We ask whether the timing of worldwide curiosity, measured by Google Trends, lines up with the positions of celestial bodies. For each of 1034 search fields - a curated atlas of skill terms (each derived from an ISCO-08 occupation, e.g. accountant -> accounting) plus learning topics - we fit 22 years of MONTHLY search interest with a strict 11-parameter model: eleven sidereal bodies, each contributing a parameter-free SINC feature sinc(delta) of its angular distance delta to a shared phase - a sinusoid oscillating at that body's own EXACT orbital period, measured day-by-day from the ephemeris - with ONE non-negative weight per body and NO intercept. Two things are chosen per keyword by grid search (not fitted coefficients): the shared phase angle (swept 5 degrees, giving the sign) and a resonance PERIOD - the kernel's inverse-width, reported in years - selected to maximise F1, the harmonic mean of the fit R2 and the sign decisiveness (rep). The phase is swept around the zodiac on a 5-degree grid to place the sign. The eleven bodies are the Sun, Mercury, Venus, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto, Chiron and the lunar node (the Moon is not included). We fit the cube root of interest - a variance-stabilising transform for the right-skewed Trends signal - by NON-NEGATIVE, robust (Huber) least squares (each body only adds to interest); the angle of minimum error assigns each field a zodiac sign, in-sample R2 measures fit, and a variance decomposition attributes it across bodies. The median in-sample R2 is carried mostly by the SLOW outer bodies acting as a low-frequency secular-trend basis (over 22 years Neptune and Pluto barely complete an arc) plus the Sun encoding annual seasonality - not by the fast bodies. Dropping the slow bodies collapses the median R2, so the high R2 is mostly trend fitting, not celestial seasonality; the sidereal alignment is best read as a compact re-description of each field's trend and annual seasonality. We present the full atlas, an open dataset, and a one-click reproduction. This is a transparent statistical curiosity: correlation is not causation, and nothing here claims a physical mechanism.
Aggregate search interest is a noisy but rich record of collective attention. We test a deliberately unusual hypothesis - that a search field's multi-decade rhythm aligns with one angle of the sidereal zodiac - purely as an exercise in transparent curve fitting and open, reproducible methodology. We hold ourselves to two self-imposed disciplines: the model carries exactly eleven numbers, and its kernel is parameter-free (a sinc at the fields tuned resonance period), so the only fitted quantities are eleven non-negative weights - one per body, with no intercept - while the shared phase (5-degree grid, the sign) and a per-keyword resonance period (chosen by max F1 of fit and decisiveness) are grid-searched selectors.
The atlas is 1034 search fields: a curated set of SKILL terms - each derived from one of the 436 ISCO-08 occupations by mapping the occupation to its most natural Google-search field (accountant -> accounting, electrician -> electrical wiring) - together with a set of learning topics. For each field we fetch its all-time MONTHLY Google-Trends series (2004 to the present, about 270 points) as a single direct export and reindex it to a fixed monthly grid - one request per field, with no finer-grained windows to splice together and rescale. The most recent twelve months are dropped before fitting (see the companion recency study), so R2 is reported in-sample on the remaining series. We model the CUBE ROOT of interest: Trends interest is strongly right-skewed - long quiet stretches punctuated by short spikes - so on the raw scale a handful of unfittable spikes dominate the residual. The cube root stabilises the variance; being monotone per series it leaves the assigned sign essentially unchanged.
Each of eleven bodies - the Sun, Mercury, Venus, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto, Chiron and the lunar node - sweeps the zodiac at its own physical speed. At a trial angle phi each body contributes a SINC of its wrapped angular distance to phi at the field's inverse-width w = 1/PERIOD: kern(d) = sinc(w*deg2rad(d)); the period is tuned per keyword (larger w = shorter period, first zero at 57.3/w deg). each body sweeps at its own EXACT orbital period - measured day-by-day from the sidereal ephemeris (Sun ~1 yr, Jupiter ~12 yr, Neptune ~165 yr). There is no width or frequency parameter, no FFT, no tuning. Fast bodies wave many times over 22 years; the slow bodies barely complete an arc, so they contribute a near-constant secular ramp rather than a wave. We fit by NON-NEGATIVE, robust Huber least squares, so every body's weight is >= 0 and each body only ADDS to interest - the weights never blow up and cancel, which makes the per-body variance decomposition trustworthy; Huber loss downweights the interest spikes that would otherwise dominate squared error. The only fitted angle is the one shared phase, swept around the zodiac on a 5-degree grid; the angle of lowest error is the field's sign, identically the angle of highest R2. The model is 11 numbers: one non-negative weight per body, with no intercept; the shared phase and the per-keyword resonance period are grid-searched selectors, not fitted coefficients.
Across 1034 fields the median in-sample R2 is 0.65. A per-body variance decomposition (flatten each body to its mean, recompute R2, take the drop; non-negative, summing to R2) shows the fit is carried mostly by the slow outer bodies - which, barely completing an arc over 22 years, act as a low-frequency secular-trend basis - plus the Sun encoding annual seasonality; the fast bodies carry little. We rank fields by representativeness, R2 times the share of fit-quality area concentrated within the assigned sign's 30 degrees rather than spread around the zodiac. Under the sinc kernel the fit-quality is spread widely around the zodiac, so sign assignments are soft rather than decisive; 15 fields exceed a runner-up ratio of 1.5. Figure 1 shows a representative fit; Figure 2 the distribution of best-fit signs.
The high R2 is mostly trend fitting, not celestial seasonality. The decisive test is to drop the slow outer bodies and refit: the median R2 collapses sharply. In other words, almost all the explained variance comes from the slow bodies' near-constant secular ramp and the Sun's annual term - exactly a trend-plus-annual-seasonality basis - and very little from any genuinely celestial sub-annual structure. The sidereal alignment is therefore best read as a compact RE-DESCRIPTION of each field's trend and annual seasonality, not as evidence of a celestial influence: a plain trend-plus-annual baseline with comparable degrees of freedom and NO astrology explains nearly as much. The model is descriptive, not predictive: R2 is in-sample, and lining up two rhythms over 22 years is not causation. We publish the data and code so readers can verify every claim.
All data underlying this study are openly available at https://artaquest.org/wp-content/uploads/research/ - the full-resolution monthly series, the sidereal ephemeris, and the result tables (atlas and recency curves). These files are the exact inputs to the analysis.
The complete analysis code is open in the ArtaQuest repository (the analysis/ directory). A one-click Google Colab notebook reproduces every figure and number from the hosted data: https://colab.research.google.com/gist/artaquest/6d2a073d195d3c075ac6d93d3c6f899d/seasonality.ipynb.
A.A. designed the study, performed the analysis, and wrote the manuscript.
The author declares no competing financial or non-financial interests.
This work received no external funding and was conducted under the ArtaQuest Foundation.