Tension in Dark Sirens

CatWISE quasar dipole systematics (reproducibility bundle) This repository supports the compact MNRAS Letter in mnras_letter/ and the accompanying reproducibility bundle archived on Zenodo (10.5281/zenodo.18719277). It contains code, small vendored reviewer-seed artifacts, and auditable report outputs under REPORTS/. A larger PRD-style figure pack workflow is also supported; see docs/OVERLEAF_PRD.md. (Short note-style manuscripts also live under radio_a/.) Quick summary (what matters) Question / claim Key quantitative summary Where to verify One-command reproduce Magnitude-limit stability diagnostic (CatWISE) D ≈ 1.6e-2 while axis drifts to ~34° by W1_max=16.6 mnras_letter/main.tex, REPORTS/Q_D_RES_2_2/master_report.md make reproduce Time-domain (epoch) diagnostic (unWISE; CatWISE parent) Dipole-only epoch amplitudes span D≈0.067–0.118, but a constrained rich bootstrap gives P(range_sim≥range_obs)=0.974 and P(χ²_sim≥χ²_obs)=0.983 REPORTS/unwise_time_domain_catwise_epoch_systematics_suite/, REPORTS/unwise_time_domain_catwise_parametric_bootstrap_constrained_constantD_rich_20260220_231118UTC/ ./.venv/bin/python scripts/run_unwise_time_domain_catwise_epoch_systematics_suite.py “Case-closed” robustness package (maximal nuisance + calibration) At W1_max=16.6: baseline D=0.01678 → maximal nuis D=0.00578 → orthog D=0.01776; CMB-fixed D_par=-0.00815; constrained-null bootstrap `p_abs=P( D_par,sim ≥ Why this work should not be dismissed Controlled falsifiers: time-domain stability and faint-limit axis drift are diagnostics that a kinematic/physical interpretation must be stable under, absent selection/estimation biases. Calibrated inference: key variability statements are calibrated with constrained parametric bootstraps and injection-through-measured-systematics tests, not just analytic χ². Multiple estimators + templates: results are checked with regression, Poisson GLMs, and cross-check estimators, reducing “it’s just the method” failure modes. Held-out validation: where applicable, nuisance fields are trained on one sky subset and evaluated on another (wedge GroupKFold), to reduce overfit arguments. Internal self-audits: the maximal-nuisance suite includes a scan-vs-single-cut consistency check (scan_rep_consistency.json) so optimizer artefacts cannot silently change headline numbers. Auditable artifacts: each suite writes a self-contained report directory with JSON summaries + plots + exact command lines for independent verification. Quickstart 1) Verify headline values (reviewer seed; no external downloads) This lightweight command reproduces the headline numbers quoted in mnras_letter/main.tex using vendored artifacts under artifacts/. It does not require the Secrest+22 tarball or other large external downloads. make reproduce This writes a timestamped folder under outputs/ containing report.md and summary.json. 2) Run the case-closed suite (maximal nuisance + calibration) ./.venv/bin/python scripts/run_case_closed_maximal_nuisance_suite.py \ --out-report-dir REPORTS/case_closed_maximal_nuisance_suite_$(date -u +%Y%m%d_%H%M%SUTC) \ --bootstrap-nsim 1000 3) Run the epoch-slicing suite (unWISE time domain; CatWISE parent) ./.venv/bin/python scripts/run_unwise_time_domain_catwise_epoch_systematics_suite.py Scope and limitations (read first) This repo is designed to support the following (core, referee-facing) claims: The CatWISE “accepted” number-count dipole amplitude is non-zero at the few×10⁻² level. The inferred dipole axis/direction is not stable under plausible magnitude-limit and depth/completeness modeling choices. Selection/completeness systematics can generate smooth magnitude-limit scans and bias both amplitude and direction in injection tests. unWISE epoch slicing provides a controlled time-domain diagnostic showing that dipole-only recovery is systematics-dominated. This repo does not currently claim or provide: A fully closed, image-level injection/recovery end-to-end completeness pipeline that “solves” the all-sky amplitude. A validated, all-sky W1-conditioned completeness model (the all-sky external proxy uses Gaia qsocand, which has no W1). How to cite (copy/paste) If you use this repository, please cite: The MNRAS Letter in mnras_letter/ (source: mnras_letter/main.tex). The Zenodo reproducibility archive: 10.5281/zenodo.18719277. Supplementary assets archive (where applicable): 10.5281/zenodo.18489200. Data requirements (minimal vs full) Minimal (reviewer seed): none (uses vendored artifacts under artifacts/). Full end-to-end reruns: require the Secrest+22 accepted CatWISE AGN bundle (10.5281/zenodo.6784602) and additional optional external data (details: docs/ADVANCED.md). Key report bundles (source of truth) These directories are the “source of truth” for the numbers quoted in the paper and table above: REPORTS/Q_D_RES_2_2/ (magnitude-limit scan + core diagnostics) REPORTS/unwise_time_domain_catwise_epoch_systematics_suite/ (epoch slicing + nuisance controls) REPORTS/unwise_time_domain_catwise_parametric_bootstrap_constrained_constantD_rich_20260220_231118UTC/ (epoch constrained null calibration) REPORTS/case_closed_maximal_nuisance_suite/ (maximal nuisance + CMB-fixed + held-out + bootstrap) For additional bundles and “how to run everything”, see docs/ADVANCED.md. References / DOIs used by this repository Quasar-dipole reproducibility archive (Zenodo), DOI: 10.5281/zenodo.18719277 Supplementary assets archive (Zenodo), DOI: 10.5281/zenodo.18489200 Legacy repository archive (Zenodo), DOI: 10.5281/zenodo.18476711 Secrest et al. 2022, ApJL 937 L31, DOI: 10.3847/2041-8213/ac88c0 Secrest+22 accepted CatWISE AGN catalog (Zenodo record), DOI: 10.5281/zenodo.6784602 CatWISE2020 (Marocco et al. 2021, ApJS 253, 8), DOI: 10.3847/1538-4365/abd805 unWISE Time-Domain Catalog (IRSA580), DOI: 10.26131/IRSA580 Environment setup Recommended (Conda, easiest for healpy): conda env create -f environment.yml conda activate quasars-systematics Alternative (pip/venv): python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt Advanced usage See docs/ADVANCED.md.