Skip to main content
Alison Thomas

Astronomy projects

Choose a star to open a project, or use the Projects list below. On a touch screen, tap once for its name and again to select it. Press Escape to close a selected project. This galaxy is an illustration, not observational data.

Choose a star, or browse the projects below
Choose from four projects

Projects

Ghostlight

A simulation of gravitational lensing tests how light from a second star affects measurements when the two stars cannot be seen separately.

Read the full case study
The offset-aware model leaves effectively zero deterministic mean residuals. The co-located-blend model leaves a strong time-dependent pattern. This is an analytic synthetic model check, not observed residual data.
Figure caption & source

Later fixed-design audit of the primary synthetic case. The offset-aware model represents the declared true mean; forcing the blend onto the source leaves a strong residual pattern. Curves are deterministic omitted-mean residuals divided by the assumed noise, not noisy observations. Expected reduced Q is 1.000 versus 18.035. Source: Ghostlight’s deterministic residual audit.

Echo Cartographer

Two studies of simulated light echoes test whether a measured delay is reliable and whether the echo is present at all.

Read the full case study
Synthetic gate admissions: disconnected bands 0 of 128, nominal genuine echo 120 of 128, sparse genuine echo 0 of 128, wider response 2 of 64, slower driver 0 of 64.
Figure caption & source

Response model plus frozen evidence/adequacy gate on held-out synthetic SV1 cohorts. Campaigns with interpretable lags: null 0/128; nominal echo 120/128; sparse echo 0/128; wider response 2/64; slower driver 0/64. Whiskers are 95% Wilson intervals conditional on this calibration. Different rows are different cohorts. A new rendering of saved results, not a new experiment.

Anatomy of a Little Red Dot

Three public JWST spectra are used to examine how faint blue light and overlapping emission affect an apparent jump in brightness.

Read the full case study
Two panels examine The Cliff. The measured spectrum rises strongly toward red wavelengths; a bending, continuous continuum plus emission follows it. Four blue-band estimates from the same exposure set overlap, and the box-sum uncertainty crosses zero, illustrating why a large red-to-blue ratio can have poorly constrained size.
Figure caption & source

The Cliff illustrates two different limits. Left: actual spectral measurements and the original bands, with an illustrative continuum that stays continuous at 3646 Å plus blended emission. Right: the faint blue-band brightness under different weightings or reductions of the same exposures. The box extraction includes zero within its formal error, leaving its positive ratio without a finite upper 95% bound. Error prescriptions differ between the panels and are identified in the figure. These are correlated alternatives and conditional models, not independent observations or a physical decomposition. NASA/ESA/CSA JWST; RUBIES GO-4233 (Anna de Graaff, Gabriel Brammer); DAWN JWST Archive / msaexp.

Gaia Pleiades Member Finder

Gaia data are used to select stars with Pleiades-like motion and distance and examine how the list changes with the selection rule.

Read the full case study
Two scatter plots show 866 selected Gaia stars in teal against the gray filtered field. Their proper motions form a compact group near 20 and minus 45 mas/year. The same candidates trace a curved color–magnitude sequence.
Figure caption & source

Left: two of the three clustering features; parallax is the third. Right: color and magnitude were withheld from the fit, but the query required usable photometry and absolute G reuses parallax. No extinction or parallax-zero-point correction is applied. The field contains 12,520 quality-filtered Gaia DR3 sources, of which 866 pass the candidate rule. Data: ESA/Gaia/DPAC, Gaia DR3. Comparison: Hunt & Reffert (2024), CDS VizieR J/A+A/686/A42.