API reference¶

Leaf modules are listed explicitly rather than recursively: several names are re-exported from their package __init__, and a recursive sweep documents those objects twice, which Sphinx reports as a duplicate description.

`specsrbench.paths`	Where the inputs and outputs live, and how they are found.
`specsrbench.methods`	The eight methods and the reference: display names and the keys they are cached under.
`specsrbench.metrics`	Reconstruction metrics, and the four guards that keep tuning honest.
`specsrbench.classical`	Classical deconvolution methods, parameterised for the log constant-R grid.
`specsrbench.data`	Loading the tuned cache: one object, loaded once, shared by every figure.
`specsrbench.sample`	The small tutorial dataset: 24 held-out spectra, fetched from the Hub.
`specsrbench.style`	The one matplotlib style every figure in the paper uses.
`specsrbench.cli`	`specsrbench` -- one entry point for every product in this repository.
`specsrbench.figures`	One module per paper figure, and the registry the CLI dispatches through.
`specsrbench.figures.fig1_toy_methods`	Figure 1 -- the eight methods on a 1D toy, where the truth is known exactly.
`specsrbench.figures.fig2_qualitative`	Figure 2 -- one held-out galaxy, every method, with an [O III] inset.
`specsrbench.figures.fig3_residual_maps`	Figure 3 -- residual maps: every held-out spectrum, sorted by redshift.
`specsrbench.figures.fig4_mae_summary`	Figure 4 -- global reconstruction fidelity, five ways.
`specsrbench.figures.fig5_per_line_snr`	Figure 5 -- per-line behaviour: S/N, line shape and position, detection.
`specsrbench.figures.fig6_redshift_mae`	Figure 6 -- reconstruction error against redshift, and why it has that shape.
`specsrbench.build`	Rebuilding the cache the figures read, from JADES DR4 and the Hub.
`specsrbench.build.predictions`	Stage 1 -- run the specsr chain over the held-out split, from Hub weights.
`specsrbench.build.sets`	Stage 2 -- cut the predictions into the three galaxy-disjoint sets.
`specsrbench.build.lsf`	Derive the prism->grating line-spread function from the instrument itself.
`specsrbench.build.tune`	Retune the classical deconvolution hyperparameters against all four guards.
`specsrbench.build.classical_cache`	Rebuild the classical caches on the log constant-R grid with retuned parameters.
`specsrbench.build.lines`	Assemble a drop-in cache on the log constant-R grid: arrays, line fits, S/N.