Data

§1Counts

Records

2,041 (1,636 simulation, 405 observational or empirical-model), 940 rankable

Points

71,727 plotted points in the points table

Simulations

22, listed with method, code and set-up in Simulations and observations

Observations

117 sources, profiled in the same page

Relations

44 in the Explorer; 44 relation codes in the records

Citations

185 distinct papers and data releases

Redshift

z = 0 to 17 for simulations

Audit

The audit script checks every record (epoch order, provenance, plausible ranges, band consistency, duplicate data, cross-source agreement) and currently reports 0 errors and 0 warnings; accepted exceptions are listed with a reason each. That checks consistency, not that every curve is the right one.

§2Records by relation

Simulations and observations count distinct sources; records count curves. Rankable: definitions and sampling allow the curve to be used for ranking. z-range is the range of epochs in the records. 0 of the relation codes are not shown in the Explorer (few curves each).

§3Evidence tiers

Every record carries one tier. A tier says how the values entered the dataset, not how good they are; how the routes work is in Methods, how values enter.

§4Ensembles

Ensembles are not curves of one simulation and are not in the exports. They live in data/ensembles/.

CAMELS SFRD

camels-sfrd.json: the cosmic star formation rate density of 2,054 CAMELS simulations, 1,027 per code for IllustrisTNG and SIMBA (1,000 Latin-hypercube and 27 cosmic-variance), 25 Mpc/h boxes, 70 redshift bins. Reduced from the public star formation logs, each file hashed. Used by the cosmic star formation study.

CAMELS closure

camels-closure.json: the stellar-mass closure check of the pipeline.

CAMELS galaxies

Binned galaxy-population summaries of the same simulations at z = 0, 0.95, 2.15, 3 and 4 (stellar mass function, quenched fraction, specific SFR, black hole mass and occupation, metallicities, sizes, gas fractions, stellar-to-halo mass ratio, halo mass function, densities). They are read from the public Subfind catalogues by range requests. camels-galaxies.json holds the SHA-256 of the arrays read from every file. Used by the all-at-once study and the dials page.

§5Downloads

Checksums are also in sim-highline-manifest.json. In the Explorer, Get data downloads exactly what is on screen, or all epochs, as the same tidy tables.

For machines. Croissant metadata (mlcommons) describes the three CSV files, their checksums and typed columns. index.json lists every export file with its size and checksum and every relation with its record count. The same facts are on this page as schema.org Dataset metadata. Release notes come as an Atom feed, and there are a sitemap and a robots.txt. All of these are generated from the saved files when the site is built, and none of them makes a request when a page loads.

The files hold the published values. No IMF, aperture or cosmology conversion is applied to a simulation curve, and the optional IMF switch in the Explorer never reaches a download. What was re-written (axis units, one h-unit conversion) and what the observation compilation had already converted is in Methods §4.

§6Python loader

import sim_highline
df  = sim_highline.load()                      # run from a folder containing data/export, or pass a URL
sel = sim_highline.select(df, relation="gsmf", z=(5.5, 6.5), kind="simulation")
g   = sel.groupby(["source", "run"]).agg(z=("z", "first"), points=("x", "size"),
                                         mass_class=("mass_class", "first"), tier=("tier", "first"))
print(g.reset_index().assign(run=lambda d: d.run.str.slice(0, 26)).to_string(index=False))
print(f"{sel.record_id.nunique()} records from {sel.source.nunique()} simulations")

Output, dataset v0.24.0

     source                        run   z  points    mass_class             tier
    COLIBRE                  m5 (L025) 6.0      10   unspecified   official-table
    COLIBRE                  m5 (L050) 6.0      12   unspecified   official-table
    COLIBRE                  m5 (L100) 6.0      18   unspecified   official-table
    COLIBRE                  m6 (L025) 6.0       5   unspecified   official-table
    COLIBRE                  m6 (L200) 6.0      13   unspecified   official-table
    COLIBRE                  m7 (L400) 6.0       9   unspecified   official-table
    COLIBRE       m7 hybrid AGN (L200) 6.0      10   unspecified   official-table
      EAGLE              Ref-L100N1504 6.0       8      aperture digitized-figure
     FIRE-2 FIRE-2 high-z zoom sample  6.0      10         other  published-table
     FLARES 40 resimulated regions (EA 6.0      15      aperture  catalog-derived
Horizon-AGN                Horizon-AGN 6.0       3   unspecified digitized-figure
Horizon-AGN              Horizon-noAGN 6.0       3   unspecified digitized-figure
  Illustris Illustris-1 (full SUBFIND  6.0      12 total-subhalo digitized-figure
  Illustris Illustris-1 (mass within 2 6.0      11      aperture digitized-figure
      SIMBA   m100n1024 (all galaxies) 5.9      10 total-subhalo digitized-figure
      SIMBA m100n1024 (central quenche 5.9       5 total-subhalo digitized-figure
      SIMBA m100n1024 (central star-fo 5.9      11 total-subhalo digitized-figure
   THESAN-1                   THESAN-1 6.0      18      aperture   official-table
18 records from 8 simulations

one = sim_highline.nearest_epoch(sim_highline.select(df, relation="sfms", kind="simulation", rankable=True), z=2.0)
x, y, lo, hi = sim_highline.curve(df, one.record_id.iloc[0])   # nearest published epoch; nothing is interpolated
print(sim_highline.bibtex(one))

§7Using the data

• Predictions at a redshift. Select a relation and an epoch. 16 simulations provide 405 curves at z ≥ 6 across relations. Read each curve’s mass definition before reading a spread as physics; for the UV luminosity function (7 simulations) and the cosmic SFR density (15), match the dust treatment and the IMF first.
• Check your own simulation. In the Explorer, Your run draws a CSV on the same axes, in the browser only, and marks which published curves share its declared definitions. In Python, sim_highline.compare_run gives the same verdicts and median offsets. It flags mismatches and does not score fits.
• Calibration targets and training sets. rankable=True keeps only curves with documented definitions and sampling. Units are log₁₀, physical and h-free, with every conversion recorded. Observational references are thinner than simulation curves and uneven across relations: check §2 first.
• Figures and papers. The Explorer exports SVG or PNG with a citation footer, a methods paragraph and BibTeX for the curves on screen. Cite the original papers listed on each curve together with the dataset.
• Gaps. The What exists view shows simulation by relation by redshift. It shows the gaps in this data, not necessarily gaps in the field.
• Use the originals instead if you need exact values, full distributions or a definition the papers did not plot: use the collaboration’s own data release or catalogue; each record’s citation leads there. Digitized curves are as precise as the plotted line.

§8Schema and licences

Records follow curve-schema.json. Units are physical, h-free and log₁₀ as published; each axis states its definition. The records table has these columns:

record_id, kind, source, run, relation, z, z_min, z_max, representation, interval_kind, connect, x_unit, y_unit, x_definition, y_definition, tier, terms, rankable, calibration, epoch_mode, population_note, mass_class, cosmology_h0, cosmology_om, imf, mass_definition, halo_mass_definition, population, sfr_timescale_myr, sfr_statistic, quenching_criterion, size_definition, metallicity_quantity, gas_phase, density_frame, citation, doi, url, checksum_sha256, attribution_kind, primary_source, retrieved, definitions_json.

If you are an author or data owner and want a record corrected, attributed differently or removed, open an issue on the repository.

§9Used in

Nobody has told us yet. If you used the data or a page in a paper, a talk, a course or code, tell us in two lines and, with your agreement, it is listed here. Until then this list is empty, and the project's usefulness is unproven.

§10How to cite

Cite the dataset and the original papers: every record carries its citation and doi, and sim-highline-citations.bib holds BibTeX for all of them. Machine-readable metadata: CITATION.cff; DOI 10.5281/zenodo.23217259.

In text

Bhatia, S. (2026). sim-highline: curated galaxy scaling relations from cosmological simulations (v0.24.0). Zenodo. https://doi.org/10.5281/zenodo.23217259. Please also cite the original paper of every curve you use.

Copy the citation Copy the BibTeX

@dataset{bhatia2026simhighline,
  author    = {Bhatia, Shubhan},
  title     = {sim-highline: curated galaxy scaling relations from cosmological simulations},
  year      = {2026},
  version   = {0.24.0},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23217259},
  url       = {https://doi.org/10.5281/zenodo.23217259}
}