joebarbere
joebarbere on Hacker News
The figures are in the zip attached to the latest release (see hi/main.pdf for the rotation curve): [
https://github.com/joebarbere/jansky-research/releases/tag/v...](
https://github.com/joebarbere/jansky-research/releases/tag/v...) or you can regenerate them yourself, python -m jansky_research.hi re-downloads the survey data and rebuilds the plot.
Sadly Claude overwrote a lot of the papers with synthetic numbers due to a bug, it will probably take a few days to rerun with the actual data. The HI paper should be corrected already though.
Regarding picking the right bump, the pipeline avoids Gaussian fitting and just takes the highest velocity where the signal is still clearly above the noise floor (fixed 2 K cutoff). Less clever than fitting, but avoids the solver wandering off and the noise bumps disappear. Claude picked that heuristic when building the module, and mentioned that it reads a few km/s high compared to proper spectral fitting (McClure-Griffiths & Dickey 2016, apparently) but it’s always high in the same direction.
I’m hoping to do a write up as soon as my Discovery Dish arrives and it sees first HI light.
The author’s raw data doesn’t appear to be published, but you can reproduce this exact measurement from public survey data: the LAB HI survey (Kalberla et al. 2005) gives calibrated 21 cm spectra for every galactic longitude, and extracting terminal velocities via the tangent-point method gives you the flat rotation curve with real error bars. I maintain an open, tested Python pipeline that does this end-to-end (github.com/joebarbere/jansky-research see the the hi module; MIT, pure NumPy/astropy) alongside a teaching course that derives the method (github.com/joebarbere/jansky). And yes, according to Claude, you can double the data points from the fourth quadrant using the largest blueshift; the method is symmetric, it’s standard practice in the professional surveys.