Begin with what the line model resolves
Read SOAP 2.0, arXiv 1409.3594v1, Sections 2.1–2.3 and Figure 1. Identify the distinction between removing flux and inhibiting convective blueshift. The authors use local-profile information richer than our equal Gaussian notches. Our live model reproduces a weighted first-moment mechanism; it does not inherit SOAP’s full bisector or cross-correlation fidelity.
Then read Haywood and colleagues, 1407.1044v1, Sections 2–3 and Table A1. Their contemporaneous photometry helps inform stellar variability. Our public replay retains the 2012 numerical table, but its RV–calcium process is a different, deliberately bounded analysis. Agreement with a published amplitude alone would not reproduce that paper’s photometric conditioning or complete inference.
Reading: Dumusque, Boisse & Santos — SOAP 2.0, arXiv:1409.3594v1 · Haywood et al. — CoRoT-7, arXiv:1407.1044v1
Follow the covariance into the result
In Rajpaul and colleagues, 1506.07304v1, follow Sections 3.1–3.4 from a shared latent process to multiple observables. Work through Appendix A.2 before looking at the tests in Sections 4.1–4.2 and Figures 3–4. Ask which observable is coupled to the process itself and which to its derivative. That question fixes the cross-covariance signs.
Separate a model’s own synthetic test from a test generated with a richer physical simulator. The former checks whether inference can recover its assumptions; the latter probes some consequences of misspecification. Neither alone demonstrates that all real stellar variability is captured. Our Gaussian and multimodal importance tests check integration, while the physical spot-only injection checks a different failure mode.
Reading: Rajpaul et al. — joint activity processes, arXiv:1506.07304v1 · Rasmussen & Williams — Gaussian Processes for Machine Learning, Chapter 2
Let the observing window enter the argument
Read “Ghost in the time series,” 1510.05598v1, Sections 2–3 and Figures 1–3. The important comparison keeps the sampling while changing the synthetic signal content. A selected periodic residual can survive a nuisance-removal procedure because of that procedure and the window. Our tutorial does not rerun the Alpha Centauri B data; its own injections demonstrate a related diagnostic on the CoRoT-7 dates.
Before treating a peak as evidence, specify the searched frequencies, nuisance model and selection rule. A conditional interval at the strongest peak does not account for the search. When an activity process absorbs an injected planet, inspect recovery across periods and priors. This tests how reliably the competing contributions can be distinguished.
Reading: Rajpaul, Aigrain & Roberts — Ghost in the time series, arXiv:1510.05598v1 · Rajpaul et al. — joint activity processes, arXiv:1506.07304v1
A modern example with several instrument states
Use Suárez Mascareño and colleagues, 2212.07332v2, especially Sections 2.1–2.2 and 4.1–4.4, Table 2, Table B.1 and Figures 9–10. The 139 observations combine CARMENES with ESPRESSO. ESPRESSO itself is separated into three states around hardware changes. Track the independent offsets and jitters before reading the orbital amplitudes.
This is published research in Astronomy & Astrophysics 670, A5 (2023); the reading link pins the corrected arXiv manuscript. The older abstract prints 20.202 days for the second planet, while the corrected version and parameter tables give 21.202. Checking a number against the tables resolves a typographical discrepancy before it becomes a physical puzzle.
Compare the raw time series, the GP components and the component-subtracted phase folds. A folded curve with other contributions removed is not the original observation. Table B.1’s priors are part of the model. The reported minimum masses near one Earth mass orbit a much lower-mass star at roughly 10.35 and 21.20 days; they are not Earth–Sun analogues with year-long periods and 0.089-metre-per-second amplitudes.
Our real tutorial contains one HARPS campaign, so it cannot empirically recover these multiple-instrument offsets. The separate artificial offset experiment is labeled synthetic. Likewise, an assumed inclination can turn minimum mass into a conditional mass, but it does not create a measured inclination.
Reading: Suárez Mascareño et al. — GJ 1002, arXiv:2212.07332v2 · GJ 1002 — published A&A 670, A5 (2023) · IAU 2015 Resolution B3 — nominal conversion constants
Write down what would change your conclusion
For each paper, record the observing span, independent channels, prior information, instrument divisions and synthetic checks. Then name one missing measurement that would most improve identifiability: longer phase coverage, an activity-sensitive line indicator with known errors, simultaneous photometry or an independent sky orbit. Choose a time and channel where the competing explanations predict different values.
The live Lab gives a surface mechanism, the frozen campaign gives an actual conditional posterior, and these papers show richer observational decisions. Keeping those three scopes explicit makes it possible to move between them without mistaking an attractive animation for a measured stellar map.
Reading: Haywood et al. — CoRoT-7, arXiv:1407.1044v1 · Suárez Mascareño et al. — GJ 1002, arXiv:2212.07332v2