davidwhogg]

• need inverse variance estimates, one per pixel
• common line-spread function

[davidwhogg]

• consistent continuum normalization based on fitting or means, but not medians or quantiles!

[annayqho]

just to make sure I understand that last point: incoming spectra should not be continuum-normalized with medians or quantiles, because we are subsequently going to continuum-normalize with medians?

[davidwhogg]

No, we will NEVER normalize with medians -- in the Cannon we only used the quantiles as a starting point, and then we iterated to get a list of wavelengths that are "continuum safe" and then we fit a curve through those points. If the input data have had anything done to them that involves medians or quantiles, it will break the model -- or distort it as a function of SNR.

[annayqho]

ok, this information lives in section 2.2 of the current draft of the paper.