Evidence, Uncertainty, and Model Comparison in Astronomy

Working definition

Astronomical knowledge is built by connecting detector records to calibrated observables and then to models with explicit likelihoods, priors, selection functions, assumptions, and uncertainties. Observation, inference, consensus, and open questions must not be collapsed into one confidence label.

What is measured

Raw counts, timestamps, detector states, and calibration exposures form the observation record.
Pipelines produce observables with statistical and systematic uncertainty.
Catalogs preserve flags, covariance, release identifiers, and provenance.

What is inferred

Physical parameters are estimated under a named model.
Population conclusions require a selection function and completeness model.
Competing explanations are compared against the same evidence.

Fact-layer dependency

The explanatory layer cannot rewrite these fields.

Fact schema →

time.utcInstantsubject.identifierscoordinates.uncertaintyprovenance.provider

What is observed

Raw counts, timestamps, detector states, and calibration exposures form the observation record. Pipelines produce observables with statistical and systematic uncertainty. Catalogs preserve flags, covariance, release identifiers, and provenance.

Calibrated measurement[1]

Scientific archives expose release documentation, quality fields, and known issues because catalog values are not context-free facts.

Boundary: Small formal error bars do not exclude systematic error.

How inference enters

Physical parameters are estimated under a named model. Population conclusions require a selection function and completeness model. Competing explanations are compared against the same evidence.

Model-dependent[1][2][3]

Population and physical claims depend on calibration, selection, uncertainty propagation, and an explicit model.

Boundary: Calibration and noise models represent the instrument.

Limits and unresolved questions

Small formal error bars do not exclude systematic error. A catalog default is not automatically the best parameter for every use. Consensus can change when new data or better models arrive.

Open question[3]

Model agreement does not identify the unique underlying mechanism when alternative models fit within present uncertainties.

Boundary: Consensus can change when new data or better models arrive.

Sources

Each source states both what it establishes and where its authority ends. Access dates record the last public verification.

[1]Gaia Archive Documentation · European Space Agency · accessed 2026-08-15
Establishes: Versioned data releases, data models, known issues, query interfaces, and access to Gaia astrometric, photometric, and spectroscopic products.
Boundary: Catalog use requires the named release, table, quality fields, covariance information, and documented known issues.
[2]NASA Exoplanet Archive Overview and Holdings · NASA Exoplanet Science Institute · accessed 2026-08-15
Establishes: Vetted exoplanet and host-star data, literature provenance, light curves, spectra, radial velocities, and distinct confirmed, candidate, and false-positive dispositions.
Boundary: Archive inclusion and default parameter selection do not eliminate publication disagreement, correlated errors, completeness limits, or later reclassification.
[3]Planck and the Cosmic Microwave Background · European Space Agency · accessed 2026-08-15
Establishes: The CMB as relic radiation, its near-blackbody spectrum, anisotropies, foregrounds, and use in constraining cosmological models.
Boundary: Cosmological parameters are conditional on a model, likelihood, foreground treatment, calibration, and combination of data sets.