Catalog Versions and Scientific Updates

A catalog statement is supported by an immutable evidence state: catalog release, search and pipeline revision, event-product versions, calibration and data-quality inputs, selection products, and a date through which later evidence was considered. A mutable label such as “the latest catalog” cannot support a reproducible event count or parameter claim.

Helpful background. From Search Triggers to Catalog Events defines the records that a release freezes; Selection Functions from Injections explains why the matching injection and threshold products are part of the same evidence state.

The identity of a catalog claim

Represent a claim as

$$Q =Q(C_v,S_v,P_v,I_v,K_v;t_*),$$

where

• $C_v$ is the catalog release and minor revision;
• $S_v$ is the search, background, and association configuration;
• $P_v$ is the event-level parameter-product set;
• $I_v$ is the injection and selection-function set;
• $K_v$ is the calibration, strain, and data-quality input set; and
• $t_*$ is the evidence cutoff.

Changing any component can change the supported claim. For example, a search reanalysis can add a candidate without changing old strain; a calibration update can change source posteriors without changing search membership; a corrected injection set can change population conclusions without changing either.

The cutoff $t_*$ answers a different question from release date. It records how late a correction, withdrawal, or independent reanalysis was considered. Two pages using the same catalog release but different cutoffs can legitimately have different evidence assessments.

Immutable snapshots and corrections

A durable release should be cited by its official title, version, DOI or stable landing page, and product identifiers. Retrieved files should also carry machine-verifiable digests. If $B$ is a downloaded byte sequence, a cryptographic hash

$$H=\operatorname{SHA256}(B)$$

checks that a later rerun used the same bytes. A hash verifies identity, not scientific correctness.

Corrections should create an explicit new state or correction notice. They should not erase the fact that an earlier paper used the previous state. A comparison can then state:

Result $Q$ used release $v$ through cutoff $t_*$. Release $v'$ revised the relevant field; the earlier value remains part of the historical record but is not the current result.

This avoids two opposite errors: silently updating an old claim, and continuing to present a superseded claim as current.

What must be versioned

Layer	Versioned items	Consequence of drift
Detector data	Strain files, sample rate, calibration, quality flags, GPS interval	Different likelihood input or valid analyzed time
Search	Pipeline/code, bank, rank, background, vetoes, threshold	Different triggers, FARs, and inclusion
Association	Coincidence window, source class, de-duplication and tie-breaks	Different event records
Event inference	Waveform, frequency band, priors, PSD, calibration nuisance, sampler	Different posterior and evidence products
Selection	Injection table, proposal density, recovery pipeline, threshold	Different efficiency and population normalization
Publication	Catalog table, release notes, DOI, corrections, cutoff	Different supported wording and provenance

A catalog row alone is insufficient provenance. Conversely, storing every software dependency without connecting it to the scientific field it affects creates volume without reproducibility. The record should make each consequential dependency traceable.

Dating catalog comparisons

Stable pages should derive methods without tying them to a mutable count. When a current example is necessary, use three sentences:

1. Version: name the exact official release and relevant product.
2. Cutoff: state the last date on which corrections and reanalyses were considered.
3. Scope: say which conclusion is release-dependent and which derivation is stable.

For this volume the empirical cutoff is 2026-08-18. The current baseline is the official GWTC-5.0 data release, DOI 10.7935/bk00-6a89, with search and product definitions in GWTC-5.0 Methods, §§ 2–5 and scope in GWTC-5.0 Observations, §§ 2–4 and release tables. Future pages must not silently reinterpret “GWTC-5.0” as a later catalog.

Comparing two catalog versions

For two catalog snapshots, normalize only representation—not scientific content—and compare by stable event identifier. Report separate sets:

$$E_{\mathrm{added}}, \quad E_{\mathrm{removed}}, \quad E_{\mathrm{changed}}, \quad F_{\mathrm{changed}}(e).$$

The comparison should preserve source files and hashes, parse with a pinned environment, validate schema and units, and emit a row-level diff. It must treat missing values, changed coordinate frames, renamed fields, and revised credible summaries as scientifically meaningful until shown otherwise.

Before quoting such a comparison, record the runtime and package versions, exact URLs or file identifiers, hashes, parser options, numerical precision, and counts for each difference class. Vary parser/version assumptions and stop for duplicate identifiers or schema changes that the parser cannot interpret.

The GWOSC API documentation is the current official interface reference at the cutoff. An API response remains version-dependent; archive the resolved identifiers rather than relying on an unversioned query to return the same state indefinitely.

Common pitfalls

Using retrieval date as dataset version. Retrieval time says when a file was obtained, not which scientific release it contains. Record both.

Comparing rounded summaries. Two medians may agree while posterior samples, priors, or interval conventions differ. Compare the actual product and metadata needed by the claim.

Recomputing old claims with new selection products. That creates a hybrid evidence state that belonged to no release. Either reproduce the original state or label the result as a new reanalysis.

Exercise

Catalog $C_1$ contains events A and B. Catalog $C_2$ contains A and C; A has unchanged search fields but a revised source-frame mass because the cosmology metadata changed.

1. Classify the row-level differences.
2. Which claims can remain unchanged?

Solution

The sets are

$$E_{\mathrm{added}}=\{C\}, \qquad E_{\mathrm{removed}}=\{B\}, \qquad E_{\mathrm{changed}}=\{A\}.$$

For A, source-frame mass and cosmology provenance belong in $F_{\mathrm{changed}}(A)$; its search rank and detector-frame quantities may remain unchanged if their underlying products are identical. A claim about A’s source-frame mass must cite $C_2$, while a method derivation independent of these rows can remain unchanged. A catalog count must specify whether it refers to $C_1$ or $C_2$.

References

• Gravitational Wave Open Science Center (2026). “Application Programming Interface.” Documentation verified 2026-08-18. Official documentation.
• LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration (2026). “GWTC-5.0: Methods for Identifying and Characterizing Gravitational-Wave Transients.” arXiv:2605.27224. Open PDF.
• LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration (2026). “GWTC-5.0: Observations from the Second Part of the Fourth LIGO-Virgo-KAGRA Observing Run and Updates to the Gravitational-Wave Transient Catalog.” arXiv:2605.27225. Open PDF.
• LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration (2026). GWTC-5.0 Data Release. DOI 10.7935/bk00-6a89. Official release.