About · release v2026.09.25
About varchive
Light curves from six surveys for 10,095,969 of the 10,304,679 stars in AAVSO's VSX catalog, each processed the same way: features, periodograms, an adopted period and a classification. Free to read in bulk, no account needed.
- VSX stars
- 10,304,679
- measurements
- 12.16 billion
- surveys
- 6
- years of data
- 29 · 1997–2026
- periodic
- 4.50 M
What's in it
Each survey's light curves keep their own bands and magnitude system.
Each bar places the survey's years on one ruler, 1997 to 2026; OGLE's bar shows its phases II, III and IV.
Stars by the number of surveys with a light curve
10,095,969 stars have at least one light curve, 5,966,107 from two or more surveys.
Times are in BJDTDB; every survey but Gaia also keeps its native time column.
How it's built
The same four steps for every star and every survey.
- 01
Clean
Each survey's own rules: quality flags, placeholder values, duplicates, camera zero points.
per survey, per band - 02
Search
A Lomb–Scargle periodogram per band, a joint multiband fit for Gaia and ZTF, a box search for eclipses.
GLS · multiband GLS · BLS - 03
Adopt
One period per star across surveys. The most confident periodic survey wins; agreement raises the confidence, disagreement is flagged.
VSX compared, never copied - 04
Classify
A class, a probability and a set of plausible classes for almost every star with a light curve; it abstains when the evidence is ambiguous.
labels weighted by sourceModel card →
Periodic, and how often by chance
4,502,369 stars are periodic. Shuffling each light curve's times keeps its cadence and noise but destroys any period; about 2% still pass.
Checked against OGLE
In v2026.09.20, adopted periods matched OGLE's own for 97.7% of shared stars, within each light curve's frequency resolution. Most of these stars' periods were measured on OGLE's own photometry, so this checks the period search rather than the data.
Release notes
What's new
v2026.09.25
- Pulsating stars at their fundamental period: 11,798 periods of long-period variables, RR Lyrae, δ Scuti stars and Cepheids moved. 2,369 Miras that sat at twice or half their VSX period now match it; OGLE's own periods side with VSX for 1,970 of 1,972 such Miras. R And moved from 203.6 to 407.2 days (VSX: 410).
- For these classes a period at twice or half VSX's no longer counts as agreeing with VSX: 2,869 periods that agreed only that way were withdrawn, and 6,494 more stars are flagged as disagreeing with VSX.
- The change also moved 1,571 stars from VSX's period, or a 2:1 harmonic of it, to one under a tenth or over ten times it (1,522 long-period variables, 164 of them Miras); all are flagged as disagreeing with VSX.
- Periods and their flags were re-released; classifications and light curves are unchanged from v2026.09.20.
v2026.09.24
- Eclipsing binaries at their full orbital period: v2026.09.20 had about 895,000 of them (41% of those with a VSX period) at half their orbit. Agreement with the fitted orbits of Rowan et al. (2022) rose from 50% to 62%, with Gaia DR3 orbits from 43% to 64%.
- The period is now doubled when the two eclipses differ in depth or the doubled period matches VSX. Detached binaries whose eclipses look alike stay at half the orbit: photometry alone cannot tell them apart.
- Six value-added catalogs matched to the archive's stars: detached eclipsing-binary fits (Rowan et al. 2022), Gaia DR3 and SB9 binary orbits, SDSS-V ASTRA, LAMOST DR11 and GALAH DR4 with per-visit radial velocities. A star's page shows its values against each catalog's population.
- Star pages load faster. A star is found by its name in any letter case or by a SIMBAD identifier, and every survey's light curve downloads as one file.
- Periods and their flags were re-released; classifications and light curves are unchanged from v2026.09.20.
v2026.09.20 · first multi-survey release
- The full VSX list and six surveys (previously ASAS-SN only, about 3 million stars).
- A varchive ID for every star; ASAS-SN IDs, VSX names and old links still work.
- Three public collections: per-star values, frozen light curves and live light curves (Data & API).
- Periods adopted across surveys, flagged where surveys agree or disagree.
- A new classifier for almost every star with a light curve; it abstains when the evidence is ambiguous.
Known limits
- Detectability (whether a star's variability should be visible at its brightness) is computed for ASAS-SN only.
- Classifications are from clf-v2026.09.20, scored on v2026.09.20's periods. Stars whose period changed in v2026.09.24 and v2026.09.25 (about 712,000 eclipsing binaries and 11,798 pulsating stars) have not been re-scored.
- About 8% of eclipsing binaries with a VSX period still sit at half of it, mostly detached pairs whose two eclipses look alike; photometry cannot tell them apart. A star's
vsx_period_ratio(adopted over VSX) reads 0.5 for these. - Some adopted periods are a multiple or a fraction of the true one: Gaia-only periods, which rest on a few dozen transits, and some semiregular variables at twice or half their VSX period. The star page flags a period that disagrees with VSX or with the class's typical range.
- A period a survey finds on its own outranks one that matches VSX, even when the first is implausible, so some long-period variables show a period of a day or less. The star page flags these as disagreeing with VSX.
- About 78% of VSX records are Gaia DR3 imports, so Gaia's scanning pattern shows as arcs on the sky map.
- Gaia is DR3; DR4 will follow its release.
Citation and licenses
Cite the archive and every survey whose data you use. Star pages and exports list the relevant references. The archive's own data (its periods, classifications and catalog) are released under CC BY 4.0; survey light curves and the catalogs it draws on stay under their own licenses.
The citation table →Access
The data are on a public bucket with no accounts or tokens. The planner and Analyze run in your browser; your location and photometry are not uploaded.
Privacy
No accounts and no cookies. Your observing site, saved targets and display settings stay in this browser. One page view in ten reports how fast it loaded: the page type, the release, the site build, the timings, how the page was reached and which page element a slow timing points to, with the country Cloudflare assigns to the request. None of it identifies you.
Cloudflare, which serves the site, sees IP addresses as any web host does. varchive uses them only for per-address rate limits: a client that reaches a limit is logged under a short hash of its address, and the request logs varchive reads to find errors are kept for seven days. A few features load from other services: the finder chart from CDS (Aladin Lite and DSS images), name lookups from CDS Sesame and Analyze's runtime from jsDelivr. Feedback goes to GitHub, under GitHub's terms.