Forward-modelling Milky Way Cepheids: selection effects and physical priors in the Gaia-HST calibration

Adam Riess; Dan Scolnic; Guilhem Lavaux; Harry Desmond; Richard Stiskalek

Forward-modelled Bayesian Cepheid calibration recovers SH0ES PL parameters and shows that omitting selection under a volume prior biases the zero-point by ~0.05 mag.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 23:55 UTC pith:P6QCVZXX

load-bearing objection Abstract-only: generative MW Cepheid model with selection recovers SH0ES PL at <0.5σ and flags a ~0.05 mag volume-prior bias; load-bearing risk is whether the analytic detection integral is complete. the 3 major comments →

arxiv 2603.09880 v2 pith:P6QCVZXX submitted 2026-03-10 astro-ph.GA astro-ph.CO

Forward-modelling Milky Way Cepheids: selection effects and physical priors in the Gaia-HST calibration

Richard Stiskalek , Adam Riess , Harry Desmond , Guilhem Lavaux , Dan Scolnic This is my paper

classification astro-ph.GA astro-ph.CO

keywords Cepheidsperiod-luminosity relationGaia parallaxesselection effectsBayesian forward modellingdistance ladderHubble constantGalactic disc geometry

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

High-precision Gaia parallaxes of Milky Way Cepheids now allow per-cent-level calibration of the local distance ladder and of H0. This paper builds a fully forward Bayesian model that jointly infers the period-luminosity (PL) relation, the Gaia parallax zero-point, and individual distances while folding in Galactic disc geometry and the explicit selection functions of the two HST SH0ES campaigns. An analytic detection probability that accounts for magnitude, parallax, period and extinction cuts is reduced to a tractable integral over distance and sky position; posterior predictive checks confirm that the generative model reproduces the observed distributions. The resulting PL parameters agree with the SH0ES maximum-likelihood values at the <0.5 σ level, a direct consequence of the small intrinsic scatter of the Cepheid PL relation. In contrast, adopting a uniform-in-volume prior without simultaneously modelling selection produces a ~0.05 mag zero-point bias and predictive distributions that are incompatible with the data; that bias is driven mainly by the missing selection model and would translate into an unjustified apparent shift in H0. A self-consistent Bayesian treatment of structure and selection therefore reinforces the local distance-ladder determination of H0 and the tension with early-Universe inferences.

Core claim

A fully forward Bayesian framework that simultaneously infers the Cepheid period-luminosity relation, the Gaia parallax zero-point and individual distances, while incorporating disc geometry and the two HST SH0ES selection functions, recovers PL parameters that match the SH0ES maximum-likelihood values at the <0.5 σ level. Omitting the selection model while using a uniform-in-volume prior produces a ~0.05 mag zero-point bias whose origin is mismodelling rather than new physics.

What carries the argument

An analytic detection probability that folds magnitude, parallax, period and extinction cuts into a tractable integral over distance and sky position; this generative selection term, together with a Galactic disc distance prior, is the central mechanism that keeps the PL zero-point unbiased.

Load-bearing premise

That the analytic detection probability fully captures the magnitude, parallax, period and extinction cuts of the two HST SH0ES campaigns and the Gaia sample, so residual incompleteness does not still bias the PL zero-point.

What would settle it

A controlled mock catalogue drawn from the same generative model but with deliberately altered or incomplete selection cuts that, when analysed with the paper's pipeline, yields a PL zero-point shift larger than ~0.05 mag or posterior predictive distributions that fail to match the real Gaia and HST observables.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

PL parameters obtained with the full forward model remain consistent with SH0ES at <0.5 σ, leaving the local H0 ladder essentially unchanged.
A ~0.05 mag zero-point bias appears when selection is omitted under a uniform-in-volume prior; that bias would masquerade as an H0 shift if left uncorrected.
Posterior predictive checks of parallax, magnitude and period distributions become incompatible with the data once the selection model is dropped.
Self-consistent treatment of Galactic structure and survey truncation therefore strengthens rather than weakens the local distance-ladder H0 and the tension with early-Universe determinations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same generative-selection machinery can be ported to other geometric calibrators (e.g., TRGB, masers) whose samples are magnitude- or extinction-limited, testing whether analogous zero-point biases appear.
If residual incompleteness beyond the analytic cuts is later quantified, the reported 0.05 mag bias sets a concrete scale against which any remaining systematic can be compared.
Future Gaia data releases with improved zero-point maps can be re-analysed inside the identical hierarchical model, isolating whether the PL zero-point moves more from better parallaxes or from better selection characterisation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Abstract-only: generative MW Cepheid model with selection recovers SH0ES PL at <0.5σ and flags a ~0.05 mag volume-prior bias; load-bearing risk is whether the analytic detection integral is complete.

read the letter

The one thing to know is that this is a methodological defense of the SH0ES MW Cepheid calibration: a joint forward Bayesian model with disc geometry and the two HST selection functions recovers the SH0ES PL parameters at <0.5σ, while a uniform-in-volume prior without selection produces a ~0.05 mag zero-point bias and bad posterior predictive checks. That bias is presented as mostly selection omission, not a real H0 shift.

What is new is the self-consistent generative treatment—analytic detection probability reduced to an integral over distance and sky position, plus simultaneous inference of PL, Gaia zero-point, and latent distances—checked against observed parallax, magnitude, and period distributions. Hierarchical Cepheid/Gaia work already exists; the distinctive piece is folding the actual SH0ES campaign cuts into the generative model and quantifying the volume-prior artifact. If the predictive checks hold, that is real credit: it is non-circular in design and directly answers a recent prior claim.

Soft spots are exactly what you expect from abstract-only. We cannot inspect the detection integral, extinction handling, period-dependent completeness, or the figures. The stress-test concern is fair: if the analytic selection under-states faint/high-extinction incompleteness, residual bias at the 0.05 mag level remains possible. That is the load-bearing assumption, not a proven flaw. Circularity looks modest; free parameters are the usual PL + zero-point + distances. No invented entities, no red-flag internal contradiction visible.

This is for people who care about local-ladder systematics and H0 tension bookkeeping. A serious referee should see the full methods, code, and checks. I would send it to peer review rather than desk-reject; the claim is field-relevant and the generative framing is the right way to settle the prior debate. Without the full text I would not cite it yet, and I would only bring it to reading group once the selection function is inspectable. Verdict: conditional accept-shaped if the paper delivers what the abstract promises.

Referee Report

3 major / 0 minor

Summary. The manuscript presents a fully forward-modelled Bayesian calibration of the Milky Way Cepheid period–luminosity (PL) relation using Gaia EDR3 parallaxes. The framework jointly infers the PL zero-point and slope (with intrinsic scatter), the Gaia parallax zero-point offset, and latent individual distances, while incorporating Galactic disc geometry through the distance prior and the selection functions of two HST SH0ES campaigns. An analytic detection probability that accounts for magnitude, parallax, period, and extinction cuts is reduced to a tractable integral over distance and sky position. Posterior predictive checks are reported to match the observed parallax, magnitude, and period distributions. The recovered PL parameters agree with SH0ES maximum-likelihood values at the <0.5 σ level; adopting a uniform-in-volume prior without selection is reported to induce a ~0.05 mag zero-point bias and an unjustified apparent H0 shift, attributed mostly to omission of the selection model.

Significance. If the generative selection model is correctly specified and the reported agreement and bias estimates survive full scrutiny of the likelihood, priors, and predictive checks, the work would reinforce the local distance-ladder H0 determination by showing that self-consistent Bayesian treatment of Galactic structure and survey truncation recovers SH0ES-consistent PL parameters, while previously claimed shifts can arise from mismodelling. The analytic detection integral and the explicit generative validation against observed distributions are methodologically valuable for Cepheid and related stellar calibrations. The result is of direct interest to the Hubble-tension literature.

major comments (3)

The central claim that the analytic detection probability fully encodes the joint magnitude, parallax, period, and extinction cuts of the two HST SH0ES campaigns and the Gaia sample is load-bearing for both the reported <0.5 σ SH0ES agreement and the attribution of the ~0.05 mag zero-point bias to omission of selection. Only the abstract is available for this review, so the explicit form of the detection integral, the treatment of extinction maps, and any period-dependent completeness cannot be inspected. Residual incompleteness at the faint or high-extinction end at the level of the claimed 0.05 mag effect cannot be ruled out without that derivation.
The abstract states that posterior predictive checks match the observed distributions of parallaxes, magnitudes, and periods, and that this validates the generative model. These checks are essential to the claim that residual selection does not bias the PL zero-point. The full manuscript must present them quantitatively (which statistics, residual structure as a function of magnitude/period/extinction) so that incompleteness at the scale of the reported bias can be assessed.
The comparison to Högås & Mörtsell (2026) attributes the ~0.05 mag shift “mostly” to omission of the selection model rather than to the uniform-in-volume prior alone. A controlled decomposition (selection on/off crossed with prior choice) is required to substantiate that attribution and the associated claim of an “unjustified” H0 shift; the abstract alone does not establish the relative contributions.

Circularity Check

0 steps flagged

Abstract-only review finds no exhibited circular reduction; generative Bayesian setup with PPC checks is self-contained, with only mild non-load-bearing SH0ES author overlap on selection functions.

full rationale

Only the abstract is available, so no equation-level reduction (Eq. X ≡ Eq. Y by construction, or fitted parameter renamed as prediction) can be exhibited. The claimed pipeline is a generative Bayesian model that jointly infers the PL relation, Gaia parallax zero-point, and distances under an explicit disc-geometry prior and analytic selection functions for the two HST SH0ES campaigns plus Gaia cuts; agreement with SH0ES ML values at <0.5σ is presented as a consequence of small intrinsic PL scatter, not as an input. Posterior predictive checks against observed parallax, magnitude, and period distributions are cited as external consistency tests. Critiquing the uniform-in-volume prior of Högås & Mörtseil (2026) without selection, and recovering a ~0.05 mag zero-point bias when that prior is used, is a comparative modelling result rather than a self-definitional loop. Author overlap with SH0ES (Riess) means the selection functions come from campaigns co-designed by a co-author, which is ordinary survey practice and not a load-bearing uniqueness theorem or self-citation chain that forces the PL zero-point. Residual risk that the analytic detection probability incompletely captures joint cuts is a correctness/assumption concern, not circularity under the stated rules. Score 1 reflects only that minor, non-load-bearing author overlap; no steps meet the quote-and-reduce threshold.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

Abstract-only: free parameters of the hierarchical model (PL slope/zero-point/scatter, Gaia zero-point offset, individual distances) are inferred rather than fixed by hand, but their exact prior forms and any hyperparameter choices are not given. Core domain assumptions are standard Galactic disc geometry and the stated SH0ES selection cuts. No new physical entities are introduced; the contribution is statistical modelling of selection and geometry.

free parameters (3)

period-luminosity zero-point and slope (and intrinsic scatter)
Inferred jointly in the hierarchical model; abstract does not report fitted values or prior widths, but these are the target parameters of the calibration.
Gaia parallax zero-point offset
Inferred simultaneously with PL parameters and distances; value not stated in the abstract.
individual stellar distances (latent)
Latent parameters with a disc-geometry prior; hierarchical structure implies many free latents regularized by the prior.

axioms (4)

domain assumption Milky Way Cepheids follow a disc geometry that can be encoded as a distance prior
Abstract states the model incorporates disc geometry through the distance prior; form of the prior is not specified here.
domain assumption Detection probability for the two HST SH0ES campaigns and Gaia cuts is fully captured by magnitude, parallax, period, and extinction selection functions reducible to an integral over distance and sky position
Central modelling claim of the abstract; correctness of the analytic reduction is load-bearing for the selection correction.
domain assumption Cepheid period-luminosity relation has small intrinsic scatter, so selection/geometry mismodelling does not strongly shift PL parameters when scatter is small
Abstract attributes <0.5 σ agreement with SH0ES to this small scatter.
standard math Standard Bayesian hierarchical inference and posterior predictive checking are valid for this generative model
Background statistical machinery assumed throughout.

pith-pipeline@v1.1.0-grok45 · 6265 in / 2910 out tokens · 19636 ms · 2026-07-14T23:55:37.385849+00:00 · methodology

0 comments

read the original abstract

The advent of high-precision Gaia parallaxes for Milky Way Cepheids enables per cent-level calibration of the local distance ladder and the Hubble constant $H_0$. We revisit the Milky Way Cepheid calibration from Gaia EDR3 parallaxes using a fully forward-modelled Bayesian framework that simultaneously infers the period--luminosity relation, the Gaia parallax zero-point offset, and individual stellar distances while explicitly incorporating the disc geometry of the Galaxy through the distance prior and the selection functions specified in two HST SH0ES campaigns. We derive an analytic treatment of the detection probability that accounts for magnitude, parallax, period, and extinction cuts and reduces it to a tractable integral over distance and sky position. Posterior predictive checks show that this generative model matches the observed distributions of parallaxes, magnitudes, and periods. Modelling Galactic structure and survey truncation self-consistently in a Bayesian framework yields period--luminosity parameters that agree with the SH0ES maximum-likelihood values at the ${<}0.5\,\sigma$ level, a consequence of the small intrinsic scatter of the Cepheid period--luminosity relation. Adopting the uniform-in-volume prior recently advocated by H\"og\r{a}s & M\"ortsell (2026), without simultaneously accounting for selection, leads to a ${\sim}\,0.05~\mathrm{mag}$ bias in the period--luminosity zero-point and posterior predictive distributions incompatible with the observed data; this shift is mostly driven by the omission of the selection model, and produces an apparent and unjustified shift in $H_0$ that reflects this mismodelling. A consistent Bayesian treatment of Galactic structure and selection effects reinforces the local distance-ladder determination of $H_0$, and hence the Hubble tension with early-Universe inferences.

discussion (0)

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Constraints on the gravitational potential from DESI DR2 BAO and its implications for the local void scenario
astro-ph.CO 2026-06 conditional novelty 6.0

A fixed gravitational-redshift contribution from a local void fits high-redshift data almost as well as ΛCDM, but the data currently favour z0=0.
Constraints on the gravitational potential from DESI DR2 BAO and its implications for the local void scenario
astro-ph.CO 2026-06 unverdicted novelty 5.0

High-redshift datasets constrain the local void gravitational redshift parameter z0 to be consistent with zero but allow the value needed for Hubble tension solution.
Late Time Dynamical Dark Energy and the CMB-Distance Ladder Tension
astro-ph.CO 2026-07 conditional novelty 4.0

The SNIa absolute-magnitude tension between the distance ladder (−19.204) and CMB+ΛCDM (−19.430) is independent of late-time expansion history, and DESI's w0–wa dark-energy hints shift H0 by only ~0.3–0.4 km/s/Mpc.